CN110458119B - Non-contact measurement concrete aggregate gradation rapid identification method - Google Patents
Non-contact measurement concrete aggregate gradation rapid identification method Download PDFInfo
- Publication number
- CN110458119B CN110458119B CN201910752374.1A CN201910752374A CN110458119B CN 110458119 B CN110458119 B CN 110458119B CN 201910752374 A CN201910752374 A CN 201910752374A CN 110458119 B CN110458119 B CN 110458119B
- Authority
- CN
- China
- Prior art keywords
- aggregate
- image
- determining
- detected
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Evolutionary Biology (AREA)
- Software Systems (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a non-contact measurement method for quickly identifying concrete aggregate gradation. The method combines digital image processing and deep learning, collects multi-angle pictures of each stone block of each large category as a first training set, obtains aggregate attributes by image processing, further determines the small category to which the aggregates belong to obtain a second training set, and trains an improved convolutional neural network model for multiple times by using a standard sample library consisting of the first training set and the second training set to ensure the accuracy of the improved convolutional neural network model.
Description
Technical Field
The invention relates to the field of aggregate identification, in particular to a non-contact measurement concrete aggregate gradation rapid identification method.
Background
Concrete aggregate refers to a granular loose material that plays a role of a skeleton or filling in concrete. The aggregate is divided into coarse aggregate and fine aggregate. The coarse aggregate refers to pebbles, broken stones and the like, and the fine aggregate refers to natural sand, artificial sand and the like. The aggregate volume accounts for 60-80% in the concrete, and the hydraulic concrete is more than 80%, and the performance of the aggregate has important influence on the performance of the concrete.
The aggregate with the particle size of more than 4.75mm is called coarse aggregate and commonly called stone. Two types of gravel and pebbles are commonly used. The crushed stone is made of natural rock or rock through mechanical crushing and sieving, and has a particle size larger than 4.75 mm. Pebbles are rock particles with a particle size of more than 4.75mm, which are formed by natural weathering, water flow transportation, sorting and stacking. The needle-shaped particles are the cobble and broken stone particles with the length 2.4 times larger than the average particle diameter of the corresponding grade of the particles; the particles having a thickness of less than 0.4 times the average particle diameter are flaky particles (the average particle diameter means the average of the upper and lower limit particle diameters of the fraction).
Aggregates with a particle size of 4.75mm or less are called fine aggregates, commonly called sand. The sand is divided into natural sand and artificial sand according to the production source. The natural sand is rock particles with the particle size of less than 4.75mm formed by natural weathering, water flow transportation and sorting and accumulation, but does not comprise particles of soft rock and weathered rock. Natural sands include river sands, lake sands, mountain sands, and desalinated sea sands. The artificial sand is a general name of machine-made sand and mixed sand which are subjected to soil removal treatment.
For concrete dams, in the process of filling the dam body, effective control of the quality of a damming material is an important measure for ensuring safe and normal operation of the damming material in a construction period and an operation period, and the grading of the damming material, namely the grading of aggregate, is an important standard for directly influencing the quality of the damming material. The grading is the distribution of particles with different particle sizes in the aggregate, and at present, the size of the grading is generally determined by a manual or mechanical screening method. Although the method has mature technology, low equipment cost, intuitive result and simple operation, the efficiency is extremely low, the construction progress is seriously influenced, manual operation is required, the accuracy and the credibility of the method completely depend on the operation level of a tester, and the deformation of the sieve pores can also influence the result to a certain extent.
Disclosure of Invention
The invention aims to provide a concrete aggregate gradation rapid identification method based on non-contact measurement.
In order to achieve the purpose, the invention provides the following scheme:
the invention provides a non-contact measurement concrete aggregate gradation rapid identification method, which comprises the following steps:
collecting multi-angle photos of each stone block of each large category as a first training sample, and generating a standard sample library;
building a convolutional neural network to obtain an initial improved convolutional neural network model;
carrying out primary training on the initial improved convolutional neural network model by using the standard sample library to obtain a trained improved convolutional neural network model;
collecting an aggregate original image, determining the small category of each aggregate in the aggregate original image by adopting an image processing mode to obtain a second training sample, and updating the standard sample library by utilizing the second training sample;
performing secondary training on the improved convolutional neural network model after the primary training by using the updated standard sample library to obtain an improved convolutional neural network model after the secondary training;
collecting an aggregate image to be detected;
inputting the aggregate image to be detected into the improved convolutional neural network model after the secondary training, and determining the classification result of each aggregate in the aggregate image to be detected;
and establishing a grading curve of the aggregate to be detected according to the classification result.
