CN117351001B - Surface defect identification method for regenerated aluminum alloy template - Google Patents

Abstract

The present invention provides a method for identifying surface defects of a recycled aluminum alloy template, which relates to the technical field of aluminum alloy production, and includes: step 1: obtaining a real-time image of a template and performing preprocessing to obtain an image to be processed; step 2: performing feature extraction on the image to be processed, generating defect feature information, and inputting the information into a preset defect recognition model for defect recognition to obtain a first result; step 3: obtaining historical defect image information, and establishing a historical defect image set, and at the same time, inputting the first result and the historical defect image set into a preset image comparison model for comparison analysis to obtain a second result; step 4: based on the defect feature information, screening a defect analysis model in a preset model database, and inputting the first result and the second result into the defect analysis model for defect analysis to obtain a final recognition result. The present invention can improve the recognition accuracy of surface defects of recycled aluminum alloy templates, thereby improving the quality control level of aluminum alloy production.

Description

A method for identifying surface defects of recycled aluminum alloy templates

Technical Field

The invention relates to the technical field of aluminum alloy production, and in particular to a method for identifying surface defects of a recycled aluminum alloy template.

Background technique

Aluminum alloy formwork is a mold or model used to manufacture various aluminum alloy products. It is usually made of aluminum alloy material and has the characteristics of high strength, wear resistance, corrosion resistance, etc. Aluminum alloy formwork is widely used in various fields, such as automotive industry, aerospace, construction and architecture, electronic and electrical industry, etc.

At present, with the improvement of energy-saving and environmental protection awareness in industrial production, recycled aluminum alloy formwork technology has gradually been applied to the field of aluminum alloy production. Recycled aluminum alloy formwork refers to aluminum alloy formwork made by recycling and reusing recycled scrap aluminum materials. Although the use of recycled aluminum alloy formwork helps promote circular economy and sustainable development, reduce resource consumption and environmental impact. However, due to the uncertainty and uneven quality of scrap aluminum materials, the physical properties and product performance of recycled aluminum alloy formworks are somewhat different from those of the original formworks. Therefore, quality control and testing of recycled aluminum alloy formworks are particularly important.

Therefore, the present invention provides a method for identifying surface defects of a recycled aluminum alloy template.

Summary of the invention

The present invention provides a method for identifying surface defects of a recycled aluminum alloy template, which is used to improve the recognition accuracy of surface defects of the recycled aluminum alloy template, thereby improving the quality level of the finished product of the recycled aluminum alloy template, thereby improving the quality control level of aluminum alloy production.

The present invention provides a method for identifying surface defects of a recycled aluminum alloy template, comprising:

Step 1: Acquire a real-time image of the template through an image acquisition device, and pre-process the real-time image to obtain an image to be processed;

Step 2: extracting features from the image to be processed to generate defect feature information, and inputting the defect feature information into a preset defect recognition model to perform defect recognition to obtain a first result;

Step 3: Obtain historical defect image information of the template and establish a historical defect image set. At the same time, input the first result and the historical defect image set into a preset image comparison model for comparison and analysis to obtain a second result;

Step 4: Based on the defect feature information, a defect analysis model is obtained by screening in a preset model database, and the first result and the second result are input into the defect analysis model to perform defect analysis to obtain a final recognition result.

Preferably, step 1 includes:

The template surface is imaged from a preset angle by an image acquisition device to obtain a real-time image, and a real-time image set is established according to the time sequence of the real-time images;

Each real-time image in the real-time image set is preprocessed, and the real-time image that meets a preset first threshold is marked as an image to be processed.

Preferably, step 2 includes:

Acquire the image to be processed in the real-time image set, and at the same time, select an adaptive feature extraction strategy from a preset strategy database;

Based on the feature extraction strategy, select a suitable feature extraction method from the preset strategy-method comparison table;

Based on the feature extraction method, feature extraction is performed on the image to be processed to generate defect feature information.

Preferably, step 2 further includes:

Based on the defect feature information, obtaining model information in a preset feature-model matching table, and selecting a corresponding preset defect recognition model in a preset model database based on the model information;

The defect feature information is input into the preset defect recognition model to perform defect recognition and obtain a first result.

Preferably, step 3 includes:

Based on the defect feature information and the first result, historical defect image information with a matching degree greater than a first matching degree is screened in a historical defect database;

Classifying the historical defect image information, and establishing a historical defect image set based on the classification results;

At the same time, a preset image comparison model is used to compare and analyze the first result and the historical defect image set to obtain a second result.

Preferably, step 3 further includes:

Performing content analysis on the defect feature information, and establishing a defect feature data packet based on the analyzed content;

Classify and analyze all defect features in the defect feature data package, and establish a defect feature classification table;

Combined with the preset category-factor matching table, the first screening factor of each feature type in the defect feature classification table is obtained;

At the same time, the defect feature data packet is input into a preset feature analysis model for parameter calculation to obtain feature parameters corresponding to each defect feature;

Based on the characteristic parameters of each defect characteristic, a second screening factor for each defect characteristic parameter is obtained using a preset parameter-factor comparison table;

In combination with the first screening factor and the second screening factor, historical defect image information having a matching degree greater than a first matching degree is screened in the historical defect database;

According to the characteristic information carried in the historical defect image information, the historical defect image information is classified and analyzed, and a historical defect image set is established according to the classification analysis result and the characteristic parameters of the corresponding defect characteristics;

Extract the defect features and corresponding feature parameters under the same classification in the defect feature classification table and the historical defect image set to establish a comparison data package;

In combination with the first result, the comparison data packet corresponding to each feature category is input into a preset image comparison analysis model for comparison analysis to obtain a second result.

