CN112017164A - Soil and stone material grading detection method based on depth threshold convolution model - Google Patents
Soil and stone material grading detection method based on depth threshold convolution model Download PDFInfo
- Publication number
- CN112017164A CN112017164A CN202010833597.3A CN202010833597A CN112017164A CN 112017164 A CN112017164 A CN 112017164A CN 202010833597 A CN202010833597 A CN 202010833597A CN 112017164 A CN112017164 A CN 112017164A
- Authority
- CN
- China
- Prior art keywords
- soil
- stone
- grading
- image
- depth threshold
- 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.)
- Pending
Links
- 239000002689 soil Substances 0.000 title claims abstract description 143
- 239000004575 stone Substances 0.000 title claims abstract description 118
- 239000000463 material Substances 0.000 title claims abstract description 66
- 238000001514 detection method Methods 0.000 title claims abstract description 30
- 238000000034 method Methods 0.000 claims abstract description 40
- 239000011435 rock Substances 0.000 claims abstract description 29
- 239000002245 particle Substances 0.000 claims description 79
- 238000012549 training Methods 0.000 claims description 28
- 238000013527 convolutional neural network Methods 0.000 claims description 21
- 238000003708 edge detection Methods 0.000 claims description 13
- 238000013528 artificial neural network Methods 0.000 claims description 12
- 238000012545 processing Methods 0.000 claims description 9
- 238000012216 screening Methods 0.000 claims description 8
- 238000006243 chemical reaction Methods 0.000 claims description 7
- 230000007797 corrosion Effects 0.000 claims description 7
- 238000005260 corrosion Methods 0.000 claims description 7
- 230000000877 morphologic effect Effects 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 6
- 238000004458 analytical method Methods 0.000 claims description 3
- 230000002146 bilateral effect Effects 0.000 claims description 3
- 238000001914 filtration Methods 0.000 claims description 3
- 230000009467 reduction Effects 0.000 claims description 3
- 238000007619 statistical method Methods 0.000 abstract description 3
- 230000008569 process Effects 0.000 description 14
- 230000006870 function Effects 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 6
- 238000011176 pooling Methods 0.000 description 6
- 238000012937 correction Methods 0.000 description 5
- 239000011159 matrix material Substances 0.000 description 5
- 230000011218 segmentation Effects 0.000 description 5
- 238000012360 testing method Methods 0.000 description 5
- 238000012795 verification Methods 0.000 description 5
- 238000013135 deep learning Methods 0.000 description 4
- 230000004913 activation Effects 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 230000018109 developmental process Effects 0.000 description 3
- 239000000284 extract Substances 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000005286 illumination Methods 0.000 description 2
- 230000001788 irregular Effects 0.000 description 2
- 238000013256 Gubra-Amylin NASH model Methods 0.000 description 1
- 235000019738 Limestone Nutrition 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 239000010426 asphalt Substances 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000003628 erosive effect Effects 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 229910052500 inorganic mineral Inorganic materials 0.000 description 1
- 239000006028 limestone Substances 0.000 description 1
- 239000011707 mineral Substances 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000010422 painting Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 238000003908 quality control method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 238000005096 rolling process Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000013526 transfer learning Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/12—Edge-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20024—Filtering details
- G06T2207/20028—Bilateral filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30132—Masonry; Concrete
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a soil and stone grading detection method based on a depth threshold convolution model. The method comprises the steps of rapidly segmenting a soil and stone image and extracting a contour by a thresholding method, and performing statistical analysis on extracted contour characteristics to obtain initial grading data; and then, taking the obtained initial grading data as a sample, constructing a depth threshold convolution model to correct the error of the thresholding image identification result, and obtaining accurate grading data, thereby realizing the rapid detection of soil and rock material grading based on the image and obviously improving the stability and accuracy of image identification.
Description
Technical Field
The invention relates to the technical field of soil and rock material grading detection, in particular to a soil and rock material grading detection method based on a depth threshold convolution model.
Background
The grading of earth and rock materials is a basic parameter for researching the performance of the earth and rock dam, and the quality of the grading directly influences the impermeability and the stability of the dam body. In the dam construction quality control process, the rapid detection of soil and stone material grading on site has important significance for ensuring the engineering safety. And traditional soil and stone material gradation detection mainly adopts the screening method, acquires through artifical sampling combination screening machine test, and the process is complicated, and the process is consuming time, is difficult to satisfy real-time, quick demand in the work progress. With the development of computer science, the digital image technology provides a new technology and a new method for rapidly detecting the grading of the soil and rock materials.