Optionally, the improved convolutional neural network model includes: 1 input layer, 10 convolutional layers, 9 pooling layers, and 1 output layer; the activation function of the improved convolutional neural network model is a Relu activation function.
Optionally, the acquiring a multi-angle photo of each stone block of each large category as a first training sample specifically includes:
selecting a plurality of sample stones from the aggregate;
determining the volume of each sample stone block by adopting a drainage method, and determining the large category of each sample stone block according to the volume of the sample stone blocks;
a multi-angle photograph of each stone block of each large category is obtained as a first training sample.
Optionally, the acquiring an aggregate original image, and determining a small category to which each aggregate in the aggregate original image belongs by using an image processing method to obtain a second training sample specifically includes:
acquiring an aggregate orthographic image and four aggregate oblique images by adopting a five-lens oblique camera set;
selecting an aggregate inclined image with the best quality from the four aggregate inclined images as an input image of a second training sample;
and determining the subclass of each aggregate in the aggregate ortho-image in an image processing mode to serve as a target output result of the second training sample.
Optionally, the determining, in an image processing manner, the subclass to which each aggregate belongs in the aggregate ortho-image is used as a target output result of the second training sample, and specifically includes:
preprocessing the aggregate orthoimage to obtain a preprocessed orthoimage;
carrying out canny operator edge detection on the preprocessed ortho-image, determining the edge of each aggregate image in the preprocessed ortho-image, and obtaining the edge-detected ortho-image;
performing expansion and corrosion treatment on the edge-detected ortho-image, and determining the outline of each aggregate in the edge-detected ortho-image;
determining the contour line of each aggregate in the aggregate orthographic image according to the contour line of each aggregate in the orthographic image after the edge detection;
performing contour fitting on the contour line of each aggregate in the aggregate ortho-image to obtain a fitted minimum circumscribed ellipse of each aggregate in the aggregate ortho-image;
and determining the small category of each aggregate in the original image according to the fitted minimum circumscribed ellipse of each aggregate in the aggregate orthoimage.
Optionally, the acquiring an aggregate image to be detected specifically includes:
acquiring an orthoscopic image of the aggregate to be detected and four inclined images of the aggregate to be detected by adopting a five-lens inclined camera set;
and selecting an image with the best quality from one aggregate orthoscopic image to be detected and four aggregate oblique images to be detected as an aggregate image to be detected.
Optionally, the establishing a grading curve of the aggregate to be tested according to the classification result specifically includes:
determining the number of aggregates of different categories according to the classification result; the categories comprise a large category and a small category;
determining the mass ratio of the aggregates of each category according to the number of the aggregates of different categories;
and drawing an aggregate grading curve according to the mass ratio of the aggregates of each category.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
the invention provides a non-contact measurement method for quickly identifying concrete aggregate gradation. The identification method comprises the following steps: collecting a multi-angle photo of each stone block of each large category as a first training set, generating a standard sample library, and carrying out primary training on an initial improved convolutional neural network model by using the standard sample library; determining the small category of each aggregate in the obtained aggregate original image by adopting an image processing mode to obtain a second training set, and updating the second training set into a standard sample library; performing secondary training on the improved convolutional neural network model after the primary training by using the updated standard sample library; inputting the aggregate image to be detected into an improved convolutional neural network model after secondary training, and determining the classification result of each aggregate in the aggregate image to be detected; and establishing a grading curve of the aggregate to be detected according to the classification result. The invention realizes the problem of fast identification of concrete aggregate gradation measured in a non-contact way, combines digital image processing with deep learning, collects multi-angle pictures of each stone block of each large category as a first training set, obtains aggregate attributes by image processing, further determines the small category to which the aggregates belong, obtains a second training set, trains an improved convolutional neural network model for multiple times by using a standard sample library consisting of the first training set and the second training set, ensures the accuracy of the improved convolutional neural network model, establishes an aggregate gradation curve by using the trained improved convolutional neural network model and the obtained aggregate images, improves the efficiency of gradation identification, and realizes the fast identification of the concrete aggregate gradation measured in a non-contact way.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
FIG. 1 is a flow chart of a non-contact measurement method for rapidly identifying concrete aggregate gradation according to the present invention;
FIG. 2 is a schematic diagram of a non-contact measurement method for rapidly identifying concrete aggregate gradation according to the present invention;
FIG. 3 is an aggregate orthophoto image provided by the present invention; wherein, (a) is an original aggregate orthophoto image; (b) the aggregate orthoimage after binaryzation is obtained; (c) the aggregate orthoimage after Gaussian filtering; (d) is the aggregate orthophoto image after threshold segmentation.
