CN118411576B

CN118411576B - Carton classification method and device based on data processing

Info

Publication number: CN118411576B
Application number: CN202410892460.3A
Authority: CN
Inventors: 陶丹
Original assignee: Jiangsu Jianshan Packaging Technology Co ltd
Current assignee: Jiangsu Jianshan Packaging Technology Co ltd
Filing date: 2024-07-04
Publication date: 2024-11-12
Anticipated expiration: 2044-07-04

Abstract

The invention discloses a carton classification method and device based on data processing. The carton classification method based on data processing relates to the technical field of data processing, and comprises the following steps: collecting image data; extracting key image data; constructing a carton classification model; testing and evaluation. According to the method, the target classification set is set according to the preset classification target of carton classification, the image data of the cartons to be classified on the conveyor belt are acquired in real time, then the key image data are extracted according to the image recognition principle, the carton classification model is constructed by combining the extracted key image data, model optimization is carried out, finally the model-optimized carton classification model is tested in a simulated classification environment, meanwhile, the performance of the carton classification model when the placement angle of the cartons to be classified is changed is evaluated, the effect of improving the classification recognition rate in the carton transmission process is achieved, and the problem that the classification recognition rate in the carton transmission process is low in the prior art is solved.

Description

Carton classification method and device based on data processing

Technical Field

The invention relates to the technical field of data processing, in particular to a carton classification method and device based on data processing.

Background

Along with the rapid development of industrial automation technology, each link in the carton production and classification flow gradually realizes automation and intellectualization. The application of automatic production lines, robotics, automatic control systems and the like enables the carton production process to be more efficient and accurate, reduces labor cost, and the existing carton classification control technology is developing towards more intelligent and automatic directions. In addition, through high-resolution cameras and image processing technology, the characteristics of the shape, the color, the size and the like of the carton can be rapidly identified and classified. Meanwhile, a large amount of carton data can be learned and trained through a deep learning algorithm, so that automatic recognition of carton types is realized, and the method not only improves classification accuracy, but also greatly improves classification efficiency. However, traditional carton sorting is largely labor-dependent, inefficient and prone to error, and particularly, the limitations of manual sorting are becoming more apparent in the face of the need for large-scale, efficient carton sorting. Therefore, it is important to develop a more efficient and intelligent carton classification method and device for data processing, which is also an urgent need for development of the automation industry.

The existing carton classification system detects the position and angle of the cartons to be classified through a sensor, captures and analyzes the appearance characteristics of the cartons to be classified on the conveyor belt through an image processing technology, and then automatically classifies the cartons to be classified by comparing the extracted shape, size, color and texture characteristics with preset classification standards.

For example, bulletin numbers: a data processing system for entity classification of the invention patent publication of CN116227495B, comprising: acquiring a target text and acquiring a coding vector of the target text by using a trained first neural network model; reasoning the coding vector of the target text by using the trained second neural network model to obtain the coding vector of the target text; carrying out unified dimension and splicing treatment on the coding vector of the target text to obtain a corresponding target coding tensor; and (3) reasoning the target coding tensor by using the trained third neural network model to obtain the corresponding target text entity type.

For example, bulletin numbers: the invention patent publication text classification model training, classification method and system of CN113177119B and data processing system, comprising: determining a classification estimated value of an unlabeled sample based on a text classification model to be trained to obtain an estimated labeling sample set; acquiring a text and a vector of a marked sample set through an encoder to be trained, and acquiring a mixed marked sample set by combining a category identifier of the marked sample; inputting the estimated marked sample set and the mixed marked sample set into a feedforward neural network, adjusting parameters of the encoder to be trained and the current feedforward neural network according to the loss value of the loss function, and simultaneously obtaining a text classification model by combining preset training times.

However, in the process of implementing the technical scheme of the embodiment of the application, the application discovers that the above technology has at least the following technical problems:

in the prior art, as the appearance of the cartons to be classified possibly changes due to factors such as abrasion, stains and the like in the transportation process, the recognition rate is reduced, the cartons are difficult to accurately distinguish, and the problem of low classification recognition rate in the carton transportation process exists.

Disclosure of Invention

The embodiment of the application solves the problem of low classification recognition rate in the carton transmission process in the prior art by providing the carton classification method and the device based on data processing, and realizes the improvement of the classification recognition rate in the carton transmission process.

The embodiment of the application provides a carton classification method based on data processing, which comprises the following steps: s1, setting a target classification set according to a preset classification target of carton classification, and simultaneously acquiring and marking image data of cartons to be classified on a conveyor belt in real time through a sensor, wherein the target classification set is used for visualizing a classification region set corresponding to the cartons to be classified and the preset classification target; s2, carrying out normalization processing on the collected image data by using an image processing method, and extracting key image data according to an image recognition principle, wherein the key image data comprises appearance information data and texture information data of the cartons to be classified; s3, constructing a carton classification model by combining the extracted key image data and performing model optimization, wherein the carton classification model is used for automatically identifying and classifying the input cartons to be classified by learning the key image data, and the classification performance is used for numerically describing classification performance of the carton classification model in the automatic classification process; s4, testing the carton classification model subjected to model optimization in a simulated classification environment, and evaluating performance of the carton classification model when the placement angle of the carton to be classified is changed according to test results, wherein the simulated classification environment is used for simulating movement and placement angle change conditions of the carton to be classified on a conveyor belt.

Further, the specific setting steps of the target classification set are as follows: analyzing physical characteristics and functional characteristics of the carton to be classified to identify preset classification characteristics, and setting corresponding classification standards by combining classification characteristic data in a preset time period, wherein the classification characteristic data comprises physical characteristic data and functional characteristic data; the corresponding relation between the classification characteristic data and the preset classification characteristic is visually displayed in a classification tree form by combining the classification standard and the classification recognition index, wherein the classification recognition index is used for measuring the matching degree of the classification characteristic data and the preset classification characteristic; and analyzing the visual display result, presenting the classification logic and the classification area of the carton to be classified in a chart, and setting a target classification set in combination with the actual classification requirement.