The digital image technology-based soil and stone material grading detection key lies in the extraction of soil and stone material particle shape characteristics and the conversion between the shape characteristics and the quality. And a painting bin and the like extract particle morphological parameters through two-dimensional images to describe the particle shapes of the soil and stone materials with irregular shapes. The average particle size and uniformity index of the particles are introduced into the coarse aggregate particles in ten thousand, and the uniformity of the particle size of the coarse aggregate particles is quantitatively described by means of a thresholding method. The detection method has the advantages that the detection of the gradation of the asphalt mixture mineral aggregate based on the image is realized by adopting differential correction on the identification error on the basis of thresholding by Shaaimin and the like. Based on thresholding and edge detection algorithms, a soil and stone grading rapid detection system is established, and industrial grading rapid detection is realized. The foregoing scholars all adopt feature extraction methods based on the conventional image recognition technology, the recognition speed is fast but the accuracy is insufficient, and with the development of artificial intelligence, the research of two-dimensional shape feature quantitative extraction methods of target objects in images has made new progress. Jonathan et al propose a full Convolutional neural network (FCN), modify the original neural network full connection layer into a Convolutional layer, promote the neural network output judgment from one-dimensional data classification to two-dimensional data classification by means of deconvolution, realize the identification of each pixel point in an image, and provide directions for deep learning based on the image. The method comprises the steps of designing a Residual error neural Network (ResNet) through Residual error learning, solving the problem of gradient disappearance or gradient explosion of the neural Network during reverse propagation, deepening the depth of the neural Network, improving the accuracy of image recognition, providing a Mask-RCNN image recognition model on the basis, and promoting the object contour segmentation in the image to a pixel level by setting an output layer Mask prediction branch, thereby providing a transfer learning basis for the recognition and segmentation of different types of objects. The development of the traditional image identification technology and the image identification technology based on deep learning provides theoretical basis and technical support for image-based soil and stone grading detection.
The content of particles with the particle size of less than 5mm in the soil material has important significance on the rationality of gradation. The relevant national specifications make different regulations of 20-50% on the content of particles below 5mm at different parts of the earth-rock dam body. The judgment of the content of the particles with the particle size of less than 5mm is the key in the classification identification of the soil and rock material, and the particles with the particle size of less than 5mm reflect the characteristics of fineness, adhesion, irregular shape and serious stacking phenomenon in the image of the soil and rock material. For the image with the adhered particles, the traditional image identification mainly adopts a mode based on thresholding, edge detection and watershed algorithm, and the deep learning image identification mainly adopts a semantic segmentation method based on FCN, VGG, ResNet and GAN models. After experiments with different algorithms it was found that: aiming at the soil and stone image, the traditional identification method has poor precision, and is difficult to identify particles with the particle size of below 5 mm. The deep learning image identification method extracts features through sample marks, the identification precision mainly depends on the number of samples and the quality of the sample marks, a large number of soil and stone particles exist in the soil and stone image, the sample marking process is complex, the requirement of model operation on computer hardware is high, and the method is greatly limited.
Disclosure of Invention
Aiming at the defects in the prior art, the invention provides a soil and rock material grading detection method based on a depth threshold convolution model.
In order to achieve the purpose of the invention, the invention adopts the technical scheme that:
a soil and rock material grading detection method based on a depth threshold convolution model comprises the following steps:
s1, acquiring soil and stone image data;
s2, preprocessing the soil and stone image;
s3, extracting initial grading data from the preprocessed soil and stone material image in a thresholding mode;
and S4, constructing a depth threshold convolution model and training, and correcting the initial grading data by using the trained depth threshold convolution model to obtain soil and stone grading data.
Preferably, the step S2 specifically includes:
setting the size of the soil and stone image, and carrying out bilateral filtering and noise reduction treatment on the soil and stone image.
Preferably, the step S3 specifically includes:
and respectively carrying out threshold conversion and edge detection on the preprocessed soil and stone material images, extracting soil and stone material particle profile information, and fitting the soil and stone material particle profile to obtain initial grading data.
Preferably, the thresholding conversion of the preprocessed soil and stone image in the step S3 specifically includes:
extracting gray values of all pixel points on the preprocessed soil and stone material image;
calculating the mean value of the gray values of the pixel points;
and determining an optimal threshold value by adopting a maximum inter-class variance method, and dividing the particle region.
Preferably, the edge detection of the preprocessed soil and stone image in the step S3 specifically includes:
respectively determining corrosion and expansion structure units, performing corrosion and expansion morphological processing on the thresholded soil and stone image, and determining the soil and stone edge;
and carrying out Canny edge detection on the soil stone image.
Preferably, the step S3 of fitting the soil and stone particle profile to obtain the initial grading data specifically includes:
taking the length of the short axis of the contour shape of the soil and stone material particles as the minimum diameter of the screening aperture, counting the number of particles and contour information in each particle size range, and carrying out ellipse fitting on each particle contour to obtain the percentage of the particles passing through the screen aperture in each particle size group range, thus obtaining the initial grading data.