Detailed Description
The invention aims to provide a method for rapidly identifying concrete aggregate gradation through non-contact measurement so as to improve the efficiency and accuracy of aggregate gradation.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
Dam stones are mostly aggregates with various sizes, aiming at the shape characteristic, a digital image photogrammetry technology is used as a background, an imaging device is used for quickly obtaining an aggregate original image and acquiring a large number of training sets, sample classification and arrangement are carried out in combination with machine learning and a model is constructed, then image preprocessing, image segmentation and recognition algorithms are carried out to obtain each attribute value of the aggregate, each aggregate in the aggregate original image is more finely classified in combination with the attribute values, the types of the samples in a standard sample library are more fine while the number of the samples in the standard sample library is increased, the updated standard sample library is used for carrying out secondary training on the integrated neural network model, the training precision is improved, the classification condition of the data to be detected is obtained by the classification model, and the method for evaluating the geometric characteristics and the grading of coarse aggregates is completed, finally obtaining the grading curve of the aggregate.
This scheme adopts image recognition and machine learning, and to the mixture that the aggregate produced behind the gradation can pass through the conveyer belt transportation, just can carry out digital formation of image at conveyer belt transportation mixture motion image with the characteristics that the cementite mixes in the mixing bunker, gathers required mixture image data.
As shown in fig. 1, the invention provides a non-contact measured concrete aggregate gradation rapid identification method, which comprises the following steps:
step 101, collecting a multi-angle photo of each stone block of each large category as a first training sample, and generating a standard sample library.
Specifically, a plurality of sample stones are selected from aggregate; determining the volume of each sample stone block by adopting a drainage method, and determining the large category of each sample stone block according to the volume of the sample stone blocks; photographs of a plurality of angles of each stone of each large category are taken as first training samples, and a standard sample library is generated.
The first training sample contains information such as volume, quality and image.
And 102, building a convolutional neural network to obtain an initial improved convolutional neural network model.
The improved convolutional neural network model comprises: 1 input layer, 10 convolutional layers, 9 pooling layers, and 1 output layer; the activation function of the improved convolutional neural network model is a Relu activation function. The size of the input image of the input layer is 224 x 224.
And 103, performing primary training on the initial improved convolutional neural network model by using the standard sample library to obtain a trained improved convolutional neural network model.
During training, training samples in a standard sample library are divided into a training set (train data is 70%), a verification set (validation data is 15%) and a test set (test data is 15%). The number of classes is chosen as training momentum (momentum term ═ 0.925). The number of iterations of the training is set to 8000 and the number of input samples per iteration is set to 128. The learning rate is set to 0.0001 and the weight decay rate is set to 0.005.
And 104, acquiring an aggregate original image, determining the small category of each aggregate in the aggregate original image by adopting an image processing mode to obtain a second training sample, and updating the standard sample library by utilizing the second training sample.
Acquiring an aggregate orthographic image and four aggregate oblique images by adopting a five-lens oblique camera set; selecting an aggregate inclined image with the best quality from the four aggregate inclined images as an input image of a second training sample; and determining the subclass of each aggregate in the aggregate ortho-image in an image processing mode to serve as a target output result of the second training sample.
Determining the subclass of each aggregate in the aggregate ortho-image in an image processing mode, wherein the subclass is used as a target output result of a second training sample, and the method specifically comprises the following steps:
1) and preprocessing the aggregate orthoimage to obtain a preprocessed orthoimage.
And (3) preprocessing operation: performing binarization processing on the orthophoto image, and separating an interested target from a background; and performing noise reduction processing on the image subjected to binarization processing by using Gaussian filtering: noise in the image is reduced, image contrast is increased, analysis accuracy is improved, and visual effect is improved; and performing segmentation of each aggregate on the image by using threshold segmentation according to the gray value of the image pixel, and segmenting the aggregates in the image from the background.
2) And carrying out canny operator edge detection on the preprocessed ortho-image, determining the edge of each aggregate image in the preprocessed ortho-image, and obtaining the edge-detected ortho-image.