Preferably, step 4 includes:

Performing cluster analysis on each defect feature in the defect feature information, and at the same time, counting the number of defect features under the same category in the cluster analysis results, and arranging them in descending order according to the number of defect features contained in each category to obtain a feature descending order table;

Extracting feature categories whose ordinal numbers are greater than the first ordinal number in the feature descending table, and obtaining a first screening parameter according to a preset category-parameter comparison table;

At the same time, each defect feature in the defect feature information is prioritized against a preset feature-priority comparison table to obtain a priority corresponding to each defect feature;

Based on each defect feature and the corresponding priority, a second screening parameter corresponding to each defect feature is obtained in combination with a preset priority-parameter comparison table;

Inputting the first screening parameter and the second screening parameter into a preset model database for model matching, and obtaining a defect analysis model with a matching degree greater than the second matching degree;

Acquire first category information corresponding to the first result, and bind the first category information to the first result to obtain first binding information;

The first binding information and the second result are input into the defect analysis model, and parameter analysis is performed on feature parameters under the same feature category to obtain a final recognition result.

Preferably, based on the feature extraction method, feature extraction is performed on the image to be processed to generate defect feature information, including:

intercepting a plurality of square grayscale images of different preset sizes and a preset filter function from the image to be processed;

Perform filtering processing on each of the grayscale images at m scales and n preset directions using a preset filtering function to obtain m*n filtering kernels corresponding to each of the square grayscale images;

Fusing the m*n filter kernels of all the square grayscale images to obtain a first texture image;

Obtaining a brightness mean value ave1 of the first texture image and a brightness mean value ave2 of the corresponding real-time image respectively;

The real-time image is adjusted by the brightness mean value ave1 to obtain a first image to be analyzed, and at the same time, the first texture image is adjusted by the brightness mean value ave2 to obtain a second image to be analyzed;

When the brightness value r1 _i of the i-th pixel in the real-time image satisfies the condition ave1-Δ1≤r1 _i ≤ave1+Δ1, the i-th pixel is replaced with the brightness mean ave1, otherwise, the i-th pixel is kept unchanged, and the first image to be analyzed is obtained, where Δ1 represents the brightness variance obtained when there is no defect on the surface of the recycled aluminum alloy template;

At the same time, when the brightness value r2 _j of the j-th pixel in the first texture image satisfies ave2-Δ2≤r2 _j ≤ave2+Δ2, the j-th pixel is replaced with the brightness mean ave2, otherwise, the j-th pixel is kept unchanged, and then the second image to be analyzed is obtained, where Δ2 represents the brightness variance of all the same coordinates where defects are preliminarily determined to exist in the first texture image and the real-time image;

Acquire a first brightness variance σ1 ² between the first image to be analyzed and the real-time image, and at the same time, acquire a second brightness variance σ2 ² between the second image to be analyzed and the first texture image;

According to the brightness mean ave1, the brightness mean ave2, the first brightness variance σ1 ² and the second brightness variance σ2 ² , the image quality W of the first texture image is obtained;

Wherein, ∝1, ∝2, ∝3 represent importance parameters; N1 represents the number of pixel brightness replacements in the first image to be analyzed; N3 represents the total number of pixels in the first image to be analyzed; N2 represents the number of pixel brightness replacements in the second image to be analyzed; N4 represents the total number of pixels in the second image to be analyzed;

If the image quality W is greater than or equal to a threshold quality, performing feature extraction based on the first texture image;

Otherwise, according to the difference between the image quality W and the threshold quality, the adjustment clarity is determined from the difference-clarity adjustment mapping table, and at the same time, the average grayscale value and the average blur value of each square grayscale image are respectively obtained, and the weak texture area and the detail texture area are obtained based on the adjustment clarity.

For the weak texture area, Perform the first definition magnification adjustment and adjust the detail texture area according to/> Perform a second definition multiple adjustment to obtain a second texture image, wherein G1(avehd,avemh) represents an adjustment function based on the average gray value avehd and the average blur value avemh; β1 and β2 represent adjustment weights; Δ(mh) represents a variable based on the blur mh; Δ(hd,bd) represents a variable based on the gray value hd and the standard gray value bd; G2(avehd,avemh,maxmh,avemh1) represents an adjustment function based on the average gray value avehd, the average blur value avemh, the maximum blur maxmh and the average blur avemh1 that satisfies the normal distribution probability; [] represents a rounding symbol;

Feature extraction is performed based on the second texture image.

The implementation principle and beneficial effects of the present invention are as follows: the present invention firstly improves the image quality of the real-time image by preprocessing the acquired aluminum alloy template image, and then extracts the defect features of the image to be processed through a preset filter function, thereby generating defect feature information and a metal surface image; then, the defects on the template surface are identified in a set defect recognition model to obtain a first result, and at the same time, a comparison and analysis is performed with the matching historical defect image to generate a second result, and the current defects are judged with reference to the historical data, which greatly improves the recognition accuracy of the defects, and then the final defect recognition result is obtained through comprehensive analysis and processing of the first result and the second result, and the defects are identified and analyzed from multiple perspectives, which reduces the possibility of misjudgment, thereby greatly improving the detection accuracy of the defects and improving the quality control level of the recycled aluminum alloy template.

Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.

The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments.

BRIEF DESCRIPTION OF THE DRAWINGS

The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

FIG1 is a schematic flow chart of a method for identifying surface defects of a recycled aluminum alloy template according to an embodiment of the present invention.

Detailed ways

The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

1 , an embodiment of the present invention provides a method for identifying surface defects of a recycled aluminum alloy template, comprising:

Step 1: Obtain a real-time image of the template through an image acquisition device, pre-process the real-time image, and obtain an image to be processed;

Step 2: extract features from the image to be processed, generate defect feature information, and input the defect feature information into a preset defect recognition model to perform defect recognition, thereby obtaining a first result;

Step 3: Obtain historical defect image information of the template and establish a historical defect image set. At the same time, input the first result and the historical defect image set into a preset image comparison model for comparison and analysis to obtain a second result;

Step 4: Based on the defect feature information, a defect analysis model is obtained by screening in a preset model database, and the first result and the second result are input into the defect analysis model for defect analysis to obtain a final recognition result.

In this embodiment, the image acquisition device: a device for acquiring image information of the recycled aluminum alloy template, such as a camera, a video camera, a thermal imager, etc.;

In this embodiment, the real-time image: an image of the recycled aluminum alloy template captured by an image acquisition device;

In this embodiment, preprocessing: before subsequent processing and analysis of the image, a series of processing steps are performed on the original image to improve the quality of the image, reduce noise, enhance the features in the image, etc., including but not limited to: image denoising, image enhancement, image sharpening, etc.;

In this embodiment, the image to be processed is: an image obtained by preprocessing the real-time image;

In this embodiment, feature extraction refers to the operation of extracting representative features from an image. The image features may be numerical representations of the local structure, texture, shape, color, etc. of the image.

In this embodiment, the defect feature information includes information of all image features generated after feature extraction;

In this embodiment, the preset defect recognition model: a model used to recognize defects in the template image according to the input defect feature information, is pre-trained with big data;

In this embodiment, the first result: a recognition result generated by defect recognition using a preset defect recognition model;

In this embodiment, the historical defect image information: that is, the historical defect image of the template;

In this embodiment, the historical defect image set: an image set consisting of a plurality of historical defect image information;

In this embodiment, the preset image comparison model: the model used for comparing and analyzing the input first result and the historical defect image set is preset;

In this embodiment, the second result: that is, the analysis result obtained by comparing and analyzing the first result and the historical defect image set through a preset image comparison model;

In this embodiment, the preset model database: a database containing a large number of data processing models for processing and analyzing image data;

In this embodiment, the defect analysis model is a model obtained by screening from a preset model database and used to perform defect analysis on the first result and the second result;

In this embodiment, the final recognition result is: the analysis result obtained after performing defect analysis on the first result and the second result through the defect analysis model.

The working principle and beneficial effects of the above technical scheme are as follows: the present invention firstly improves the image quality of the real-time image by preprocessing the acquired aluminum alloy template image, and then extracts the defect features of the processed image through a preset filter function, thereby generating defect feature information and a metal surface image; then, the defects on the template surface are identified in a set defect recognition model to obtain a first result, and at the same time, a comparison and analysis is performed with the matching historical defect image to generate a second result, and the current defects are judged with reference to the historical data, which greatly improves the recognition accuracy of the defects, and then the final defect recognition result is obtained through comprehensive analysis and processing of the first result and the second result, and the defects are identified and analyzed from multiple perspectives, which reduces the possibility of misjudgment, thereby greatly improving the detection accuracy of defects and improving the quality control level of the recycled aluminum alloy template.

The embodiment of the present invention provides a method for identifying surface defects of a recycled aluminum alloy template, wherein step 1 comprises:

The image acquisition device is used to acquire images of the template surface from a preset angle to obtain a real-time image, and a real-time image set is established according to the time sequence of the real-time images;

Each real-time image in the real-time image set is preprocessed, and the real-time image that meets a preset first threshold is marked as an image to be processed.

In this embodiment, the preset angle: that is, the pre-set shooting angle, can capture images of the template surface from multiple spatial angles, reducing the possibility of misjudgment of defects;

In this embodiment, the time sequence refers to the sequence of the real-time image shooting time;

In this embodiment, the real-time image set: an image set in which a plurality of groups of real-time images are arranged in the order of shooting time;

In this embodiment, a first threshold is preset: a threshold used to determine whether the image after preprocessing meets a preset condition, for example, thresholds for parameters such as image size, signal-to-noise ratio, and image contrast are preset.

The working principle and beneficial effects of the above technical solution are as follows: the present invention can improve the recognition accuracy of defect characteristics on the surface of the aluminum alloy template by collecting images of the surface of the recycled aluminum alloy template from multiple preset angles, and reduce the possibility of misjudgment in defect recognition from a single angle. At the same time, by preprocessing the real-time image set established in chronological order, not only the image quality is improved, but also it can be determined whether there is an error in the current image by comparing the images before and after time, further reducing the possibility of misjudgment, thereby improving the recognition accuracy of aluminum alloy template defects.