Preferably, the constructing of the depth threshold convolution model in step S4 specifically includes:
and taking initial grading data obtained by thresholding the soil and stone material image in the steps S1 to S3 as an input sample, taking the real grading data as a training target, extracting local features of sample data by adopting a convolutional neural network, and correcting the error of a thresholding image identification result to obtain final soil and stone material grading data.
Preferably, the training of the depth threshold convolution model in step S4 specifically includes:
under the same grading condition, different soil and stone images formed after multiple times of turning processing are used as identification samples, initial grading data of multiple times of identification are used as training samples of a convolutional neural network, real grading data are used as training targets of the convolutional neural network, and neural network training is carried out on thresholding image identification results.
Preferably, in the step S4, the mean square error loss function is used to perform error analysis on the output result of the depth threshold convolution model, and the corrected soil and rock material grading data is obtained through error back propagation adjustment.
Preferably, the mean square error loss function is specifically represented as:
wherein E represents the mean square error loss, E represents the prediction result output by the fully-connected layer of the convolutional neural network, y represents the real grading data, and n represents the total number of all output results.
The invention has the following beneficial effects:
the method comprises the steps of rapidly segmenting a soil and stone image and extracting a contour by a thresholding method, and performing statistical analysis on extracted contour characteristics to obtain initial grading data; and then, taking the obtained initial grading data as a sample, constructing a depth threshold convolution model to correct the error of the thresholding image identification result, and obtaining accurate grading data, thereby realizing the rapid detection of soil and rock material grading based on the image and obviously improving the stability and accuracy of image identification.
Drawings
FIG. 1 is a schematic flow chart of a soil and rock material grading detection method based on a depth threshold convolution model according to the present invention;
FIG. 2 is a schematic view illustrating image recognition of soil and stone materials according to an embodiment of the present invention; wherein, the graph (a) is an original image, the graph (b) is a threshold value transformation image, the graph (c) is an erosion and expansion processing image, and the graph (d) is a Canny edge detection image;
fig. 3 is a schematic structural diagram of a convolutional neural network of a depth threshold convolutional model in an embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention is provided to facilitate the understanding of the present invention by those skilled in the art, but it should be understood that the present invention is not limited to the scope of the embodiments, and it will be apparent to those skilled in the art that various changes may be made without departing from the spirit and scope of the invention as defined and defined in the appended claims, and all matters produced by the invention using the inventive concept are protected.
As shown in fig. 1, an embodiment of the present invention provides a depth threshold convolution model-based soil and rock material grading detection method, including the following steps S1 to S5:
s1, acquiring soil and stone image data;
in the embodiment, the limestone and stone which are widely used in engineering, have dark colors and are difficult to recognize images are used as soil and stone typical samples, and the soil and stone samples are reduced in a certain proportion according to requirements of actual engineering environments and model establishment, wherein 60mm is used as the maximum control grain size, and 1mm is used as the minimum control grain size. In order to simplify the grading data processing process, the content of particles with the particle diameter of less than 1mm accounts for 0% of the total particles, and the content of particles with the particle diameter of less than 60mm accounts for 100% of the total particles.
Considering the engineering site environment, the invention adopts an industrial camera as an image acquisition tool of soil and stone materials, and the basic parameters are shown in table 1.
TABLE 1 Camera and Lighting System parameters
In order to reduce the influence caused by illumination conditions, the LED auxiliary lighting system is arranged, and the brightness and the definition of an image are ensured.
S2, preprocessing the soil and stone image;
in this embodiment, since the field environment of the soil and rock material grading detection project is relatively complex, in order to ensure the accuracy of the model identification effect, certain preprocessing needs to be performed on the soil and rock material image.
The method for preprocessing the soil and stone image specifically comprises the following steps:
setting the size of the soil and stone image, and carrying out bilateral filtering and noise reduction treatment on the soil and stone image.
Through the preprocessing, the preprocessed soil and stone images have uniform shape specifications, and the influence of noise information of the images caused by factors such as illumination, angles and the like is reduced.
S3, extracting initial grading data from the preprocessed soil and stone material image;
in this embodiment, the method performs thresholding on the preprocessed soil and stone material image by using a maximum inter-class variance method, and performs preliminary segmentation; determining the profile information of the soil and stone particles by a morphological processing method of corrosion and expansion; and finally, carrying out statistical analysis on the extracted contour features to obtain initial grading data.
The step S3 specifically includes:
and respectively carrying out threshold conversion and edge detection on the preprocessed soil and stone material images, extracting soil and stone material particle profile information, and fitting the soil and stone material particle profile to obtain initial grading data.
The thresholding conversion of the preprocessed soil and stone image specifically comprises the following steps:
extracting gray values of all pixel points on the preprocessed soil and stone material image;
calculating the mean value of the gray values of the pixel points;
and determining an optimal threshold value by adopting a maximum inter-class variance method, and dividing the particle region.