And (3) carrying out canny operator edge detection on the preprocessed image: the purpose of canny edge detection is to find a set of pixels with severe brightness change in the preprocessed image, which is often a contour. If the edges in the image can be accurately measured and located, this means that the actual aggregate can be located and measured, including the area of the aggregate, the diameter of the aggregate, the shape of the object, etc. Canny edge detection is mainly performed in four steps:
firstly, filtering an image by using a Gaussian filter; secondly, carrying out convolution on the Sobel horizontal and vertical detectors and the image to calculate a gradient and a direction angle; thirdly, non-maximum value inhibition, namely searching the local maximum value of a pixel point in the image, and setting the gray value corresponding to the non-maximum value point as 0, so that most non-edge points can be eliminated; the fourth step of hysteresis thresholding often has the problem of breaking at edges that should be continuous due to the influence of noise. Hysteresis thresholding sets two thresholds: one high threshold Th and one low threshold Tl. If the effect of any pixel edge operator exceeds a high threshold, marking those pixels as edges; pixels that affect more than the low threshold (between the high and low thresholds) are also marked as edges if 4-or 8-adjacent to the pixel that has been marked as an edge. Namely, it is
If the gradient value of the pixel is less than Tl, the pixel is a non-edge pixel;
if the gradient value of the pixel is larger than Th, the pixel is an edge pixel;
if the gradient value of the pixel is between Tl and Th, 8 points in the 3 × 3 neighborhood of the pixel need to be further detected, if one or more point gradients in the 8 points exceed Th, the pixel is an edge pixel, otherwise, the pixel is not an edge pixel.
3) And performing expansion and corrosion treatment on the ortho-image after the edge detection, and determining the outline of each aggregate in the ortho-image after the edge detection.
Carrying out expansion and corrosion operations on the ortho-image after edge detection: dilation is the operation of finding a local maximum, kernel B (kernel can be of any shape and size, which has a separately defined reference point, which we call anchor point) in most cases, kernel is a small middle with a reference point and a solid square or disk, in which we can consider kernel as template or mask) convolved with the image, i.e. the maximum of the pixel points in the area covered by kernel B is calculated and assigned to the pixel specified by the reference point. This results in a gradual increase in the highlight areas in the image. Erosion is the opposite of swelling and erosion is the operation of finding a local minimum. And (4) performing expansion and corrosion operations on the image subjected to canny edge detection to obtain a more obvious aggregate profile.
4) And determining the contour line of each aggregate in the aggregate orthoimage according to the contour line of each aggregate in the orthoimage after the edge detection.
The outline of the aggregate of the ortho-image after edge detection is detected by using a findContours function: the contour lines of the aggregates are found in the edge map and the detected contours are ordered from left to right using the sort _ constraints function.
5) And performing contour fitting on the contour line of each aggregate in the aggregate ortho-image to obtain a fitted minimum circumscribed ellipse of each aggregate in the aggregate ortho-image.
Establishing circulation for one pair of detected outline fitting minimum external ellipses to obtain the related attribute data of the aggregate: and calling a minAreaRect function in a computer open source vision library to fit a minimum external rectangle of the aggregate, obtaining 4 vertex coordinates (x, y), width, height and rotation angle of the minimum external rectangle, and further calculating the area of the fitted external rectangle. And calling a fitEllipse function to fit a minimum external ellipse of the aggregate to obtain the center point coordinate, the length of the long axis and the length of the short axis of the external ellipse, thereby calculating the area of the external ellipse for fitting the aggregate.
6) And determining the category of each aggregate in the original image according to the fitted minimum circumscribed ellipse of each aggregate in the aggregate orthoimage.
The key point of the image processing is to extract the outline boundary of the independent aggregate so as to realize the detection and estimation of the characteristic parameters. For the aggregate particles which are distributed discretely, the gray level of stones is obviously higher than that of the surrounding background, and a good segmentation result can be obtained by using a threshold segmentation method. As shown in fig. 3, the images (a), (b), (c), and (d) are processed images at respective stages.
105, performing secondary training on the improved convolutional neural network model after the primary training by using the updated standard sample library to obtain an improved convolutional neural network model after the secondary training;
and 106, acquiring an aggregate image to be detected.
The method specifically comprises the following steps: acquiring an orthoscopic image of the aggregate to be detected and four inclined images of the aggregate to be detected by adopting a five-lens inclined camera set; and selecting an image with the best quality from one aggregate orthoscopic image to be detected and four aggregate oblique images to be detected as an aggregate image to be detected.
And 107, inputting the aggregate image to be detected into the secondarily trained improved convolutional neural network model, and determining the classification result of each aggregate in the aggregate image to be detected.