The embodiment of the present invention provides a method for identifying surface defects of a recycled aluminum alloy template, wherein step 2 comprises:

Obtain the image to be processed in the real-time image set, and at the same time, select the adapted feature extraction strategy from the preset strategy database;

Based on the feature extraction strategy, select the appropriate feature extraction method from the preset strategy-method comparison table;

Based on the feature extraction method, features are extracted from the image to be processed to generate defect feature information.

In this embodiment, the preset strategy database: a database containing a large number of strategies for extracting image features, which is preset;

In this embodiment, the feature extraction strategy is a strategy obtained by matching from a preset strategy database and used to extract image features from the image to be processed;

In this embodiment, the preset strategy-method comparison table: a table containing mapping relationships between feature extraction strategies and feature extraction methods, used to assign corresponding feature extraction methods to corresponding images to be processed according to the input feature extraction strategies;

In this embodiment, the feature extraction method is a method for extracting features of the image to be processed obtained by inputting the feature extraction strategy into a preset strategy-method comparison table.

The working principle and beneficial effects of the above technical solution are as follows: the present invention first obtains a strategy for feature extraction of the image to be processed in the real-time image set by screening in a preset strategy database, and can match the corresponding feature extraction strategy according to the type or feature of the image, and then obtain the corresponding image feature extraction method through a preset strategy-method comparison table, which not only increases the compatibility with the image to be processed, but also can adapt more accurate feature extraction strategies and methods according to the image category, thereby improving the extraction efficiency and recognition accuracy of image features.

The embodiment of the present invention provides a method for identifying surface defects of a recycled aluminum alloy template, wherein step 2 further includes:

Based on the defect feature information, model information is obtained in a preset feature-model matching table, and a corresponding preset defect recognition model is selected in a preset model database based on the model information;

The defect feature information is input into a preset defect recognition model to perform defect recognition and obtain a first result.

In this embodiment, the preset feature-model matching table: a matching table including mapping relationships between defect features and defect recognition models, which is preset and used to obtain corresponding defect recognition model information according to input defect feature information;

In this embodiment, model information: information of a defect recognition model obtained by inputting defect feature information into a preset feature-model matching table is used to screen and obtain a corresponding defect recognition model in a preset model database.

The working principle and beneficial effects of the above technical solution are as follows: the present invention first inputs the defect feature information containing a large number of recycled aluminum alloy template image features into a preset feature-model matching table to obtain model information, and then uses the model information to filter out a defect recognition model that matches the defect feature information in a preset model database, so that the defect feature information is highly matched with the defect recognition model, thereby increasing the accuracy of defect recognition.

The embodiment of the present invention provides a method for identifying surface defects of a recycled aluminum alloy template, wherein step 3 comprises:

Based on the defect feature information and the first result, historical defect image information with a matching degree greater than the first matching degree is screened in the historical defect database;

Classify the historical defect image information into defects, and establish a historical defect image set based on the classification results;

At the same time, a preset image comparison model is used to compare and analyze the first result and the historical defect image set to obtain a second result.

In this embodiment, the historical defect database is a database storing a large number of aluminum alloy template defect images, which is used to compare with the defect characteristics of the current aluminum alloy template;

In this embodiment, the first matching degree is a threshold used to screen the historical defect images matching the defect feature information and the first result in the historical defect database, which is preset;

In this embodiment, defect classification: the operation of classifying historical defect image information according to its defect category facilitates better understanding and management of the defects of the aluminum alloy template.

The working principle and beneficial effects of the above technical solution are as follows: the present invention obtains historical defect image information that meets preset conditions by screening in a historical defect database according to the defect feature information of the recycled aluminum alloy template and the first result, and then divides the historical defect information according to the defect type to facilitate subsequent comparison with the first result, and then compares and analyzes the historical defect image set after type division with the first result through a preset image comparison model, and analyzes the current recognition result with the historical defect image as a reference, thereby reducing the possibility of misjudgment, wrong judgment, and missed judgment, thereby improving the accuracy of defect recognition.

The embodiment of the present invention provides a method for identifying surface defects of a recycled aluminum alloy template, wherein step 3 further includes:

Analyze the defect feature information and create a defect feature data package based on the analyzed content;

Classify and analyze all defect features in the defect feature data package, and establish a defect feature classification table based on the analysis results;

Combined with the preset category-factor matching table, the first screening factor of each feature type in the defect feature classification table is obtained;

At the same time, the defect feature data packet is input into the preset feature analysis model for parameter calculation to obtain the feature parameters corresponding to each defect feature;

Based on the characteristic parameters of each defect characteristic, a second screening factor for each defect characteristic parameter is obtained using a preset parameter-factor comparison table;

In combination with the first screening factor and the second screening factor, historical defect image information having a matching degree greater than the first matching degree is screened out in the historical defect database;

According to the characteristic information carried in the historical defect image information, the historical defect image information is classified and analyzed, and a historical defect image set is established according to the classification analysis results and the characteristic parameters of the corresponding defect characteristics;

Extract the defect features and corresponding feature parameters under the same classification in the defect feature classification table and the historical defect image set, and establish a comparison data package;

Combined with the first result, the comparison data packet corresponding to each feature category is input into a preset image comparison analysis model for comparison analysis to obtain a second result.