In the invention, the soil and stone image is abstracted into a two-dimensional function f (x, y), wherein x and y correspond to the positions of all points in the image, and the amplitude f of the image at each point is the pixel value of the corresponding point, and the value range is 0-255, also called as the gray value.
The maximum inter-class variance method adopted by the invention is a thresholding segmentation method taking the maximization of the variance between a target class and a background class as a standard, and the fast identification of soil and stone particles can be realized by solving the inter-class difference between the two classes for the soil and stone image, wherein the target class represents the soil and stone particles, and the background class represents the shadow part at the edge.
The soil and stone material image f (x, y) comprises L gray levels, wherein the number of pixel points with the gray level value i is N, the total number of pixel points is N, and then the probability p that the gray level of the pixel points is iiComprises the following steps:
assuming a certain threshold k, the soil and stone images are divided into two categories, wherein the region with the pixel gray value larger than k is called a foreground target region, and the region smaller than k is called a background edge region. The probability of two types of regions occurring at this time can be expressed as:
wherein, w1Representing the probability of occurrence of background edge regions, w2Representing the probability of the occurrence of the target soil and stone particle region.
The average gray levels in the background region and the target region are:
wherein u is1Representing the average gray level, u, of the edge area of the background2Representing target soil and stone particle areasAverage gray scale, uTAnd the integral average gray scale of the soil and stone image is represented.
The inter-class variance of the two classes of regions can be expressed as:
σ2＝w1(u1-uT)2+w2(u2-uT)
iterating the threshold k in a traversal fashion, with the between-class variance σ2And taking the k value corresponding to the maximum value as an optimal threshold, replacing the gray value of the background edge region by 0, and replacing the gray value of the target soil and stone particle region by 255, thereby completing the basic identification process of the soil and stone particles.
The edge detection of the preprocessed soil and stone image specifically comprises the following steps:
respectively determining corrosion and expansion structure units, performing corrosion and expansion morphological processing on the thresholded soil and stone image, and determining the edges of soil and stone particles;
and carrying out Canny edge detection on the soil stone image.
The step of fitting the soil and stone particle profile to obtain initial grading data specifically comprises the following steps:
taking the length of the short axis of the contour shape of the soil and stone material particles as the minimum diameter of the screening aperture, counting the number of particles and contour information in each particle size range, and carrying out ellipse fitting on each particle contour to obtain the percentage of the particles passing through the screen aperture in each particle size group range, thus obtaining the initial grading data.
As shown in fig. 2, the maximum between-class variance thresholding of the soil and stone images taken by the industrial camera can determine the base range of the soil and stone particles. At the moment, a large amount of adhesion phenomena exist among the soil and stone particles, so that the area where the soil and stone are located can be reduced by adopting a corrosive morphological treatment method, and the mutual adhesion phenomena among the particles are reduced. And then expanding the edge of the region in an expansion mode to restore the edge characteristics of the particles. And (4) carrying out Canny edge detection on the morphologically processed soil and stone image to obtain the contour information of the soil and stone particles.
The invention assumes that the shape of the soil and stone material particles is elliptical and the density is consistent, the length of the minor axis is the minimum diameter capable of passing through the screening aperture, the number and the outline information of the particles in each particle size range are counted, and the ellipse fitting is carried out on each particle outline, so that the percentage of the particles passing through the screen aperture in each particle size group range can be obtained, and the preliminary grading data is obtained.
And S4, constructing a depth threshold convolution model and training, and correcting the initial grading data by using the trained depth threshold convolution model to obtain soil and stone grading data.
In this embodiment, the constructing a depth threshold convolution model (Deep obtus Convolutional Network, DO-CNN) in the present invention specifically includes:
and taking initial grading data obtained by thresholding the soil and stone material image in the steps S1 to S3 as an input sample, taking real grading data as a training target, adopting a convolutional neural network to extract local characteristics of sample data, and correcting the error of a thresholding image identification result to obtain final soil and stone material grading data, thereby realizing the rapid detection of the soil and stone material grading based on the image.
Because the particle size in the soil and stone material image is small, the phenomena of rolling, covering and the like can be generated in the soil and stone material carrying process, and thus the grading data acquired by the thresholding image identification algorithm has certain errors. However, for the same group of graded soil and stone particles under the same mining condition, the grading obtained by the image after continuous overturning through the thresholding algorithm should be kept unchanged theoretically, so that the image recognition error distribution rule can be summarized through an artificial intelligent algorithm to realize error correction. The error distribution rule of the grading data has a certain relation with the particle size range, so that the method adopts a Convolutional Neural Network (CNN) capable of extracting the local characteristics of the data as an error correction model.