The stone to be recognized can be selected into an optimal image through equipment shooting, the stone to be recognized can be rapidly classified through an improved convolutional neural network model after secondary training, the particle size range is obtained according to classification information, and the category of the aggregate to be recognized is obtained. And setting updating time, updating and recording the identified data set and the corresponding identification result into a standard sample library in an updating period, and updating the identification network model.
And 108, establishing a grading curve of the aggregate to be detected according to the classification result.
The method specifically comprises the following steps: determining the number of aggregates of different categories according to the classification result; the categories comprise a large category and a small category; determining the mass ratio of the aggregates of each category according to the number of the aggregates of different categories; and drawing an aggregate grading curve according to the mass ratio of the aggregates of each category.
Wherein the major and minor categories of the present invention are relative. The classification range of the large category is larger, and the classification range of the small category is finer. For example, the classification range of the large class is 1: 0-40mm, 2: at 40-120mm … …, the small categories may be classified in the range of 1.1: 0-10mm, 1.2: 10-20mm, 1.3: 20-30mm, 1.4: 30-40 mm; 2.1: 40-50mm, 2.2: 50-60mm, 2.3: 60-70mm, 2.4: 70-80mm, 2.5: 80-90mm, 2.6: 90-100mm, 2.7: 100-; the large category and the small category of the invention can mutually verify, and the accuracy of identification is further improved.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
1. a large number of sample stones are taken to be shot in multiple angles to construct an early-stage training set, the volume of the samples is obtained through measurement by a drainage method and is stored in a standard sample library, and the standard sample library contains information such as the volume, the quality and images of the samples. And performing classification training by machine learning through data integration of the sample library to obtain the CNN model. With the supplement and update of the data of the sample library, the CNN model is continuously optimized, and the accuracy is continuously improved.
2. The method also comprises the steps of obtaining the concrete aggregate orthoimage to be identified on the conveyor belt, obtaining respective attributes through the steps of preprocessing, image segmentation, identification algorithm and the like, obtaining a more refined classification result according to the attribute information, supplementing the classification result back to the standard sample base again, and realizing the expansion and optimization of the sample base again. And simultaneously, images can be selected from multi-angle oblique images by using a conditional random field, and classification of coarse aggregates is realized through a CNN model. Because the accurate volume of the stone cannot be obtained from the attribute information obtained by image recognition, the accurate volume of the stone can be obtained by combining machine learning and obtaining a classification result based on a CNN model, and accordingly, the accurate volume of the stone can be obtained.
3. Obtaining the aggregate volume according to the classification information so as to obtain the mass ratio of each classification of the whole aggregate to be detected (the mass m of each classification)1(m2，m3…)/M total percentage), and finally drawing a grading curve to realize the quick identification of concrete aggregate grading of non-contact measurement.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other.
The principle and the implementation manner of the present invention are explained by applying specific examples, the above description of the embodiments is only used to help understanding the method of the present invention and the core idea thereof, the described embodiments are only a part of the embodiments of the present invention, not all embodiments, and all other embodiments obtained by a person of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the protection scope of the present invention.
Claims (5)
1. A non-contact measured concrete aggregate gradation rapid identification method is characterized by comprising the following steps:
collecting multi-angle photos of each stone block of each large category as a first training sample, and generating a standard sample library;
building a convolutional neural network to obtain an initial improved convolutional neural network model;
carrying out primary training on the initial improved convolutional neural network model by using the standard sample library to obtain a trained improved convolutional neural network model;
collecting an aggregate original image, determining the small category of each aggregate in the aggregate original image by adopting an image processing mode to obtain a second training sample, and updating the standard sample library by utilizing the second training sample;
performing secondary training on the improved convolutional neural network model after the primary training by using the updated standard sample library to obtain an improved convolutional neural network model after the secondary training;
collecting an aggregate image to be detected;
inputting the aggregate image to be detected into the improved convolutional neural network model after the secondary training, and determining the classification result of each aggregate in the aggregate image to be detected;
establishing a grading curve of the aggregate to be detected according to the classification result;
the method comprises the steps of collecting an aggregate original image, determining the small category of each aggregate in the aggregate original image in an image processing mode, and obtaining a second training sample, wherein the method specifically comprises the following steps: acquiring an aggregate orthographic image and four aggregate oblique images by adopting a five-lens oblique camera set; selecting an aggregate inclined image with the best quality from the four aggregate inclined images as an input image of a second training sample; determining the subclass of each aggregate in the aggregate ortho-image in an image processing mode to serve as a target output result of a second training sample;
the determining, by using an image processing method, the subclass to which each aggregate belongs in the aggregate ortho-image is used as a target output result of the second training sample, and specifically includes: preprocessing the aggregate orthoimage to obtain a preprocessed orthoimage; carrying out canny operator edge detection on the preprocessed ortho-image, determining the edge of each aggregate image in the preprocessed ortho-image, and obtaining the edge-detected ortho-image; performing expansion and corrosion treatment on the edge-detected ortho-image, and determining the outline of each aggregate in the edge-detected ortho-image; determining the contour line of each aggregate in the aggregate orthographic image according to the contour line of each aggregate in the orthographic image after the edge detection; performing contour fitting on the contour line of each aggregate in the aggregate ortho-image to obtain a fitted minimum circumscribed ellipse of each aggregate in the aggregate ortho-image; and determining the small category of each aggregate in the original image according to the fitted minimum circumscribed ellipse of each aggregate in the aggregate orthoimage.