Further, the specific obtaining steps of the classification identification index are as follows: extracting key information data from classified feature data, setting corresponding allocation weights according to similarity matching degree and classification dimension to obtain a classified identification index, wherein the key information data is used for describing physical characteristic information and functional characteristic information of the classified feature data, the similarity matching degree is used for measuring similarity degree of the classified feature data and the key information data, the classification dimension is used for quantifying bearing capacity and compression resistance of a carton to be classified in a classification process to obtain corresponding bearing capacity data and compression resistance data, the allocation weights are used for describing influence degree of the physical feature data, the functional feature data, the bearing capacity data and the compression resistance data relative to the classified identification index, and the allocation weights comprise a first weight, a second weight, a third weight and a fourth weight; the classification index is calculated by the following formula:

wherein y is the number of the carton to be classified, Y is the total number of cartons to be sorted,A classification recognition index corresponding to the y-th carton to be classified is represented, e is a natural constant,Representing preset physical characteristic data corresponding to the y-th carton to be classified,Representing the physical characteristic data corresponding to the y-th carton to be classified,Representing the reference deviation of the physical characteristic data,A first weight is indicated and a second weight is indicated,Representing the preset functional characteristic data corresponding to the y-th carton to be classified,Representing functional characteristic data corresponding to the y-th carton to be classified,Representing the deviation of the reference of the functional characteristic data,A second weight is indicated as being indicative of a second weight,Representing the bearing capacity data corresponding to the y-th carton to be classified,Indicating the reference data of the load bearing capacity,Representing the reference deviation of the load bearing capacity data,A third weight is indicated as being indicative of a third weight,Representing the compressive capacity data corresponding to the y-th carton to be classified,Representing the reference data for the resistance to pressure,Representing the reference deviation of the pressure resistance data,Representing a fourth weight.

Further, the specific extraction step of the key image data includes: encoding the acquired image data and mapping the image data to pixel point positions corresponding to the coordinate images; correcting the deviation of the rotation angle of the cartons to be classified in the conveying process of the conveyor belt, detecting the main direction of the image data on the coordinate image and determining the rotation angle between the main direction and the expected main direction; rotating the image data according to the determined rotation angle and by utilizing the rotation transformation matrix to obtain regular image data; extracting the outline of the carton to be classified by using an edge detection method, and simultaneously, performing filtering processing on the regular image data by using a filter to determine the texture characteristics of the carton to be classified; inputting the outline of the carton to be classified and the texture features of the carton to be classified into the gray level co-occurrence matrix to extract key image data.

Further, the specific step of constructing the carton classification model by combining the extracted key image data comprises the following steps: obtaining a target classification set region according to the target classification set and combining actual classification requirements, and simultaneously mapping pixel points of key image data in the gray level co-occurrence matrix with a preset target classification set region according to an image correction threshold; converting the key image data into a model numerical form, and designing a model network architecture by combining the data complexity of the key image data, wherein the data complexity represents the complexity of the key image data obtained by analyzing the discrete degree of the pixel points of the key image data in a preset target classification set region, and the model network architecture is used for regularizing the model numerical form of the key image data to eliminate the model numerical form with over-fitting and under-fitting; and constructing a carton classification model according to the designed model network architecture and combining model architecture coincidence coefficients, wherein the model architecture coincidence coefficients are used for measuring the adaptation degree between the model network architecture and the carton classification model.

Further, the model architecture coincidence coefficients are obtained by the following method: constructing a classification reference model based on an image correction threshold value and a classification recognition index, and performing model training on the constructed classification reference model by using key image data, wherein the classification reference model is used for visualizing the performance of the to-be-classified cartons in simulated classification on a conveyor belt; evaluating the classification reference model according to a preset evaluation index, judging whether the performance of the simulation classification of the carton to be classified meets the preset evaluation index according to the evaluation result, acquiring a model framework conforming coefficient according to data complexity and data similarity if the performance meets the preset evaluation index, otherwise retraining the constructed classification reference model until the preset evaluation index is met, wherein the data similarity represents the similarity degree of the key image data acquired by analyzing the pixel gray values of the key image data in a preset target classification set region; the model architecture coincidence coefficients are calculated by the following formula:

Wherein i is the number of training times of the model, M is the total number of times of model training,Representing the model architecture coincidence coefficient corresponding to the ith model training, e is a natural constant,Representing the brightness of the key image data corresponding to the ith model training,Representing the reference brightness of the key image data,A reference deviation representing the brightness of the key image data,Representing the contrast of the key image data corresponding to the ith model training,Representing the reference contrast of the key image data,A reference deviation representing the contrast of the key image data,Representing the data complexity corresponding to the ith model training,Representing the complexity of the reference to the data,Representing the reference deviation of the complexity of the data,Representing the data similarity corresponding to the ith model training,Representing the degree of similarity of the data references,Representing the data similarity reference deviation.

Further, the specific steps of the model optimization are as follows: performing corner detection on the placement angle of the cartons to be classified on the conveyor belt, and simultaneously comparing the placement angle of the corner detection with a preset standard angle in a target classification set area; generating integrated rotation information according to the comparison result and combining a rotation coincidence coefficient, wherein the rotation coincidence coefficient is used for measuring the adaptation degree of the cartons to be classified and the target classification set region after the placement angle is changed; and carrying out secondary training on the carton classification model by combining the integrated rotation information so as to realize automatic classification when the placement angle of the cartons to be classified changes.

Further, the rotation coincidence coefficient is obtained by the following method: evaluating a result of the carton classification model after secondary training and acquiring a confidence coefficient and a confidence coefficient evaluation factor by combining a reference confidence coefficient, wherein the confidence coefficient is used for measuring the confidence coefficient of the carton to be classified belonging to a target classification area under a placement angle, and the confidence coefficient evaluation factor represents the influence of the confidence coefficient on the placement angle when the carton to be classified is subjected to a classification process; obtaining a corner matching degree and an angle difference value according to a corner detection result, wherein the corner matching degree is used for measuring the matching degree of the position of a placed corner detected by the corner and the position of an expected corner in a target classification set area, and the angle difference value is used for measuring the deviation degree of the placed angle detected by the corner and a preset standard angle in the target classification set area; and obtaining the rotation coincidence coefficient by combining the relative deviation of the angular point matching degree and the relative deviation of the angle difference value.

Further, the specific step of testing the carton classification model after model optimization in the simulated classification environment comprises the following steps: constructing a simulated classification environment according to an actual application scene, and simultaneously acquiring a test data set by combining classification performance of a carton classification model, wherein the test data set represents image data of a carton to be classified under a preset category and a preset placement angle; loading the carton classification model subjected to model optimization into a simulated classification environment, setting test parameters of batch sequence and test round number according to actual test requirements, and simultaneously evaluating classification performance of the carton classification model by combining performance evaluation indexes; and adjusting the test parameters according to the evaluation result and performing iterative optimization on the carton classification model until the performance evaluation index is met.