As shown in fig. 3, the convolutional neural network is a deep neural network having a convolutional structure, and its basic structure includes an input layer, a convolutional layer (convolutional layer), a pooling layer (posing layer), a fully connected layer (fully connected layer), and an output layer.
In the convolutional layer, a convolutional kernel slides on a feature graph output from the previous layer, products and summations are carried out on the feature graph output from the previous layer and elements in a corresponding region of the convolutional kernel, primary output can be obtained through bias correction and activation of an activation function, and local feature information of data can be obtained through adjustment of the size of the convolutional kernel, and the local feature information is expressed as:
wherein denotes a convolution;
representing the convolution result in a convolution kernel region; mjRepresenting a subset of the output characteristic diagram of the previous layer, namely a corresponding area when the convolution kernel slides;
represents MjThe ith element;
representing a convolution kernel;
represents a bias correction magnitude;
represents the jth output in the present layer convolution process; f (-) represents the activation function. Convolution kernel corresponding to each characteristic graph
May be different.
The pooling layer achieves the invariant nature of the data within the region by reducing the resolution of the feature plane. Maximum pooling (Maxpool), i.e.pooling in such a way that a maximum value in the pool kernel range is retained in the pooling, is used herein. The convolution layer and the pooling layer are repeatedly arranged to form a multi-layer middle hidden layer of the neural network, high-dimensional summarization of regional data features is achieved, and then convolution kernels with the same size as the feature graph output by the upper layer are utilized to perform convolution with all the output of the upper layer in the full-connection layer, so that summarization of the whole feature graph is achieved to serve as output.
The inevitable vibration of soil and stone particles in the transportation process causes that soil and stone images formed by the soil and stone under the same grading condition under different shooting conditions are different, and grading data obtained through image identification are different correspondingly. In order to solve the problem of soil and stone image identification errors caused by vibration, the invention trains grading data of image identification by adopting a convolutional neural network and corrects the identification grading result.
According to the method, under the same grading condition, different soil and stone images formed after 20 times of turning processing are used as identification samples, initial grading data identified for 20 times are used as training samples of a convolutional neural network, real grading data are used as training targets of the convolutional neural network, and neural network training is carried out on thresholding image identification results. The rounding processing is used for reflecting the states of soil and stone materials formed under different vibration conditions and is used for training to improve the accuracy of model identification.
Each soil and stone grading curve of the invention contains 10 kinds of control particle sizes of 1mm, 5mm, 10mm, 20mm, 30mm, 40mm, 45mm, 50mm, 55mm and 60mm, and 10 particle size ranges are correspondingly formed, so that a 10 x 20 sample data matrix is formed after 20 times of image recognition, and a feature matrix formed after convolution can be rapidly reduced due to smaller matrix specification, thereby causing edge data feature loss. Therefore, the depth threshold convolution model of the invention takes 0 as the filling (Padding) of each intermediate layer feature matrix in the convolution process, and prevents the feature matrix from being shrunk too fast. And (3) performing Error analysis on the output result of the convolutional neural network by adopting a Mean Square Error Loss function (MSELoss), and obtaining corrected grading data through Error back propagation adjustment.
The mean square error loss function is specifically expressed as:
wherein E represents the mean square error loss, E represents the prediction result output by the fully-connected layer of the convolutional neural network, y represents the real grading data, and n represents the total number of all output results.
In order to verify the accuracy and stability of the depth threshold convolution model, a manual screening test is adopted to obtain the real grade of the soil and rock material and the soil and rock material image. The invention totally carries out 18 groups of screening tests under different grading conditions, and each group of soil and rock material is turned over for 20 times to obtain 18 multiplied by 20 different soil and rock material image data. 16 groups of data are taken as model training samples, and the rest 2 groups are taken as model verification.
In the model training process, different parameter settings have different influences on the model recognition result. The main parameters include: epoch, representing the number of times sample data is passed in model operation; iteration, which represents the number of iterations performed in one pass; batch-size, which represents the number of samples put into each training process; file-number, representing the total number of groups participating in the training sample; learning-rate, which represents the neural network learning rate, is uniformly set to 0.0005. In order to ensure the diversity and the difference of sample data, samples are randomly selected from all training samples to be put into the neural network in each batch of training process, and 10 groups of comparison working conditions are set in consideration of the influence of different parameters, as shown in table 2. The test parameters of the working condition 4 and the working condition 5 are consistent, the stability of the model identification result under the same parameters is mainly verified, the number of the working condition 8 sample groups is reduced, and the influence of different numbers of samples on the model identification is mainly verified.
Table 210 test parameter settings under operating conditions
Comparing the model detection results under 10 working conditions, and finally selecting the model parameter under the working condition 4 with the best recognition result as the model parameter during the final recognition.