In this embodiment, content analysis refers to an operation of analyzing defect feature information to obtain information content therein;

In this embodiment, the defect feature data packet is: a data packet generated by converting the analyzed content obtained by performing content analysis on the defect feature information;

In this embodiment, classification analysis: the operation of classifying and analyzing all defect features contained in the defect feature data packet according to preset standards, so as to better understand the nature, cause and impact of the defect and facilitate subsequent improvements;

In this embodiment, the defect feature classification table is a classification table established according to the analysis results generated after classifying and analyzing all defect features in the defect feature data packet, and includes defect features of different categories;

In this embodiment, the preset category-factor matching table: a table containing mapping relationships between defect categories and first screening factors, used to obtain corresponding first screening factors according to the category of the input defect feature, is preset;

In this embodiment, the first screening factor: a screening condition for screening the historical defect image information in the historical defect database;

In this embodiment, the preset feature analysis model: a model used to calculate parameters of an input defect feature data packet is preset;

In this embodiment, parameter calculation: an operation of measuring a parameter value corresponding to each defect feature in a defect feature data packet;

In this embodiment, characteristic parameters: i.e., parameter values corresponding to each defect characteristic, such as size, shape, location, distribution, quantity, severity, etc.;

In this embodiment, the preset parameter-factor comparison table: a comparison table including a mapping relationship between characteristic parameters and second screening factors, used to obtain the corresponding second screening factors according to the input defect characteristic parameters;

In this embodiment, the second screening factor: a screening condition for screening the historical defect image information in the historical defect database, corresponding to the first screening factor;

In this embodiment, the feature information: defect feature information contained in the historical defect image information;

In this embodiment, the same classification refers to the intersection of the defect feature classification table and the historical defect image set, in other words, the same defect feature category in the defect feature classification table and the historical defect image set;

In this embodiment, the comparison data packet is a data packet generated by extracting defect features and corresponding feature parameters in the defect feature classification table and the historical defect image set that are included in the same defect feature category.

The working principle and beneficial effects of the above technical scheme are as follows: the present invention first establishes a defect feature data packet for data processing according to the parsed content of the current defect feature information, then classifies and analyzes the defect features in the defect feature data packet to obtain a defect feature classification table, and then obtains a first screening factor according to the defect-category matching table, and at the same time uses a preset feature analysis model to calculate the feature parameters of the defect, and then obtains a second screening factor according to a preset parameter-factor comparison table, and then combines the first screening factor and the second screening factor to screen out historical reference image information that meets preset conditions in the historical defect database, and further compares and analyzes the current defect features and historical defect features and corresponding feature parameters under the same classification through a preset image comparison analysis model to generate a second result, which greatly improves the recognition accuracy of the defect features, and then improves the accuracy of template defect recognition, thereby ensuring the quality control level of recycled aluminum alloy production.

The embodiment of the present invention provides a method for identifying surface defects of a recycled aluminum alloy template, wherein step 4 comprises:

Perform cluster analysis on each defect feature in the defect feature information. At the same time, count the number of defect features in the same category in the cluster analysis results, and arrange them in descending order according to the number of defect features contained in each category to obtain a feature descending order table.

Extract the feature categories whose ordinal numbers are greater than the first ordinal number in the feature descending table, and obtain the first screening parameter according to the preset category-parameter comparison table;

At the same time, each defect feature in the defect feature information is prioritized against a preset feature-priority comparison table to obtain the priority corresponding to each defect feature;

Based on each defect feature and the corresponding priority, a second screening parameter corresponding to each defect feature is obtained in combination with a preset priority-parameter comparison table;

Inputting the first screening parameter and the second screening parameter into a preset model database for model matching, and obtaining a defect analysis model with a matching degree greater than the second matching degree;

Obtaining first category information corresponding to the first result, and binding the first category information with the first result to obtain first binding information;

The first binding information and the second result are input into the defect analysis model, and parameter analysis is performed on the feature parameters under the same feature category to obtain the final recognition result.

In this embodiment, cluster analysis: the operation of analyzing and processing similar defect features helps to discover patterns and associations between defects;

In this embodiment, the same category: that is, the same cluster category;

In this embodiment, the number of defect features is: the sum of the number of all defect features in the same category in the cluster analysis results;

In this embodiment, descending order is used to arrange the clustering results in descending order of the number of defect features contained in each category;

In this embodiment, the feature descending table is a category sorting table generated by arranging all categories in descending order according to the number of defect features contained;

In this embodiment, the first ordinal number is used to extract the threshold value that meets the preset condition in the feature descending table. For example, if the first ordinal number is 3, the defect categories with ordinal numbers 1 and 2 are extracted, and the rest are not extracted.

In this embodiment, the preset category-parameter comparison table: a comparison table containing mapping relationships between feature categories and first screening parameters, which is preset and used to obtain corresponding first screening parameters according to the input defect feature category;

In this embodiment, the first screening parameter is used to screen the parameters of the defect analysis model that meets the preset conditions in the preset model database;

In this embodiment, the preset feature-priority comparison table: a comparison table containing mapping relationships between defect features and priorities, which is preset and used to obtain corresponding priorities according to input defect features;

In this embodiment, priority: the defect feature is input into the preset feature-priority comparison table to obtain the corresponding priority level. The higher the priority, the more important the defect feature is, and the higher the impact ratio among all defect features is.