The embodiment of the application provides a carton classification device based on data processing, which comprises: the system comprises an image data acquisition module, a key image data extraction module, a model optimization module and a test evaluation module; the image data acquisition module is used for setting a target classification set according to a preset classification target of carton classification, acquiring image data of cartons to be classified on the conveyor belt in real time through a sensor and marking the image data, and the target classification set is used for visualizing a classification area set corresponding to the cartons to be classified and the preset classification target; the key image data extraction module is used for carrying out normalization processing on the collected image data by utilizing an image processing method and extracting key image data according to an image recognition principle, wherein the key image data comprises appearance information data and texture information data of the carton to be classified; the model optimization module is used for constructing a carton classification model by combining the extracted key image data and performing model optimization, the carton classification model is used for automatically identifying and classifying the input carton to be classified by learning the key image data, and the classification performance is used for numerically describing classification performance of the carton classification model in the automatic classification process; the test evaluation module is used for testing the carton classification model subjected to model optimization in a simulated classification environment, and evaluating performance of the carton classification model when the carton to be classified is processed to change the placement angle according to test results, and the simulated classification environment is used for simulating movement and placement angle change conditions of the carton to be classified on the conveyor belt.

One or more technical solutions provided in the embodiments of the present application at least have the following technical effects or advantages:

1. The method comprises the steps of setting a target classification set through a preset classification target of carton classification, collecting image data of cartons to be classified on a conveyor belt in real time through a sensor and marking, extracting key image data according to an image recognition principle, constructing a carton classification model and optimizing the model, testing the model-optimized carton classification model in a simulated classification environment, and evaluating performance of the carton classification model in an automatic classification process, so that more accurate evaluation of classification performance is realized, further improvement of classification recognition rate in the carton transmission process is realized, and the problem of low classification recognition rate in the carton transmission process in the prior art is effectively solved.

2. The method comprises the steps of obtaining a target classification set region through a target classification set and combining actual classification requirements, simultaneously mapping pixel points of key image data in a gray level co-occurrence matrix with a preset target classification set region by combining an image correction threshold, then converting the key image data into a model numerical form, designing a model network architecture by combining data complexity of the key image data, and finally constructing a carton classification model according to the designed model network architecture and combining model architecture coincidence coefficients, thereby realizing accurate acquisition of the model architecture coincidence coefficients and further realizing more accurate construction of the carton classification model.

3. Through carrying out the angular point detection with the angle of placing of waiting to classify the carton on the conveyer belt, compare the angle of placing of angular point detection with the preset standard angle in the target classification collection region simultaneously, then according to the result of comparing and combine rotatory coincidence coefficient to generate integrated rotatory information, combine integrated rotatory information to carry out secondary training to the carton classification model so as to realize waiting to classify automatic classification when the angle changes of placing of carton to the improvement of angular point detection accuracy has been realized, and then more accurate classification of waiting to classify the carton has been realized.

Drawings

Fig. 1 is a flowchart of a method for classifying cartons based on data processing according to an embodiment of the present application;

Fig. 2 is a flow chart of the construction of a carton classification model according to an embodiment of the present application;

fig. 3 is a test flow chart of a carton classification model provided in an embodiment of the application;

fig. 4 is a schematic structural diagram of a carton classification device based on data processing according to an embodiment of the present application.

Detailed Description

The embodiment of the application solves the problem of low classification recognition rate in the carton transmission process in the prior art by providing the carton classification method and the device based on data processing, firstly sets a target classification set according to a preset classification target of carton classification so as to visualize a corresponding classification area set between a carton to be classified and the preset classification target, simultaneously acquires image data of the carton to be classified on a conveyor belt in real time through a sensor and marks the image data, then performs normalization processing on the acquired image data by using an image processing method, extracts appearance information data and texture information data of the carton to be classified according to an image recognition principle, then obtains a target classification set area according to the target classification set and combining with actual classification requirements, then converts the appearance information data and the texture information data into a model numerical form so as to design a model network architecture, simultaneously constructs a carton classification model by combining with the designed model network architecture, finally tests the model-optimized carton classification model in a simulated classification environment, simultaneously monitors the movement and placement angle change condition of the carton to be classified on the conveyor belt in real time, and evaluates the performance of the carton classification model when the carton placement angle change is processed according to test results, thereby realizing improvement of the recognition rate in the carton transmission process.

The technical scheme in the embodiment of the application aims to solve the problem of low classification recognition rate in the carton transmission process, and the overall thought is as follows:

The method comprises the steps of setting a target classification set through a preset classification target of carton classification, collecting image data of cartons to be classified on a conveyor belt in real time through a sensor and marking, then constructing a carton classification model by combining key image data extracted by an image recognition principle and optimizing the model, finally testing the model-optimized carton classification model in a simulated classification environment, and evaluating performance of the carton classification model when the placement angle of the cartons to be classified is changed according to test results, so that the effect of improving the classification recognition rate in the carton transmission process is achieved.

In order to better understand the above technical solutions, the following detailed description will refer to the accompanying drawings and specific embodiments.

As shown in fig. 1, a flowchart of a carton classification method based on data processing according to an embodiment of the present application is provided, and the carton classification method based on data processing provided by the present application includes the following steps: s1, setting a target classification set according to a preset classification target of carton classification, and simultaneously collecting and marking image data of cartons to be classified on a conveyor belt in real time through a sensor, wherein the target classification set is used for visualizing a classification region set corresponding to the cartons to be classified and the preset classification target; s2, carrying out normalization processing on the collected image data by using an image processing method, and extracting key image data according to an image recognition principle, wherein the key image data comprises appearance information data and texture information data of the cartons to be classified; s3, constructing a carton classification model by combining the extracted key image data and performing model optimization, wherein the carton classification model is used for automatically identifying and classifying the input cartons to be classified by learning the key image data, and the classification performance is used for quantitatively describing classification performance of the carton classification model in the automatic classification process; s4, testing the carton classification model subjected to model optimization in a simulated classification environment, and evaluating performance of the carton classification model when the placement angle of the carton to be classified is changed according to test results, wherein the simulated classification environment is used for simulating movement and placement angle change conditions of the carton to be classified on a conveyor belt.

In this embodiment, after setting the target classification set, the sensor is used to collect image data of the cartons to be classified on the conveyor belt in real time, and by selecting the sensor device with stable performance and reasonably arranging the sensor device at the key position of the conveyor belt, the size classification information, the material classification information and the usage classification information of the cartons to be classified can be ensured to be comprehensively captured, in addition, the influence of illumination conditions and background interference on the image quality needs to be considered, corresponding measures are taken to improve, after the image data is collected, advanced image processing technology is adopted to carry out deep analysis and processing, for example, the size of the paper box can be classified by utilizing an image recognition algorithm, the material of the paper box is recognized by utilizing a material recognition algorithm, the application of the paper box is classified by utilizing a natural language processing technology, a feedback mechanism can be introduced, the classification result is monitored and adjusted in real time, and when the recognition rate is low, the sensor parameters, the optimization algorithm model or the update target classification set can be adjusted in time.