2. The method for rapidly identifying concrete aggregate gradation according to the non-contact measurement of claim 1, wherein the improved convolutional neural network model comprises: 1 input layer, 10 convolutional layers, 9 pooling layers, and 1 output layer; the activation function of the improved convolutional neural network model is a Relu activation function.
3. The method for non-contact measured concrete aggregate grading quick identification according to claim 1, wherein the step of collecting multi-angle photos of each stone block of each large category as a first training sample comprises:
selecting a plurality of sample stones from the aggregate;
determining the volume of each sample stone block by adopting a drainage method, and determining the large category of each sample stone block according to the volume of the sample stone blocks;
a multi-angle photograph of each stone block of each large category is obtained as a first training sample.
4. The method for rapidly identifying concrete aggregate gradation through non-contact measurement according to claim 1, wherein the step of obtaining the aggregate image to be measured specifically comprises the steps of:
acquiring an orthoscopic image of the aggregate to be detected and four inclined images of the aggregate to be detected by adopting a five-lens inclined camera set;
and selecting an image with the best quality from one aggregate orthoscopic image to be detected and four aggregate oblique images to be detected as an aggregate image to be detected.
5. The method for rapidly identifying concrete aggregate gradation through non-contact measurement according to claim 1, wherein the establishing of the gradation curve of the aggregate to be measured according to the classification result specifically comprises:
determining the number of aggregates of different categories according to the classification result; the categories comprise a large category and a small category;
determining the mass ratio of the aggregates of each category according to the number of the aggregates of different categories;
and drawing an aggregate grading curve according to the mass ratio of the aggregates of each category.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910752374.1A CN110458119B (en) | 2019-08-15 | 2019-08-15 | Non-contact measurement concrete aggregate gradation rapid identification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910752374.1A CN110458119B (en) | 2019-08-15 | 2019-08-15 | Non-contact measurement concrete aggregate gradation rapid identification method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110458119A CN110458119A (en) | 2019-11-15 |
| CN110458119B true CN110458119B (en) | 2020-08-18 |
Family
ID=68486674
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910752374.1A Active CN110458119B (en) | 2019-08-15 | 2019-08-15 | Non-contact measurement concrete aggregate gradation rapid identification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110458119B (en) |
Families Citing this family (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111652270B (en) * | 2020-04-23 | 2022-03-29 | 中南大学 | Coarse-grained soil filler gradation automatic identification method based on image matching and application system |
| CN111523616B (en) * | 2020-05-14 | 2021-01-29 | 中国铁道科学研究院集团有限公司铁道建筑研究所 | Coarse-grained soil filler grading identification method based on convolutional neural network and application system |
| CN111751253B (en) * | 2020-07-06 | 2022-10-14 | 重庆理工大学 | Forming method and quality detection method of concrete aggregate detection model |
| CN112017164A (en) * | 2020-08-18 | 2020-12-01 | 中国水利水电科学研究院 | Soil and stone material grading detection method based on depth threshold convolution model |
| CN112183308A (en) * | 2020-09-25 | 2021-01-05 | 中国水利水电科学研究院 | Cemented sand gravel material online identification and grading optimization method and system |
| CN112201101A (en) * | 2020-09-29 | 2021-01-08 | 北京科东电力控制系统有限责任公司 | Education training system and training method based on augmented reality technology |
| CN112924654B (en) * | 2021-01-15 | 2023-06-02 | 重庆交通大学 | Non-contact intelligent real-time detection method and device for concrete plasticity-maintaining performance |
| CN113160125B (en) * | 2021-02-25 | 2023-11-03 | 同济大学 | Asphalt pavement mixture grading prediction method based on three-dimensional texture features |