The depth threshold convolution model identifies each soil and stone image in the operation process to generate corresponding grading data, namely the grading data obtained only through a thresholding image identification algorithm, and by taking a verification group 1 of two verification groups in 18 groups of experiments as an example, 20 different soil and stone images are formed by being turned under the same grading condition, wherein the grading data obtained by identifying 10 images through thresholding images is shown in table 3.
TABLE 3 thresholded image identification results
Through the identification training of 320 images in 16 groups, the model extracts the characteristics of soil and stone grading distribution rules and image identification result errors in different vibration states under the same grading condition. The 2 validation set samples not participating in model training were tested with weights saved in model training, and the recognition results and Mean Absolute Percentage Error (MAPE) under the same training parameters are shown in table 4.
TABLE 4 DO-CNN model identification results
According to the identification results of the verification group 1 and the verification group 2, compared with a single image identification algorithm, the depth threshold convolution model can greatly improve the accuracy and stability of soil and stone material image grading identification, and the model image identification part is based on a thresholding algorithm, so that the image identification speed can be obviously improved, and the grading rapid detection based on the soil and stone material image is realized.
It will be appreciated by those of ordinary skill in the art that the embodiments described herein are intended to assist the reader in understanding the principles of the invention and are to be construed as being without limitation to such specifically recited embodiments and examples. Those skilled in the art can make various other specific changes and combinations based on the teachings of the present invention without departing from the spirit of the invention, and these changes and combinations are within the scope of the invention.
Claims (10)
1. A soil and rock material grading detection method based on a depth threshold convolution model is characterized by comprising the following steps:
s1, acquiring soil and stone image data;
s2, preprocessing the soil and stone image;
s3, extracting initial grading data from the preprocessed soil and stone material image in a thresholding mode;
and S4, constructing a depth threshold convolution model and training, and correcting the initial grading data by using the trained depth threshold convolution model to obtain soil and stone grading data.
2. The depth threshold convolution model-based soil and rock material grading detection method according to claim 1, wherein the step S2 specifically includes:
setting the size of the soil and stone image, and carrying out bilateral filtering and noise reduction treatment on the soil and stone image.
3. The depth threshold convolution model-based soil and rock material grading detection method according to claim 1, wherein the step S3 specifically includes:
and respectively carrying out threshold conversion and edge detection on the preprocessed soil and stone material images, extracting soil and stone material particle profile information, and fitting the soil and stone material particle profile to obtain initial grading data.
4. The depth threshold convolution model-based soil and stone grading detection method as claimed in claim 3, wherein the thresholding conversion of the preprocessed soil and stone image in the step S3 specifically includes:
extracting gray values of all pixel points on the preprocessed soil and stone material image;
calculating the mean value of the gray values of the pixel points;
and determining an optimal threshold value by adopting a maximum inter-class variance method, and dividing the particle region.
5. The depth threshold convolution model-based soil and stone grading detection method according to claim 3, wherein the step S3 of performing edge detection on the preprocessed soil and stone image specifically includes:
respectively determining corrosion and expansion structure units, performing corrosion and expansion morphological processing on the thresholded soil and stone image, and determining the soil and stone edge;
and carrying out Canny edge detection on the soil stone image.
6. The depth threshold convolution model-based soil and rock grading detection method as claimed in claim 3, wherein the step S3 of fitting the soil and rock grain profile to obtain the initial grading data specifically includes:
taking the length of the short axis of the contour shape of the soil and stone material particles as the minimum diameter of the screening aperture, counting the number of particles and contour information in each particle size range, and carrying out ellipse fitting on each particle contour to obtain the percentage of the particles passing through the screen aperture in each particle size group range, thus obtaining the initial grading data.
7. The method for detecting grading of earth and rock materials based on the depth threshold convolution model as claimed in claim 1, wherein the constructing the depth threshold convolution model in the step S4 specifically includes:
and taking initial grading data obtained by thresholding the soil and stone material image in the steps S1 to S3 as an input sample, taking the real grading data as a training target, extracting local features of sample data by adopting a convolutional neural network, and correcting the error of a thresholding image identification result to obtain final soil and stone material grading data.
8. The method as claimed in claim 7, wherein the training of the depth threshold convolution model in step S4 specifically includes:
under the same grading condition, different soil and stone images formed after multiple times of turning processing are used as identification samples, initial grading data of multiple times of identification are used as training samples of a convolutional neural network, real grading data are used as training targets of the convolutional neural network, and neural network training is carried out on thresholding image identification results.
9. The method as claimed in claim 8, wherein the step S4 is implemented by performing error analysis on the output result of the depth threshold convolution model using a mean square error loss function, and obtaining the modified soil and rock grading data through error back propagation adjustment.
10. The depth threshold convolution model-based soil and rock grading detection method as claimed in claim 9, wherein the mean square error loss function is specifically expressed as:
wherein E represents the mean square error loss, E represents the prediction result output by the fully-connected layer of the convolutional neural network, y represents the real grading data, and n represents the total number of all output results.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010833597.3A CN112017164A (en) | 2020-08-18 | 2020-08-18 | Soil and stone material grading detection method based on depth threshold convolution model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010833597.3A CN112017164A (en) | 2020-08-18 | 2020-08-18 | Soil and stone material grading detection method based on depth threshold convolution model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112017164A true CN112017164A (en) | 2020-12-01 |
Family
ID=73504939
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010833597.3A Pending CN112017164A (en) | 2020-08-18 | 2020-08-18 | Soil and stone material grading detection method based on depth threshold convolution model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112017164A (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112884739A (en) * | 2021-02-19 | 2021-06-01 | 武汉大学 | Rapid detection method for rockfill filling compactness based on deep learning network |
| CN113487643A (en) * | 2021-07-19 | 2021-10-08 | 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 |
| CN113554222A (en) * | 2021-07-19 | 2021-10-26 | 中国水利水电科学研究院 | Dynamic optimization and intelligent regulation and control configuration method for bonding dam generalized bonding material |
| CN113591981A (en) * | 2021-07-30 | 2021-11-02 | 上海建工四建集团有限公司 | Artificial intelligence-based existing terrazzo information survey method and system |
| CN113610771A (en) * | 2021-07-15 | 2021-11-05 | 三峡大学 | Rockfill material image detection grading curve correction method based on B-spline curve and particle swarm optimization |
| CN113658117A (en) * | 2021-08-02 | 2021-11-16 | 浙江大学 | Method for identifying and dividing aggregate boundaries in asphalt mixture based on deep learning |
| CN113674201A (en) * | 2021-07-09 | 2021-11-19 | 福建南方路面机械股份有限公司 | Aggregate particle size sampling monitoring method |
| CN113740216A (en) * | 2021-09-09 | 2021-12-03 | 北京师范大学 | Air-ground integrated detection method for wide-gradation mixed aggregate |
| CN113813839A (en) * | 2021-02-24 | 2021-12-21 | 晋江华宝石业有限公司 | Quick grading method and grading device for square stock |
| CN114612450A (en) * | 2022-03-17 | 2022-06-10 | 无锡东如科技有限公司 | Image detection segmentation method and system based on data augmentation machine vision and electronic equipment |
| JP7496858B2 (en) | 2022-08-08 | 2024-06-07 | 基礎地盤コンサルタンツ株式会社 | Particle size distribution prediction device, machine learning device, particle size distribution prediction method, and machine learning method |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106228536A (en) * | 2016-07-12 | 2016-12-14 | 四川大学 | The earth and rockfill dam dam material grain composition method of inspection based on Digital Image Processing |
| CN109060813A (en) * | 2018-06-05 | 2018-12-21 | 中国矿业大学(北京) | Earth and stone material gradation automatic testing method based on image optimization classification |
| CN110458119A (en) * | 2019-08-15 | 2019-11-15 | 中国水利水电科学研究院 | A kind of aggregate gradation method for quickly identifying of non-contact measurement |
-
2020
- 2020-08-18 CN CN202010833597.3A patent/CN112017164A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106228536A (en) * | 2016-07-12 | 2016-12-14 | 四川大学 | The earth and rockfill dam dam material grain composition method of inspection based on Digital Image Processing |
| CN109060813A (en) * | 2018-06-05 | 2018-12-21 | 中国矿业大学(北京) | Earth and stone material gradation automatic testing method based on image optimization classification |
| CN110458119A (en) * | 2019-08-15 | 2019-11-15 | 中国水利水电科学研究院 | A kind of aggregate gradation method for quickly identifying of non-contact measurement |
Non-Patent Citations (2)
| Title |
|---|
| 于沭 等: "基于图像识别技术的土石料级配检测***", 《中国水利水电科学研究院学报》 * |
| 王思雨: "基于DBN模型的土石料碾压参数分析与***实现", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112884739A (en) * | 2021-02-19 | 2021-06-01 | 武汉大学 | Rapid detection method for rockfill filling compactness based on deep learning network |
| CN112884739B (en) * | 2021-02-19 | 2023-08-18 | 武汉大学 | Deep learning network-based method for rapidly detecting filling compactness of rock-fill body |
| CN113813839A (en) * | 2021-02-24 | 2021-12-21 | 晋江华宝石业有限公司 | Quick grading method and grading device for square stock |
| WO2023279556A1 (en) * | 2021-07-09 | 2023-01-12 | 福建南方路面机械股份有限公司 | Aggregate particle size sampling and monitoring method |
| CN113674201A (en) * | 2021-07-09 | 2021-11-19 | 福建南方路面机械股份有限公司 | Aggregate particle size sampling monitoring method |
| CN113610771A (en) * | 2021-07-15 | 2021-11-05 | 三峡大学 | Rockfill material image detection grading curve correction method based on B-spline curve and particle swarm optimization |
| CN113610771B (en) * | 2021-07-15 | 2024-05-28 | 三峡大学 | Correction method for rock-fill image detection grading curve based on B-spline curve and particle swarm optimization |
| CN113554222B (en) * | 2021-07-19 | 2023-11-28 | 中国水利水电科学研究院 | Dynamic optimization and intelligent regulation configuration method for wide-source cementing material of cementing dam |
| CN113554222A (en) * | 2021-07-19 | 2021-10-26 | 中国水利水电科学研究院 | Dynamic optimization and intelligent regulation and control configuration method for bonding dam generalized bonding material |
| CN113487643A (en) * | 2021-07-19 | 2021-10-08 | 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 |
| CN113591981A (en) * | 2021-07-30 | 2021-11-02 | 上海建工四建集团有限公司 | Artificial intelligence-based existing terrazzo information survey method and system |
| CN113591981B (en) * | 2021-07-30 | 2024-02-09 | 上海建工四建集团有限公司 | Existing terrazzo information investigation method and system based on artificial intelligence |
| CN113658117B (en) * | 2021-08-02 | 2023-09-15 | 安徽省交通控股集团有限公司 | Method for identifying and dividing aggregate boundary in asphalt mixture based on deep learning |
| CN113658117A (en) * | 2021-08-02 | 2021-11-16 | 浙江大学 | Method for identifying and dividing aggregate boundaries in asphalt mixture based on deep learning |
| CN113740216B (en) * | 2021-09-09 | 2022-05-24 | 北京师范大学 | Air-ground integrated detection method for wide-gradation mixed aggregate |
| CN113740216A (en) * | 2021-09-09 | 2021-12-03 | 北京师范大学 | Air-ground integrated detection method for wide-gradation mixed aggregate |
| CN114612450A (en) * | 2022-03-17 | 2022-06-10 | 无锡东如科技有限公司 | Image detection segmentation method and system based on data augmentation machine vision and electronic equipment |
| CN114612450B (en) * | 2022-03-17 | 2023-09-05 | 无锡东如科技有限公司 | Image detection segmentation method and system based on data augmentation machine vision and electronic equipment |
| JP7496858B2 (en) | 2022-08-08 | 2024-06-07 | 基礎地盤コンサルタンツ株式会社 | Particle size distribution prediction device, machine learning device, particle size distribution prediction method, and machine learning method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112017164A (en) | Soil and stone material grading detection method based on depth threshold convolution model | |
| CN110516561B (en) | SAR image target identification method based on DCGAN and CNN | |
| CN108961217B (en) | Surface defect detection method based on regular training | |
| CN108765412B (en) | Strip steel surface defect classification method | |
| CN104990925B (en) | One kind is based on gradient multi thresholds optimization defect inspection method | |
| Deb et al. | An efficient method of vehicle license plate recognition based on sliding concentric windows and artificial neural network | |
| CN101510262B (en) | Automatic measurement method for separated-out particles in steel and morphology classification method thereof | |
| CN114627383B (en) | Small sample defect detection method based on metric learning | |
| CN106096610A (en) | A kind of file and picture binary coding method based on support vector machine | |
| CN109859187B (en) | Explosive-pile ore rock particle image segmentation method | |
| CN104268600B (en) | A kind of mineral flotation foam image texture analysis and operating mode's switch method based on Minkowski distances | |
| CN111754538B (en) | Threshold segmentation method for USB surface defect detection | |
| Mousavirad et al. | Design of an expert system for rice kernel identification using optimal morphological features and back propagation neural network | |
| CN113658117B (en) | Method for identifying and dividing aggregate boundary in asphalt mixture based on deep learning | |
| CN114926407A (en) | Steel surface defect detection system based on deep learning | |
| CN111539957A (en) | Image sample generation method, system and detection method for target detection | |
| CN111738367A (en) | Part classification method based on image recognition | |
| Han et al. | An enhanced image binarization method incorporating with Monte-Carlo simulation | |
| CN114219773A (en) | Pre-screening and calibration method for bridge crack detection data set | |
| CN1252588C (en) | High spectrum remote sensing image combined weighting random sorting method | |
| CN101510265A (en) | Association moving number character intelligent identification method | |
| CN109829511B (en) | Texture classification-based method for detecting cloud layer area in downward-looking infrared image | |
| Kusanti et al. | Combination of otsu and canny method to identify the characteristics of solo batik as Surakarta traditional batik | |
| CN103093241B (en) | Based on the remote sensing image nonuniformity cloud layer method of discrimination of homogeneity process | |
| CN104063874B (en) | Method for quantitatively evaluating gray level image binarization algorithms based on weight model |
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 |