It is to be understood that the normalization processing includes coordinate centering and rotation normalization, and the normalization processing can effectively eliminate data inconsistency caused by acquisition conditions and equipment differences among different images, so that accuracy of subsequent image analysis and recognition is improved, wherein the coordinate centering is to transform characteristic points or key areas in the images relative to a coordinate system of the whole image, so that all images are aligned on a uniform coordinate origin, position offset possibly generated in an acquisition process of the images can be eliminated, the rotation normalization is to eliminate differences of the images caused by different angles, and in a process of classifying the cartons to be classified, rotation correction is needed on the images due to the fact that the cartons to be classified are possibly presented on a conveyor belt at different angles, so that all the images are consistent in directions, and improvement of classification recognition rate in a carton transmission process is realized.

In this embodiment, the node dividing condition and the branch trend of the classification tree are determined according to the classification standard and the classification recognition index, for example, if a certain classification feature has a larger influence on the classification result, the classification tree can be used as an earlier node, and the branch with a higher classification recognition index is more likely to belong to a certain category, then, by visually displaying the classification tree, how the carton to be classified is gradually divided into different categories according to the physical feature and the functional feature can be clearly seen, and the display mode is not only convenient for understanding and analyzing the classification process, but also can find potential classification problems and improvement points, and in addition, additional information such as the classification accuracy and the misjudgment rate of each node can be marked on the classification tree so as to further evaluate the classification performance; the classification logic and the classification area for graphically presenting the cartons to be classified specifically comprise the following contents: the classification logic graphically presents: marking operation and judgment conditions of each step by using a flow chart with an arrow and a text box, displaying the whole flow from inputting the characteristics to outputting the classification result, including the steps of characteristic extraction, judgment, classification and the like, then using a table form, wherein rows represent characteristic value ranges, columns represent categories, and the corresponding classification result is filled in a cell, including the mapping relation between the characteristic value ranges and the corresponding categories; the classification area is graphically presented: the area map or the line map with the colors is used, different categories are filled with different colors, boundary lines are clearly marked, meanwhile, a two-dimensional or three-dimensional scatter map is drawn according to key features, each point represents one carton, the colors or the shapes distinguish different categories, the gathering area and the distribution mode of each category of cartons can be intuitively seen, dark colors represent areas with high correlation degree or high density, light colors represent areas with low correlation degree or low density, and more accurate setting of target classification sets is achieved.

Further, physical characteristic data, functional characteristic data, bearing capacity data and compressive capacity data are input into a machine learning algorithm for learning and training, and the actual classification requirements and preset data are combined for corresponding prediction in sequence, and meanwhile, the difference between the actual classification requirements and the preset classification targets is automatically identified according to the prediction results, so that corresponding classification identification indexes are obtained; in addition to the classification recognition index obtained by the machine learning algorithm described above, the acquisition and calculation can be performed by the following methods and formulas: the specific acquisition steps of the classification identification index are as follows: extracting key information data from the classified feature data, setting corresponding allocation weights according to similarity matching degree and classification dimension to obtain a classified identification index, wherein the key information data is used for describing physical characteristic information and functional characteristic information of the classified feature data, the similarity matching degree is used for measuring similarity degree of the classified feature data and the key information data, the classification dimension is used for quantifying bearing capacity and compression resistance of a carton to be classified in a classification process to obtain corresponding bearing capacity data and compression resistance data, the allocation weights are used for describing influence degree of the physical feature data, the functional feature data, the bearing capacity data and the compression resistance data relative to the classified identification index, and the allocation weights comprise a first weight, a second weight, a third weight and a fourth weight; the class identification index is calculated by the following formula:

In this embodiment, physical feature data such as the size, shape and material of the cartons are the basis of classification and identification, when the measurement and recording of these data are more accurate, the types of the cartons can be more accurately identified, and by counting and analyzing a large number of physical feature data of the cartons, differences of the cartons in physical features of different types can be found, so that the first weight in the classification and identification index is generally higher, where the first weight represents the weight of the physical feature data relative to the classification and identification index; the functional characteristic data describe the purpose and design characteristics of the cartons, which are important for classification and identification, and when the functional characteristic data are more detailed and accurate, the purpose and characteristics of the cartons can be better understood, and the functional differences and similar points of the cartons of different categories can be found by analyzing the functional characteristic data, and powerful basis is provided for a classification algorithm according to the differences and similar points and in combination with a second weight, wherein the second weight represents the influence degree of the functional characteristic data relative to a classification and identification index; the bearing capacity data are important indexes reflecting the performance of the paper box, the performance of the paper box in actual use is better evaluated, and the paper box with high bearing capacity can be rapidly identified by testing and analyzing the bearing capacity data, so that in the classifying and identifying process, a third weight is set according to actual requirements, wherein the third weight represents the influence degree of the bearing capacity data relative to the classifying and identifying index; the pressure resistance data describe the stability of the paper box when the paper box is subjected to external force, the paper box can be subjected to the actions of extrusion, collision and the like in the transportation and storage processes, and the performance of the paper box in the transportation and storage processes can be predicted better when the pressure resistance data are more accurate, so that the fourth weight is usually set according to the use environment and the requirements of the paper box, wherein the fourth weight represents the influence degree of the pressure resistance data relative to the classification identification index; in summary, the physical feature data, the functional feature data, the load-bearing capacity data and the compression-resistant capacity data are directly proportional to the influence of the classification index, and the ratio of the physical feature data to the compression-resistant capacity data is the highest, generally, the first weight accounts for 30%, the second weight accounts for 20%, the third weight accounts for 20%, the fourth weight accounts for 30%, the specific ratio weight is set according to the actual requirement, and the accuracy and the integrity of the data are improved, so that the accuracy of the classification index can be remarkably improved, and the improvement of the classification accuracy and the reliability of the cartons to be classified is realized.

In this embodiment, in the transportation process of the conveyor belt, the rotation angle deviation may be generated due to various reasons, which may adversely affect the accuracy of classification, so that correction of the rotation angle deviation is critical, the main direction of the image data on the coordinate image may be detected and compared with the expected main direction by using an image processing technology, so as to determine the rotation angle, once the rotation angle is determined, the image data may be rotated by using a rotation transformation matrix, where the rotation transformation matrix may accurately rotate the image according to the given rotation angle to obtain regular image data, that is, the image data with the rotation angle deviation eliminated is filtered by a filter, and then some frequency components in the image may be highlighted or suppressed, so that the texture feature of the carton to be classified is extracted, and after the contour and texture feature of the carton to be classified are extracted, these information are input into a gray level co-occurrence matrix, where the co-occurrence matrix is a statistical method for describing the image texture, and the frequency of the appearance of the different gray level pairs in the texture image may be reflected by the different gray level in the texture image, so that the statistical key information of the carton may be effectively extracted; the contour and the texture features are combined, so that more comprehensive and accurate carton information to be classified can be provided for people, and more accurate recognition of the cartons to be classified is realized.

Further, as shown in fig. 2, a flowchart of constructing a carton classification model according to an embodiment of the present application includes the specific steps of: obtaining a target classification set region according to the target classification set and combining actual classification requirements, and simultaneously mapping pixel points of key image data in the gray level co-occurrence matrix with a preset target classification set region according to an image correction threshold; converting the key image data into a model numerical form, and designing a model network architecture by combining the data complexity of the key image data, wherein the data complexity represents the complexity of the key image data obtained by analyzing the discrete degree of the pixel points of the key image data in a preset target classification set region, and the model network architecture is used for regularizing the model numerical form of the key image data to eliminate the model numerical form of which fitting and under fitting occur; and constructing a carton classification model according to the designed model network architecture and combining model architecture coincidence coefficients, wherein the model architecture coincidence coefficients are used for measuring the adaptation degree between the model network architecture and the carton classification model.

In this embodiment, the image correction threshold is used to measure the difference degree of brightness and contrast between the key image data, which helps to eliminate noise and inconsistency in the image, improve the quality of the image data, map the pixels of the key image data into the target classification set area by setting a proper threshold (usually between 0 and 1, and specifically determined according to the actual situation), and convert the key image data into a model numerical form after mapping is completed, which is to enable the image data to be understood and processed by a machine learning model, generally, we can use an image feature extraction method (such as a convolutional neural network) to convert the image data into feature vectors or matrices as the input of the model, and these feature vectors or matrices can capture the key information in the image, so as to provide an effective basis for training and classification of the model; for a carton classification task, the data complexity generally comprises the diversity of the shape of the carton to be classified, the complexity of textures and the change of illumination conditions, and when a network architecture of a model is designed, the proper number of network layers and node numbers are required to be selected by taking the factors into consideration so as to construct a carton classification model capable of effectively processing complex data, thereby improving the classification efficiency of the carton classification model.

Further, a super-parameter optimization tool in an automatic machine learning method is used for automatically searching and evaluating a preset model structure, the most suitable model structure is selected and tested by combining actual classification requirements and key image data, and then the degree of conformity between the model structure and the preset model classification performance is evaluated according to the test result and by comparing the classification performance of the model structure, so that a model structure conformity coefficient is obtained; in addition to obtaining the model architecture compliance coefficients through the automated machine learning algorithm described above, the model architecture compliance coefficients may also be obtained and calculated by the following methods and formulas: the model architecture coincidence coefficients are obtained by the following method: constructing a classification reference model based on an image correction threshold value and a classification recognition index, and performing model training on the constructed classification reference model by using key image data, wherein the classification reference model is used for visualizing the performance of the simulated classification of the carton to be classified on the conveyor belt; evaluating the classification reference model according to a preset evaluation index, judging whether the performance of the simulated classification of the carton to be classified meets the preset evaluation index according to the evaluation result, acquiring a model framework conforming coefficient according to data complexity and data similarity if the performance meets the preset evaluation index, otherwise retraining the constructed classification reference model until the preset evaluation index is met, wherein the data similarity represents the similarity degree of the key image data acquired by analyzing the pixel gray values of the key image data in a preset target classification set region; the model architecture fit coefficients are calculated by the following formula:

，

In this embodiment, a classification reference model is constructed according to the complexity of an actual classification task and in combination with an image correction threshold, and a super parameter of an initial learning rate is set, wherein the data is used for controlling the speed of model training, meanwhile, a loss function corresponding to the classification task is selected, the loss function is used for measuring the difference between the classification performance of the classification reference model in the process of executing the classification task and the preset effect of the actual classification task, and then, a data preprocessing flow of key image data is set, which generally comprises data loading, classification output, back propagation and parameter updating, wherein the back propagation is used for carrying out parameter updating on the classification reference model in the model training process so as to minimize the loss function; setting a preset evaluation index according to task requirements, comparing the performance index of the simulated classification with the preset evaluation index, if the performance index of the simulated classification meets or exceeds the preset evaluation index, considering that the performance is good, otherwise, further analyzing possible reasons (such as insufficient model complexity, insufficient training data or excessive noise), adjusting a model framework, adding the training data and carrying out noise processing according to analysis results, and carrying out model training again by using the adjusted model and a new data set, and repeating the steps to iterate and optimize the model until the performance of the simulated classification meets the preset evaluation index; the reference value in the calculation formula of the model architecture coincidence coefficient is usually obtained by carrying out statistics on a large amount of data with good historical operation and then carrying out weighted average, wherein the data complexity generally refers to the number of categories in a data set, the correlation among features, the noise level and the distribution condition of the data, and the higher the data complexity is, the higher the expression capability and the more complex structure are required by the carton classification model to capture the modes and rules in the data; the data similarity refers to the similarity degree between the training data and the data to be classified, if the training data is very similar to the data to be classified, the carton classification model may learn effective characteristic representation from the training data more easily, and obtain good performance on the data to be classified, otherwise, if the similarity between the training data and the data to be classified is lower, the carton classification model may be difficult to generalize to new data, so that the classification performance is reduced; the more accurate acquisition of the model architecture coincidence coefficient is realized.

Further, the specific steps of model optimization are as follows: performing corner detection on the placement angle of the cartons to be classified on the conveyor belt, and simultaneously comparing the placement angle of the corner detection with a preset standard angle in a target classification set area; generating integrated rotation information according to the comparison result and combining a rotation coincidence coefficient, wherein the rotation coincidence coefficient is used for measuring the adaptation degree of the cartons to be classified and the target classification set region after the placement angle is changed; and carrying out secondary training on the carton classification model by combining the integrated rotation information so as to realize automatic classification when the placement angle of the cartons to be classified changes.