In this embodiment, the preset priority-parameter comparison table: a comparison table including a mapping relationship between priorities and second screening parameters, used to obtain the corresponding second screening parameters according to the input priority;

In this embodiment, the second screening parameter is used to screen the parameters of the defect analysis model that meets the preset conditions in the preset model database, corresponding to the first screening parameter;

In this embodiment, model matching: a matching process of obtaining a matching defect analysis model by screening in a preset model database according to the input first screening parameter and second screening parameter;

In this embodiment, the second matching degree is used to filter out a threshold value of a defect analysis model that meets the preset conditions in the preset model database. For example, if the second matching degree is 90%, only defect analysis models with a matching degree greater than 90% can be selected.

In this embodiment, the first category information: that is, the defect category corresponding to the defect recognition result;

In this embodiment, the first binding information is information generated by binding the first result with the corresponding first category information;

In this embodiment, parameter analysis: an operation of comparing and analyzing the characteristic parameters under the same characteristic category in the first binding information and the second result, so as to obtain the final recognition result of the defect of the recycled aluminum alloy template.

The working principle and beneficial effects of the above technical solution are as follows: the present invention can discover the similarities and differences between defects by clustering the defect features, which is helpful to identify common defect patterns and root causes of problems, thereby continuously improving the recognition accuracy and recognition speed of defects; at the same time, by combining the first screening parameter corresponding to the feature category with an ordinal number greater than the first ordinal number in the feature descending table and the second screening parameter corresponding to the priority, the defect analysis model is accurately matched in the preset model database, thereby improving the adaptability between the model and the data, thereby improving the efficiency of data processing; then, the first binding information and the second result are subjected to parameter analysis to obtain the final recognition result of the defect, which greatly reduces the occurrence of misjudgment and the like, and improves the accuracy of defect recognition for recycled aluminum alloy templates, thereby improving the quality control level of aluminum alloy production.

The embodiment of the present invention provides a method for identifying surface defects of a recycled aluminum alloy template. Based on a feature extraction method, feature extraction is performed on an image to be processed to generate defect feature information, including:

intercepting a plurality of square grayscale images of different preset sizes and a preset filter function from the image to be processed;

Each grayscale image is filtered at m scales and n preset directions by using a preset filter function to obtain m*n filter kernels corresponding to each square grayscale image;

The m*n filter kernels of all square grayscale images are fused to obtain a first texture image;

Obtain the brightness mean value ave1 of the first texture image and the brightness mean value ave2 of the corresponding real-time image respectively;

The real-time image is adjusted with the brightness mean value ave1 to obtain the first image to be analyzed, and at the same time, the first texture image is adjusted with the brightness mean value ave2 to obtain the second image to be analyzed;

When the brightness value r1 _i of the i-th pixel in the real-time image satisfies the condition ave1-Δ1≤r1 _i ≤ave1+Δ1, the i-th pixel is replaced with the brightness mean ave1, otherwise, the i-th pixel is kept unchanged, and then the first image to be analyzed is obtained, where Δ1 represents the brightness variance obtained when there is no defect on the surface of the recycled aluminum alloy template;

At the same time, when the brightness value r2 _j of the j-th pixel in the first texture image satisfies ave2-Δ2≤r2 _j ≤ave2+Δ2, the j-th pixel is replaced with the brightness mean ave2, otherwise, the j-th pixel is kept unchanged, and then the second image to be analyzed is obtained, where Δ2 represents the brightness variance of all the same coordinates where defects are preliminarily determined to exist in the first texture image and the real-time image;

Obtain a first brightness variance σ1 ² between the first image to be analyzed and the real-time image, and at the same time, obtain a second brightness variance σ2 ² between the second image to be analyzed and the first texture image;

According to the brightness mean ave1, the brightness mean ave2, the first brightness variance σ1 ² and the second brightness variance σ2 ² , the image quality W of the first texture image is obtained;

Wherein, ∝1, ∝2, ∝3 represent importance parameters; N1 represents the number of pixel brightness replacements in the first image to be analyzed; N3 represents the total number of pixels in the first image to be analyzed; N2 represents the number of pixel brightness replacements in the second image to be analyzed; N4 represents the total number of pixels in the second image to be analyzed;

If the image quality W is greater than or equal to the threshold quality, feature extraction is performed based on the first texture image;

Otherwise, according to the difference between the image quality W and the threshold quality, the adjustment clarity is determined from the difference-clarity adjustment mapping table. At the same time, the average grayscale value and the average blur value of each square grayscale image are obtained respectively, and the weak texture area and the detail texture area are obtained based on the adjustment clarity.

For weak texture areas, Perform the first definition magnification adjustment and adjust the detail texture area according to/> Perform a second definition multiple adjustment to obtain a second texture image, wherein G1(avehd,avemh) represents an adjustment function based on the average gray value avehd and the average blur value avemh; β1 and β2 represent adjustment weights; Δ(mh) represents a variable based on the blur mh; Δ(hd,bd) represents a variable based on the gray value hd and the standard gray value bd; G2(avehd,avemh,maxmh,avemh1) represents an adjustment function based on the average gray value avehd, the average blur value avemh, the maximum blur maxmh and the average blur avemh1 that satisfies the normal distribution probability; [] represents a rounding symbol;

Feature extraction is performed based on the second texture image.