| CN112949463B (en) * | 2021-02-26 | 2023-08-04 | 长安大学 | Method and system for establishing and detecting aggregate grading rapid detection model |
| CN113591830A (en) * | 2021-07-12 | 2021-11-02 | 汉谷云智(武汉)科技有限公司 | Aggregate specification intelligent identification method and equipment based on queue competition algorithm |
| CN113487643B (en) * | 2021-07-19 | 2022-06-28 | 华电西藏能源有限公司大古水电分公司 | Cemented sand gravel stock yard sampling and determining method |
| CN113486592B (en) * | 2021-07-19 | 2023-09-01 | 中国水利水电科学研究院 | Cementing material structural performance evolution prediction method considering multi-factor influence |
| CN113740216B (en) * | 2021-09-09 | 2022-05-24 | 北京师范大学 | Air-ground integrated detection method for wide-gradation mixed aggregate |
| CN113997410A (en) * | 2021-10-21 | 2022-02-01 | 怀化明达建材有限公司 | Automatic change material platform |
| CN114118266B (en) * | 2021-11-24 | 2024-08-20 | 华侨大学 | Visual detection classification method and system for recycled aggregate with mortar on surface |
| CN116674091B (en) * | 2023-08-03 | 2023-12-19 | 福建南方路面机械股份有限公司 | Automatic aggregate feeding method for post-material yard of concrete mixing station |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101609515A (en) * | 2009-06-25 | 2009-12-23 | 福州大学 | Concrete mixing proportion design method based on artificial intelligence |
| CN103514370A (en) * | 2013-09-18 | 2014-01-15 | 天津大学 | Optimization construction algorithm for aggregate grading of resin concrete |
| CN104966281A (en) * | 2015-04-14 | 2015-10-07 | 中测新图(北京)遥感技术有限责任公司 | IMU/GNSS guiding matching method of multi-view images |
| CN105510195A (en) * | 2015-12-07 | 2016-04-20 | 华侨大学 | On-line detection method for particle size and shape of stacked aggregate |
| CN107633229A (en) * | 2017-09-21 | 2018-01-26 | 北京智芯原动科技有限公司 | Method for detecting human face and device based on convolutional neural networks |
| CN108460341A (en) * | 2018-02-05 | 2018-08-28 | 西安电子科技大学 | Remote sensing image object detection method based on integrated depth convolutional network |
| CN109740657A (en) * | 2018-12-27 | 2019-05-10 | 郑州云海信息技术有限公司 | A kind of training method and equipment of the neural network model for image data classification |
| CN109993766A (en) * | 2019-03-18 | 2019-07-09 | 南京理工大学 | Sandstone image droplet measurement method based on deep learning |
| CN110047072A (en) * | 2019-04-30 | 2019-07-23 | 福建南方路面机械有限公司 | A kind of gravel size identification processing system and processing method based on mobile interchange |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105538510B (en) * | 2016-01-19 | 2019-03-22 | 中国水利水电科学研究院 | A kind of control system and method for concrete raw material match ratio |
| CN108090434B (en) * | 2017-12-13 | 2020-06-23 | 赣州好朋友科技有限公司 | Rapid ore identification method |
-
2019
- 2019-08-15 CN CN201910752374.1A patent/CN110458119B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101609515A (en) * | 2009-06-25 | 2009-12-23 | 福州大学 | Concrete mixing proportion design method based on artificial intelligence |
| CN103514370A (en) * | 2013-09-18 | 2014-01-15 | 天津大学 | Optimization construction algorithm for aggregate grading of resin concrete |
| CN104966281A (en) * | 2015-04-14 | 2015-10-07 | 中测新图(北京)遥感技术有限责任公司 | IMU/GNSS guiding matching method of multi-view images |
| CN105510195A (en) * | 2015-12-07 | 2016-04-20 | 华侨大学 | On-line detection method for particle size and shape of stacked aggregate |
| CN107633229A (en) * | 2017-09-21 | 2018-01-26 | 北京智芯原动科技有限公司 | Method for detecting human face and device based on convolutional neural networks |
| CN108460341A (en) * | 2018-02-05 | 2018-08-28 | 西安电子科技大学 | Remote sensing image object detection method based on integrated depth convolutional network |
| CN109740657A (en) * | 2018-12-27 | 2019-05-10 | 郑州云海信息技术有限公司 | A kind of training method and equipment of the neural network model for image data classification |
| CN109993766A (en) * | 2019-03-18 | 2019-07-09 | 南京理工大学 | Sandstone image droplet measurement method based on deep learning |
| CN110047072A (en) * | 2019-04-30 | 2019-07-23 | 福建南方路面机械有限公司 | A kind of gravel size identification processing system and processing method based on mobile interchange |
Non-Patent Citations (2)
| Title |
|---|
| 基于卷积神经网络的混凝土裂缝识别;温作林等;《低温建筑技术》;20190628;9-12、21 * |