In this embodiment, the specific steps of corner detection are: according to the shape, the size and the characteristics of a conveyor belt of the cartons to be classified, applying an angular point detection algorithm to key image data to extract angular point features of the cartons, wherein the angular points are points with severe gray value changes in the images and can represent edges and contours of the cartons; determining corresponding corner points of the same carton in different images through corner point matching between adjacent frame images; determining the placement angle of the cartons to be classified on the conveyor belt by calculating vector relations or geometric transformations among the corner points according to the matched corner points, which generally involve vector angle calculation and matrix transformation; the specific steps of the secondary training of the carton classification model are as follows: extracting placement angle information in key image data and converting the placement angle information into a model readable form, for example, converting an angle value into a vector form through an angle coding layer; fusing the encoded angle information with the extracted placement angle information by splicing and using an attention mechanism; carrying out primary training on the carton classification model by using a data set with labels and rotation information, so that the carton classification model learns how to classify according to the characteristics and the placement angle of the cartons; evaluating the performance of the model after the initial training on a verification set, and analyzing the classification accuracy of the carton classification model under different placement angles; performing secondary training on the carton classification model according to the performance of the model after primary training and by utilizing the data set after rotation enhancement; the trained carton classification model is deployed into an actual application scene, for example, the carton classification model is integrated into a machine vision system on a conveyor belt, and meanwhile, the automatic classification performance of the carton classification model when the placement angle is changed is verified, so that more accurate optimization and recognition of the carton classification model are realized.

Further, the rotation coincidence coefficient is obtained by the following method: evaluating a result of the carton classification model after secondary training and acquiring a confidence coefficient and a confidence coefficient evaluation factor by combining a reference confidence coefficient, wherein the confidence coefficient is used for measuring the confidence coefficient of the carton to be classified belonging to a target classification area under the placement angle, and the confidence coefficient evaluation factor represents the influence of the confidence coefficient on the placement angle when the carton to be classified is subjected to the classification process; acquiring a corner matching degree and an angle difference value according to a corner detection result, wherein the corner matching degree is used for measuring the matching degree of the position of a placed corner detected by the corner and the position of an expected corner in a target classification set area, and the angle difference value is used for measuring the deviation degree of the placed angle detected by the corner and a preset standard angle in the target classification set area; and obtaining the rotation coincidence coefficient by combining the relative deviation of the angular point matching degree and the relative deviation of the angle difference value.

In this embodiment, the actual rotation angle distribution of the carton images to be classified in the test set is counted, the similarity of the distribution and the preset standard angle distribution is compared, then the difference between the two angle distributions is quantified by using a statistical index (such as chi-square test), and a rotation coincidence coefficient is obtained according to the result of the distribution difference, wherein in the actual application, the coefficient can be a function of a difference value (such as the inverse of the difference value and the normalized value of the difference value); the rotation coincidence coefficient is obtained by the statistical analysis method, and can be calculated by the following formula:

，

wherein r is the number of the corner detection times, R is the total number of times of corner detection,Indicating that the rotation coincidence coefficient corresponding to the detection of the r-th angle point, e is a natural constant,The confidence level assessment factor is represented as such,Indicating the confidence level corresponding to the detection of the r-th corner,The confidence level of the reference is indicated,Indicating the corner matching degree corresponding to the r-th corner detection,Represents the degree of matching of the corner references,The value of the angle reference difference is indicated,Representing the angle difference value; matching the detected corner points with expected corner point positions in the reference image, wherein the matching process can be realized by comparing feature descriptions among the corner points (such as calculating Euclidean distance or cosine similarity among the feature descriptions), and comparing the feature descriptions with feature descriptors of the expected corner points, wherein the more the number of successful matches is, the higher the matching degree is; similarly, the angle difference value generally compares the detected angle with the expected angle to obtain a difference value or an absolute value difference of the angle, meanwhile, the angle point reference matching degree and the angle reference difference value are obtained by carrying out weighted average after counting a large amount of data with good historical operation, and the rotation coincidence coefficient is comprehensively evaluated by combining the obtained values.

Specifically, the comprehensive evaluation also needs to combine the confidence coefficient and the confidence coefficient evaluation factor of the placement angle of the carton to be classified, and the specific acquisition content comprises: setting a test set according to the key image data and combining the secondary training result, inputting the test set into a carton classification model to output probability distribution belonging to each category, selecting the category with the highest probability as a prediction category, taking the highest probability as a reference confidence level, setting a confidence level threshold according to actual application requirements, and considering that the classification result of the model has higher reliability when the reference confidence level is higher than the threshold, otherwise, considering that the classification result may not be reliable; the confidence coefficient evaluation factor is mainly used for measuring the influence of the placement angle on the confidence coefficient, and can be a function of the angle difference value (such as an inverse proportion function and an exponential function) according to the difference value between the placement angle obtained by angular point detection and the actual placement angle in the test set, so that when the angle difference is smaller, the factor is larger, and when the angle difference is larger, the factor is smaller, finally, the reference confidence coefficient is combined with the confidence coefficient evaluation factor, the classification probability of the model and the influence of the placement angle on the classification result are comprehensively considered, and the improvement of the accuracy and the reliability of the rotation coincidence coefficient is realized.

Further, as shown in fig. 3, a test flow chart of the carton classification model provided in the embodiment of the application includes the specific steps of testing the model-optimized carton classification model in a simulated classification environment: constructing a simulated classification environment according to an actual application scene, and simultaneously acquiring a test data set by combining the classification performance of the carton classification model, wherein the test data set represents image data of the carton to be classified under a preset category and a preset placement angle; loading the carton classification model subjected to model optimization into a simulated classification environment, setting test parameters of batch sequence and test round number according to actual test requirements, and simultaneously evaluating classification performance of the carton classification model by combining performance evaluation indexes; and adjusting the test parameters according to the evaluation result and performing iterative optimization on the carton classification model until the performance evaluation index is met.

In this embodiment, the batch order is generally set by determining the input order of the test data, which may be random or may be performed in a specific order (e.g., by category, by placement angle); setting the number of test rounds generally, namely the number of times the model predicts the test data, so as to obtain more stable and reliable evaluation results, multiple rounds of tests can be performed and statistical analysis can be performed on the test results; the performance evaluation index is generally set according to the performance of the carton classification model in the training and testing process, and meanwhile, based on the performance evaluation result, a targeted optimization suggestion is provided: if the overall performance is poor, the model structure is considered to be adjusted, training data is added or a training strategy is optimized; if some categories perform poorly, an attempt may be made to increase the number of samples for those categories, improve feature extraction methods, or introduce category balancing strategies; if the model is sensitive to the placement angle, a data enhancement strategy of adding angle change in the training process can be considered; a more accurate assessment of the sorting performance of the cartons to be sorted is achieved.

As shown in fig. 4, a schematic structural diagram of a carton classification device based on data processing according to an embodiment of the present application is provided, where the carton classification device based on data processing according to the embodiment of the present application includes: the system comprises an image data acquisition module, a key image data extraction module, a model optimization module and a test evaluation module; the image data acquisition module is used for setting a target classification set according to a preset classification target of carton classification, acquiring image data of cartons to be classified on the conveyor belt in real time through a sensor and marking the image data, and the target classification set is used for visualizing a classification region set corresponding to the cartons to be classified and the preset classification target; the key image data extraction module is used for carrying out normalization processing on the collected image data by utilizing an image processing method and extracting key image data according to an image recognition principle, wherein the key image data comprises appearance information data and texture information data of the carton to be classified; the model optimizing module is used for constructing a carton classification model by combining the extracted key image data and carrying out model optimization, the carton classification model is used for automatically identifying and classifying the input carton to be classified by learning the key image data, and the classification performance is used for numerically describing the classification performance of the carton classification model in the automatic classification process; the test evaluation module is used for testing the carton classification model after model optimization in a simulated classification environment, and evaluating the performance of the carton classification model when the placement angle of the carton to be classified is changed according to the test result, wherein the simulated classification environment is used for simulating the movement and placement angle change condition of the carton to be classified on the conveyor belt.

In this embodiment, in practical application, since the placement angle of the carton to be classified may be changed due to human factors or operation of an automation device, the carton classification model needs to have robustness to angle change, through the test evaluation module, carton images of different angles can be simulated, and classification performance of the carton classification model under the changes can be observed, so as to determine whether the carton classification model can effectively challenge the angle change, and secondly, in the test process, the prediction result of the carton classification model under different angles can be collected and compared with the actual label, and classification effects of the carton classification model under different categories and different angles can be more intuitively displayed through the visualization tool, so that more accurate classification of the carton to be classified is realized.

The technical scheme provided by the embodiment of the application at least has the following technical effects or advantages: relative to the bulletin number: according to the data processing system for entity classification disclosed by the application patent publication CN116227495B, a target classification set region is obtained through a target classification set and by combining actual classification requirements, meanwhile, pixels of key image data in a gray level co-occurrence matrix are mapped with a preset target classification set region by combining an image correction threshold value, then the key image data are converted into a model numerical form, a model network architecture is designed by combining data complexity of the key image data, and finally, a carton classification model is constructed according to the designed model network architecture and by combining model architecture coincidence coefficients, so that accurate acquisition of the model architecture coincidence coefficients is realized, and more accurate construction of the carton classification model is realized; relative to the bulletin number: according to the text classification model training and classifying method and system and the data processing system disclosed by the patent publication of CN113177119B, the embodiment of the application is characterized in that the angular point detection is carried out on the placement angle of the cartons to be classified on the conveyor belt, meanwhile, the placement angle of the angular point detection is compared with the preset standard angle in the target classification set area, then integrated rotation information is generated according to the comparison result and in combination with the rotation coincidence coefficient, and finally, the secondary training is carried out on the carton classification model in combination with the integrated rotation information so as to realize automatic classification when the placement angle of the cartons to be classified changes, so that the angular point detection accuracy is improved, and further, more accurate classification of the cartons to be classified is realized.

It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. It is therefore intended that the following claims be interpreted as including the preferred embodiments and all such alterations and modifications as fall within the scope of the invention.

It will be apparent to those skilled in the art that various modifications and variations can be made to the present invention without departing from the spirit or scope of the invention. Thus, it is intended that the present invention also include such modifications and alterations insofar as they come within the scope of the appended claims or the equivalents thereof.

Claims

1. A carton classification method based on data processing, comprising the steps of:

S1, setting a target classification set according to a preset classification target of carton classification, and simultaneously acquiring and marking image data of cartons to be classified on a conveyor belt in real time through a sensor, wherein the target classification set is used for visualizing a classification region set corresponding to the cartons to be classified and the preset classification target;

S2, carrying out normalization processing on the collected image data by using an image processing method, and extracting key image data according to an image recognition principle, wherein the key image data comprises appearance information data and texture information data of the cartons to be classified;

S3, constructing a carton classification model by combining the extracted key image data and performing model optimization, wherein the carton classification model is used for automatically identifying and classifying the input cartons to be classified by learning the key image data, and the classification performance is used for numerically describing classification performance of the carton classification model in the automatic classification process;

S4, testing the model-optimized carton classification model in a simulated classification environment, and evaluating performance of the carton classification model when the placement angle of the carton to be classified is changed according to test results, wherein the simulated classification environment is used for simulating the movement and placement angle change condition of the carton to be classified on a conveyor belt;

The specific setting steps of the target classification set are as follows:

Analyzing physical characteristics and functional characteristics of the carton to be classified to identify preset classification characteristics, and setting corresponding classification standards by combining classification characteristic data in a preset time period, wherein the classification characteristic data comprises physical characteristic data and functional characteristic data;

the corresponding relation between the classification characteristic data and the preset classification characteristic is visually displayed in a classification tree form by combining the classification standard and the classification recognition index, wherein the classification recognition index is used for measuring the matching degree of the classification characteristic data and the preset classification characteristic;

Analyzing the visual display result, presenting classification logic and classification areas of the cartons to be classified by a chart, and setting a target classification set by combining with actual classification requirements;

The specific acquisition steps of the classification identification index are as follows:

extracting key information data from classified feature data, setting corresponding allocation weights according to similarity matching degree and classification dimension to obtain a classified identification index, wherein the key information data is used for describing physical characteristic information and functional characteristic information of the classified feature data, the similarity matching degree is used for measuring similarity degree of the classified feature data and the key information data, the classification dimension is used for quantifying bearing capacity and compression resistance of a carton to be classified in a classification process to obtain corresponding bearing capacity data and compression resistance data, the allocation weights are used for describing influence degree of the physical feature data, the functional feature data, the bearing capacity data and the compression resistance data relative to the classified identification index, and the allocation weights comprise a first weight, a second weight, a third weight and a fourth weight;

the classification index is calculated by the following formula:

，

wherein y is the number of the carton to be classified, Y is the total number of cartons to be sorted,A classification recognition index corresponding to the y-th carton to be classified is represented, e is a natural constant,Representing preset physical characteristic data corresponding to the y-th carton to be classified,Representing the physical characteristic data corresponding to the y-th carton to be classified,Representing the reference deviation of the physical characteristic data,A first weight is indicated and a second weight is indicated,Representing the preset functional characteristic data corresponding to the y-th carton to be classified,Representing functional characteristic data corresponding to the y-th carton to be classified,Representing the deviation of the reference of the functional characteristic data,A second weight is indicated as being indicative of a second weight,Representing the bearing capacity data corresponding to the y-th carton to be classified,Indicating the reference data of the load bearing capacity,Representing the reference deviation of the load bearing capacity data,A third weight is indicated as being indicative of a third weight,Representing the compressive capacity data corresponding to the y-th carton to be classified,Representing the reference data for the resistance to pressure,Representing the reference deviation of the pressure resistance data,Representing a fourth weight;

the specific extraction step of the key image data comprises the following steps:

Encoding the acquired image data and mapping the image data to pixel point positions corresponding to the coordinate images;

Correcting the deviation of the rotation angle of the cartons to be classified in the conveying process of the conveyor belt, detecting the main direction of the image data on the coordinate image and determining the rotation angle between the main direction and the expected main direction;

Rotating the image data according to the determined rotation angle and by utilizing the rotation transformation matrix to obtain regular image data;

Extracting the outline of the carton to be classified by using an edge detection method, and simultaneously, performing filtering processing on the regular image data by using a filter to determine the texture characteristics of the carton to be classified;

Inputting the outline of the carton to be classified and the texture features of the carton to be classified into a gray level co-occurrence matrix to extract key image data;

the specific steps of constructing the carton classification model by combining the extracted key image data include:

obtaining a target classification set region according to the target classification set and combining actual classification requirements, and simultaneously mapping pixel points of key image data in the gray level co-occurrence matrix with a preset target classification set region according to an image correction threshold;

Converting the key image data into a model numerical form, and designing a model network architecture by combining the data complexity of the key image data, wherein the data complexity represents the complexity of the key image data obtained by analyzing the discrete degree of the pixel points of the key image data in a preset target classification set region, and the model network architecture is used for regularizing the model numerical form of the key image data to eliminate the model numerical form with over-fitting and under-fitting;

Constructing a carton classification model according to a designed model network architecture and combining model architecture coincidence coefficients, wherein the model architecture coincidence coefficients are used for measuring the adaptation degree between the model network architecture and the carton classification model;

The model architecture coincidence coefficient is obtained by the following method:

Constructing a classification reference model based on an image correction threshold value and a classification recognition index, and performing model training on the constructed classification reference model by using key image data, wherein the classification reference model is used for visualizing the performance of the to-be-classified cartons in simulated classification on a conveyor belt;

Evaluating the classification reference model according to a preset evaluation index, judging whether the performance of the simulation classification of the carton to be classified meets the preset evaluation index according to the evaluation result, acquiring a model framework conforming coefficient according to data complexity and data similarity if the performance meets the preset evaluation index, otherwise retraining the constructed classification reference model until the preset evaluation index is met, wherein the data similarity represents the similarity degree of the key image data acquired by analyzing the pixel gray values of the key image data in a preset target classification set region;

the model architecture coincidence coefficients are calculated by the following formula:

，

2. The carton classification method based on data processing as claimed in claim 1, wherein the specific steps of model optimization are:

Performing corner detection on the placement angle of the cartons to be classified on the conveyor belt, and simultaneously comparing the placement angle of the corner detection with a preset standard angle in a target classification set area;

Generating integrated rotation information according to the comparison result and combining a rotation coincidence coefficient, wherein the rotation coincidence coefficient is used for measuring the adaptation degree of the cartons to be classified and the target classification set region after the placement angle is changed;

and carrying out secondary training on the carton classification model by combining the integrated rotation information so as to realize automatic classification when the placement angle of the cartons to be classified changes.

3. A method of sorting cartons based on data processing as claimed in claim 2 and wherein: the rotation coincidence coefficient is obtained by the following method:

evaluating a result of the carton classification model after secondary training and acquiring a confidence coefficient and a confidence coefficient evaluation factor by combining a reference confidence coefficient, wherein the confidence coefficient is used for measuring the confidence coefficient of the carton to be classified belonging to a target classification area under a placement angle, and the confidence coefficient evaluation factor represents the influence of the confidence coefficient on the placement angle when the carton to be classified is subjected to a classification process;

Obtaining a corner matching degree and an angle difference value according to a corner detection result, wherein the corner matching degree is used for measuring the matching degree of the position of a placed corner detected by the corner and the position of an expected corner in a target classification set area, and the angle difference value is used for measuring the deviation degree of the placed angle detected by the corner and a preset standard angle in the target classification set area;

and obtaining the rotation coincidence coefficient by combining the relative deviation of the angular point matching degree and the relative deviation of the angle difference value.

4. The method for classifying cartons based on data processing according to claim 1, wherein the specific step of testing the model optimized carton classification model in the simulated classification environment comprises the steps of:

Constructing a simulated classification environment according to an actual application scene, and simultaneously acquiring a test data set by combining classification performance of a carton classification model, wherein the test data set represents image data of a carton to be classified under a preset category and a preset placement angle;

Loading the carton classification model subjected to model optimization into a simulated classification environment, setting test parameters of batch sequence and test round number according to actual test requirements, and simultaneously evaluating classification performance of the carton classification model by combining performance evaluation indexes;

and adjusting the test parameters according to the evaluation result and performing iterative optimization on the carton classification model until the performance evaluation index is met.

5. An apparatus for applying a data processing based carton classification method according to any of claims 1-4, comprising: the system comprises an image data acquisition module, a key image data extraction module, a model optimization module and a test evaluation module;

the image data acquisition module is used for setting a target classification set according to a preset classification target of carton classification, acquiring image data of cartons to be classified on the conveyor belt in real time through a sensor and marking the image data, and the target classification set is used for visualizing a classification area set corresponding to the cartons to be classified and the preset classification target;

The key image data extraction module is used for carrying out normalization processing on the collected image data by utilizing an image processing method and extracting key image data according to an image recognition principle, wherein the key image data comprises appearance information data and texture information data of the carton to be classified;

The model optimization module is used for constructing a carton classification model by combining the extracted key image data and performing model optimization, the carton classification model is used for automatically identifying and classifying the input carton to be classified by learning the key image data, and the classification performance is used for numerically describing classification performance of the carton classification model in the automatic classification process;

The test evaluation module is used for testing the carton classification model subjected to model optimization in a simulated classification environment, and evaluating performance of the carton classification model when the carton to be classified is processed to change the placement angle according to test results, and the simulated classification environment is used for simulating movement and placement angle change conditions of the carton to be classified on the conveyor belt.