In this embodiment, the preset size is a preset size of a square grayscale image, such as 4*4, 5*5, 6*6, etc.;

In this embodiment, the square grayscale image is an image having a square shape and containing grayscale information obtained by intercepting the image to be processed;

In this embodiment, the preset filter function: the function used for filtering the grayscale image in each direction is preset, such as the Gabor function used for image texture feature extraction;

In this embodiment, the filter kernel is a matrix generated according to a preset filter function and used to perform a filtering operation on each square grayscale image, which is consistent with the size of the directional grayscale image. For example, if the size of the square grayscale image is 4*4, the size of the filter kernel is also 4*4, and is used to calculate the elements at the same position in the square grayscale image;

In this embodiment, the first texture image is an image containing texture features of the recycled aluminum alloy template obtained by fusing m*n filter kernels of all square grayscale images;

In this embodiment, the first image to be analyzed is: an image obtained by adjusting the real-time image by the brightness mean value of the first texture image;

In this embodiment, the second image to be analyzed is: an image obtained by adjusting the first texture image by the brightness mean value of the real-time image;

In this embodiment, pixel: the most basic unit in an image, which is the smallest visible element in the image. Each pixel has a specific position and contains image attribute information such as color, transparency, grayscale, depth, etc. For example, if the size of an image is 5*5, then the image contains 5*5=25 pixels;

In this embodiment, threshold quality: a threshold used to determine whether the image quality W meets a preset condition;

In this embodiment, the difference value is the numerical difference between the two when the image quality is less than the threshold quality;

In this embodiment, the difference-definition adjustment mapping table: a mapping table including a mapping relationship between a difference and an adjustment definition, and used to obtain a corresponding adjustment definition according to an input difference;

In this embodiment, the average gray value is the average value obtained by averaging the sum of the gray values of all pixels in each square gray image;

In this embodiment, the average fuzzy value is the average of the fuzzy values of all pixels in each square grayscale image;

In this embodiment, the fuzzy value: the average value of the pixel values of the area around each pixel in the image is taken as the fuzzy value of the pixel;

In this embodiment, weak texture area: an area in the image that is relatively lacking in obvious texture features or has unclear texture details, for example, an area where the grayscale difference between each pixel is small;

In this embodiment, the detailed texture area includes an area with obvious texture or rich texture details, for example, an area with a large grayscale difference between each pixel;

In this embodiment, the definition adjustment is to continue to zoom in or zoom out the image to adjust the definition of the image;

In this embodiment, the second texture image is a texture image obtained by performing a first definition magnification adjustment on the weak texture area and a second definition magnification adjustment on the detail texture area.

The working principle and beneficial effects of the above technical solution are as follows: the present invention uses a preset filter function to filter a plurality of square grayscale images of different sizes, thereby obtaining a first texture image containing the texture feature information of the recycled aluminum alloy template, and then analyzing and processing the brightness values corresponding to the first texture image and the real-time image, respectively, to obtain the image quality of the first texture image, which not only greatly improves the image quality of the texture image, but also reduces the difficulty of subsequent texture recognition. At the same time, the present invention can further improve the image quality of the first texture image by adjusting the clarity of the weak texture area and the detail texture area, thereby obtaining a second texture image with higher clarity, further reducing the difficulty of subsequent texture feature extraction.

Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims (6)

1. The method for identifying the surface defects of the regenerated aluminum alloy template is characterized by comprising the following steps of:

Step 1: acquiring a real-time image of a template through image acquisition equipment, and preprocessing the real-time image to obtain an image to be processed;

Step 2: extracting features of the image to be processed to generate defect feature information, and inputting the defect feature information into a preset defect recognition model to perform defect recognition to obtain a first result;

step 3: acquiring historical defect image information of a template, establishing a historical defect image set, and simultaneously inputting the first result and the historical defect image set into a preset image comparison model for comparison analysis to obtain a second result;

Step 4: screening in a preset model database based on the defect characteristic information to obtain a defect analysis model, and inputting the first result and the second result into the defect analysis model to perform defect analysis to obtain a final identification result;

Wherein, step 2 includes:

Acquiring images to be processed in the real-time image set, and screening out an adaptive feature extraction strategy from a preset strategy database;

selecting a proper feature extraction method from a preset strategy-method comparison table based on the feature extraction strategy;

Based on the feature extraction method, feature extraction is carried out on the image to be processed, and defect feature information is generated, and the method comprises the following steps:

Intercepting a plurality of square gray images with different preset sizes from the image to be processed and presetting a filtering function;

Filtering each gray image in m scales and n preset directions through a preset filter function to obtain m x n filter kernels corresponding to each square gray image;

Fusing all m x n filter kernels of the square gray images to obtain a first texture image;

respectively acquiring a brightness average ave1 of the first texture image and a brightness average ave2 of a corresponding real-time image;

Adjusting the real-time image with the brightness average ave1 to obtain a first image to be analyzed, and simultaneously adjusting the first texture image with the brightness average ave2 to obtain a second image to be analyzed;

When the brightness value of the ith pixel point in the real-time image Satisfy/>Replacing the ith pixel point with a brightness average ave1 under the condition, otherwise, keeping the ith pixel point unchanged, and further obtaining a first image to be analyzed, wherein/>Representing the brightness variance obtained when the surface of the regenerated aluminum alloy template has no defect;

meanwhile, when the brightness value of the jth pixel point in the first texture image Satisfy/>2, Replacing the jth pixel point with a brightness average ave2, otherwise, keeping the jth pixel point unchanged, and further obtaining a second image to be analyzed, wherein/>2, Preliminarily determining the brightness variance of the first texture image and the real-time image under all the same coordinates with defects;

acquiring a first brightness variance of the first image to be analyzed and the real-time image Simultaneously, a second brightness variance/>, of the second image to be analyzed and the first texture image is obtained；

According to the brightness average ave1, the brightness average ave2 and the first brightness varianceSecond luminance variance/>Acquiring the image quality W of the first texture image;

;

Wherein, 、/>、/>Representing an importance parameter; n1 represents the number of pixel brightness substitutions existing in the first image to be analyzed; n3 represents the total number of pixel points in the first image to be analyzed; n2 represents the number of pixel brightness substitutions in the second image to be analyzed; n4 represents the total number of pixel points in the second image to be analyzed;

If the image quality W is greater than or equal to a threshold quality, extracting features based on the first texture image;

otherwise, determining the adjustment definition from a difference-definition adjustment mapping table according to the difference value of the image quality W and the threshold quality, respectively acquiring an average gray value and an average fuzzy value of each square gray image, and analyzing based on the adjustment definition to obtain a weak texture region and a detail texture region;

Following the weak texture region First definition multiple adjustment and detail texture area adjustment according to/>Performing second definition multiple adjustment to obtain a second texture image, wherein/>The representation is based on the average gray value/>And mean blur value/>Is a function of the adjustment of (2); /(I)、/>Representing an adjustment weight; /(I)A variable representing a blur degree mh; /(I)A variable representing a gray value hd and a standard gray value bd; /(I)The representation is based on the average gray value/>Average blur value/>An adjustment function of the maximum ambiguity maxmh and the average ambiguity avemh1 satisfying the normal distribution probability; [ ] Representing a rounding symbol;

And extracting features based on the second texture image.

2. The method for identifying surface defects of a recycled aluminum alloy template according to claim 1, wherein in step 1, the method comprises the following steps:

Image acquisition is carried out on the surface of the template from a preset angle through image acquisition equipment to obtain a real-time image, and a real-time image set is established according to the time sequence of the real-time image;

Preprocessing each real-time image in the real-time image set, and calibrating the real-time image meeting a preset first threshold value as an image to be processed.

3. The method for identifying surface defects of a recycled aluminum alloy template according to claim 1, wherein in step 2, the method further comprises:

Acquiring model information from a preset feature-model matching table based on the defect feature information, and selecting a corresponding preset defect identification model from a preset model database based on the model information;

And inputting the defect characteristic information into the preset defect identification model to identify the defects, so as to obtain a first result.

4. The method for identifying surface defects of a recycled aluminum alloy template according to claim 1, wherein in step3, the method comprises the following steps:

Based on the defect characteristic information and the first result, screening historical defect image information with the matching degree larger than the first matching degree from a historical defect database;

performing defect classification on the historical defect image information, and establishing a historical defect image set based on classification results;

and meanwhile, comparing and analyzing the first result and the historical defect image set by using a preset image comparison model to obtain a second result.

5. The method for identifying surface defects of a recycled aluminum alloy template according to claim 4, wherein in step 3, the method further comprises:

performing content analysis on the defect characteristic information, and establishing a defect characteristic data packet based on analysis content;

classifying and analyzing all defect characteristics in the defect characteristic data packet, and establishing a defect characteristic classification table based on analysis results;

Combining a preset category-factor matching table to obtain a first screening factor of each feature category in the defect feature classification table;

meanwhile, inputting the defect characteristic data packet into a preset characteristic analysis model for parameter calculation to obtain characteristic parameters corresponding to each defect characteristic;

Acquiring a second screening factor of each defect characteristic parameter by using a preset parameter-factor comparison table based on the characteristic parameter of each defect characteristic;

Screening historical defect image information with the matching degree larger than the first matching degree in a historical defect database by combining the first screening factor and the second screening factor;

classifying and analyzing the historical defect image information according to the characteristic information carried in the historical defect image information, and establishing a historical defect image set according to the classifying and analyzing result and the characteristic parameters of the corresponding defect characteristics;

Extracting the defect characteristics under the same classification and corresponding characteristic parameters in the defect characteristic classification table and the historical defect image set, and establishing a comparison data packet;

and inputting the comparison data packet corresponding to each characteristic category into a preset image comparison analysis model for comparison analysis by combining the first result to obtain a second result.

6. The method for identifying surface defects of a recycled aluminum alloy template according to claim 1, wherein in step 4, the method comprises the following steps:

Performing cluster analysis on each defect feature in the defect feature information, simultaneously counting the number of defect features in the same category in a cluster analysis result, and performing descending arrangement according to the number of defect features contained in each category to obtain a feature descending list;

Extracting feature categories with ordinal numbers larger than the first ordinal number in the feature descending list, and obtaining first screening parameters according to a preset category-parameter comparison list;

Meanwhile, judging the priority of each defect feature in the defect feature information and a preset feature-priority comparison table to obtain the priority corresponding to each defect feature;

based on each defect feature and the corresponding priority, acquiring a second screening parameter corresponding to each defect feature by combining a preset priority-parameter comparison table;

inputting the first screening parameters and the second screening parameters into a preset model database for model matching to obtain a defect analysis model with matching degree larger than the second matching degree;

Acquiring first type information corresponding to the first result, and binding the first type information with the first result to obtain first binding information;

Inputting the first binding information and the second result into the defect analysis model, and carrying out parameter analysis on the characteristic parameters under the same characteristic category to obtain a final identification result.