| 基于深度学习的再生混凝土抗压强度预测;高蔚;《混凝土》;20181127(第11期);58-61、70 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110458119A (en) | 2019-11-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110458119B (en) | Non-contact measurement concrete aggregate gradation rapid identification method | |
| CN110118775B (en) | Method for rapidly detecting composition of plant-mixed cement stabilized macadam aggregate | |
| Guo et al. | A method of blasted rock image segmentation based on improved watershed algorithm | |
| Zheng et al. | Roundness and sphericity of soil particles in assemblies by computational geometry | |
| Sun et al. | Improved watershed analysis for segmenting contacting particles of coarse granular soils in volumetric images | |
| CN112964712A (en) | Method for rapidly detecting state of asphalt pavement | |
| CN109584286B (en) | Asphalt pavement structure depth calculation method based on generalized regression neural network | |
| CN102494976A (en) | Method for automatic measurement and morphological classification statistic of ultra-fine grain steel grains | |
| CN109785307B (en) | Unmanned aerial vehicle image road damage assessment method based on vector guidance | |
| CN107314957B (en) | Method for measuring rock block size distribution | |
| Dan et al. | Application of deep learning-based image recognition technology to asphalt–aggregate mixtures: Methodology | |
| CN104657978A (en) | Road extracting method based on shape characteristics of roads of remote sensing images | |
| CN115564771A (en) | Concrete crack identification method based on building foundation column | |
| CN105654482B (en) | Based on the determination method of the soil-rock mixture characterization unit body scale of Digital image technology | |
| Kim et al. | 3D image segmentation of aggregates from laser profiling | |
| Papanicolaou et al. | The use of fractals to quantify the morphology of cluster microforms | |
| Wang et al. | Fragmentation calculation method for blast muck piles in open-pit copper mines based on three-dimensional laser point cloud data | |
| CN109146888B (en) | Soil crack fractal dimension calculation method based on difference quotient analysis | |
| CN115147401A (en) | Digital image processing-based intelligent detection method for dam material qualification of earth and rockfill dam | |
| Azarafza et al. | Application of an image processing-based algorithm for river-side granular sediment gradation distribution analysis | |
| Wang | Rock particle image segmentation and systems | |
| CN107024416A (en) | With reference to similitude and the quasi-circular particle mean size detection method of discontinuity | |
| Xie et al. | Bridge defect detection technology based on machine vision and embedded | |
| Shao et al. | Crack detection and measurement using PTZ camera–based image processing method on expressways | |
| CN108335321B (en) | Automatic ground surface gravel size information extraction method based on multi-angle photos |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB03 | Change of inventor or designer information |
Inventor after: Lei Tianjie Inventor after: Jia Jinsheng Inventor after: Zheng Cuiying Inventor after: Wang Jiabao Inventor after: Yang Huichen Inventor before: Lei Tianjie Inventor before: Shen Li Inventor before: Gong Adu Inventor before: Lv Juan Inventor before: Song Wenlong Inventor before: Zhang Ju Inventor before: Li Shican Inventor before: Zhou Lei Inventor before: Chen Qiang Inventor before: Li Aili Inventor before: Wang Yang Inventor before: Jia Jinsheng Inventor before: Lou Hezhen Inventor before: Cheng Ziyi Inventor before: Feng Wei Inventor before: Shi Wanli Inventor before: Zhao Chun Inventor before: Wan Jinhong Inventor before: Liu Zhongwei Inventor before: Lu Jingxuan Inventor before: Qu Wei Inventor before: Li Xiangyu Inventor before: Wang Jiabao Inventor before: Chen Wenjin Inventor before: Li Mingyu Inventor before: Li Yang Inventor before: Cheng Hui Inventor before: Huang Jintao Inventor before: Zhao Linhong Inventor before: Xu Ruirui Inventor before: Zhang Pengpeng Inventor before: Zheng Cuiying Inventor before: Li Shuguang Inventor before: Yang Huichen Inventor before: Zhang Yazhen Inventor before: Song Hongquan Inventor before: Yue Jianwei |
|
| CB03 | Change of inventor or designer information | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |