CN111507391A - Intelligent identification method for nonferrous metal broken materials based on machine vision - Google Patents
Intelligent identification method for nonferrous metal broken materials based on machine vision Download PDFInfo
- Publication number
- CN111507391A CN111507391A CN202010287169.5A CN202010287169A CN111507391A CN 111507391 A CN111507391 A CN 111507391A CN 202010287169 A CN202010287169 A CN 202010287169A CN 111507391 A CN111507391 A CN 111507391A
- Authority
- CN
- China
- Prior art keywords
- ferrous metal
- model
- machine vision
- identification method
- training
- 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
Images
Classifications
-
- 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
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B07—SEPARATING SOLIDS FROM SOLIDS; SORTING
- B07C—POSTAL SORTING; SORTING INDIVIDUAL ARTICLES, OR BULK MATERIAL FIT TO BE SORTED PIECE-MEAL, e.g. BY PICKING
- B07C5/00—Sorting according to a characteristic or feature of the articles or material being sorted, e.g. by control effected by devices which detect or measure such characteristic or feature; Sorting by manually actuated devices, e.g. switches
- B07C5/34—Sorting according to other particular properties
- B07C5/342—Sorting according to other particular properties according to optical properties, e.g. colour
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
-
- 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
- G06V10/56—Extraction of image or video features relating to colour
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)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Multimedia (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an intelligent non-ferrous metal crushed material identification method based on machine vision, which comprises the following steps of 1) data acquisition, wherein in an actual separation scene, images of two non-ferrous metal types of copper and aluminum are acquired, a multi-target image data set of the non-ferrous metal crushed material is established and divided into a training set, a verification set and a test set, 2) image enhancement processing is carried out on the training set, 3) neural network model building and parameter initialization are carried out by adopting a YO L Ov3 model, 4) a loss function and an IOU threshold value of the model are determined, 5) an improved YO L Ov3 target detection model is obtained by training, and 6) the separation of the non-ferrous metal crushed material is realized by adopting the improved YO L Ov3 target detection model.
Description
Technical Field
The invention relates to a machine vision identification technology, in particular to an intelligent identification method for a non-ferrous metal broken material based on machine vision.
Background
With the continuous development of economy, the automobile retirement amount is larger and larger while the automobile keeping amount in China is increased year by year. The increase of the automobile scrapped quantity in a well-spraying mode causes a series of problems of environmental protection, resource regeneration and the like. If the number of scrapped automobiles which is increasingly expanded cannot be reasonably treated, not only can environmental pollution be caused, but also the serious waste of renewable resources in scrapped automobiles can be caused, and the retired automobiles contain a large number of aluminum and copper parts. The nonferrous metal crushed material for the retired automobile has extremely high market value, and the research on the sorting technology for nonferrous metal crushing of the retired automobile is beneficial to maximally recycling waste metal resources and reducing the environmental pollution caused by improper treatment of the waste metal.
At present, the separation of the nonferrous metal materials of the scraped car after crushing is mainly manually separated. The manual operation mode can realize the separation of aluminum and copper, but has low separation efficiency and high labor intensity, generates environmental pollution and influences the human health at the same time, and does not meet the requirements of the modern automobile industry and the social life development. In recent years, machine vision is adopted to solve the problem in order to reduce labor cost and improve the accuracy of identifying the waste metal.
Chinese patent document CN107876429A discloses an automatic sorting system for waste non-ferrous metals based on machine vision, which utilizes the difference of different kinds of metals in color and adopts a multilayer perceptron neural network to train non-ferrous metals, thereby identifying non-ferrous metals such as aluminum copper. However, this method has the following disadvantages: (1) a large number of samples are needed for training, and the real-time performance is not realized; (2) the method only carries out sorting through the color feature, the extracted features are few, and the recognition accuracy is poor.
Disclosure of Invention
The invention aims to solve the technical problem of providing an intelligent identification method of a non-ferrous metal broken material based on machine vision aiming at the defects in the prior art.
The technical scheme adopted by the invention for solving the technical problems is as follows: a non-ferrous metal broken material intelligent identification method based on machine vision comprises the following steps:
1) data acquisition, namely acquiring images of copper and aluminum two non-ferrous metal types in an actual sorting scene, and establishing a multi-target image data set of a non-ferrous metal crushed material, wherein the multi-target image data set is divided into a training set and a testing set;
2) carrying out image enhancement processing on the training set;
2.1) converting the image color space from RGB to HSV;
2.2) partitioning the image, and carrying out histogram equalization operation on the brightness (V) component of each partition in the HSV color space, wherein the operation cuts and equally divides gray level pixels exceeding a threshold value in the histogram to each gray level;
2.3) splicing the V component after the adjustment and the H, S component to obtain an enhanced image.
3) Establishing a neural network model and initializing parameters by adopting a YO L Ov3 model;
the construction of the YO L Ov3 model is completed by adopting a Darknet-53 network, the whole YO L O v3 structure has no pooling layer and a full connection layer, only uses a convolution layer and is a full convolution neural network, in forward propagation, the size transformation of tensor is realized by changing the step of convolution kernel, and better effect is ensured to be achieved on the basis of real-time performance (fps > 60);
initializing the model parameters according to the trained YO L Ov3 network model parameters of the COCO data set;
4) determining a loss function and an IOU threshold of the model;
loss function:
using leaky Re L U as the activation function of YO L Ov3, the specific formula is as follows:
the two-class loss function was used as the loss function for L oss function YO L O v3, as follows:
the derivation can be:
when the error is large, the weight is updated quickly, and when the error is small, the weight is updated slowly, so that the training speed can be greatly improved;
IOU threshold value:
obtaining a bounding box prediction result by adopting the following formula
bx=σ(tx)+cx
by=σ(ty)+cy
Wherein, the parameter that the network actually needs to learn is tx,ty,tw,thCalculating the coordinates of the center point and the width and the height b of the prediction frame according to the four formulasx,by,bw,bhWherein c isx,cyFor the number of grid offset of the current grid relative to the upper left corner grid, the function of sigma () is a logistic function, the coordinates are normalized to be between 0 and 1, and the finally obtained bx,byNormalized value against grid cell, pw,phThe width and height of the anchor box (prior box) with the largest coincidence with the grountruth. The model identification accuracy is improved by continuously adjusting the threshold value of the IOU.
5) Training to obtain an improved YO L Ov3 target detection model;
6) the improved YO L Ov3 target detection model is adopted to realize the sorting of the non-ferrous metal crushed materials.
The invention has the following beneficial effects:
1. on the basis of the original YO L Ov3 model suitable for a small sample, image enhancement processing is carried out, the copper and aluminum identification accuracy is improved by about 4%, and the identification effect is effectively improved;
2. on the basis of data enhancement processing, the best matching Focal L oss is selected to further improve the prediction accuracy;
3. based on the original YO L Ov3 model, the method adopts a series of improvements of data enhancement, improved Focal L oss and determination of the IOU optimal threshold value, so that the aluminum recognition accuracy can reach 95.3 percent and the copper recognition accuracy can reach 91.4 percent finally.
Drawings
The invention will be further described with reference to the accompanying drawings and examples, in which:
FIG. 1 is a flow chart of a method of an embodiment of the present invention;
FIG. 2 is a schematic diagram of a neural network architecture according to an embodiment of the present invention;
FIG. 3 is a diagram illustrating bounding box detection according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
Under the condition that the accuracy of general classification recognition algorithms CNN, RCNN and Fast-CNN is greatly reduced, a YO L Ov3 model is used to solve the problem by adopting a transfer learning mode, in order to improve the recognition accuracy, the conventional neural network classification recognition model cannot well meet the requirement under the condition that the recognition real-time requirement is high, therefore, the intelligent non-ferrous metal crushed material recognition method based on the YO L O (You Only L ook Once) v3 model is adopted in the invention.
As shown in FIG. 1, the intelligent identification method of the nonferrous metal broken materials based on machine vision comprises the following steps:
1) data acquisition, namely acquiring images of copper and aluminum two non-ferrous metal types in an actual sorting scene, establishing a multi-target image data set of the non-ferrous metal crushed material, labeling the types of the images, and dividing the images into a training set, a verification set and a test set;
and in the step 1), a CCD industrial camera is adopted to acquire images of the actual scene.
2) Carrying out image enhancement processing on the training set;
this phenomenon can degrade picture quality because of uneven lighting or insufficient light during image acquisition in natural scenes. In order to overcome the problem, the invention adopts an image enhancement preprocessing algorithm which is improved on an adaptive histogram equalization algorithm and limits contrast histogram equalization to realize image enhancement processing, and the image enhancement preprocessing algorithm specifically comprises the following steps:
2.1) converting the image color space from RGB to HSV;
2.2) partitioning the image, and carrying out histogram equalization operation on the brightness (V) component of each partition in the HSV color space, wherein the operation cuts and equally divides gray level pixels exceeding a threshold value in the histogram to each gray level;
2.3) splicing the adjusted and increased V component and H, S component to obtain an enhanced image;
after the preprocessing, the influence of illumination on the image quality can be reduced, the contrast is improved, the illumination diversity of the sample is increased, the image quality can be effectively improved, and the accuracy and the recall rate of model identification are improved.
3) Establishing a neural network model and initializing parameters by adopting a YO L Ov3 model;
as shown in FIG. 2, the construction of the YO L Ov3 model is completed by adopting a Darknet-53 network, the whole YO L O v3 structure has no pooling layer and a full connection layer, only uses convolution layer and is a full convolution neural network, in forward propagation, the size transformation of tensor is realized by changing the step of convolution kernel, and better effect is ensured to be achieved on the basis of real-time performance (fps > 60);
the generalization capability of the model is realized through transfer learning, the model parameters are initialized according to the YO L Ov3 network model parameters trained by the COCO data set, and the high accuracy can be achieved only by training a small sample.
4) Determining a loss function and an IOU threshold of the model;
loss function:
using leaky Re L U as the activation function of YO L Ov3, the specific formula is as follows:
the two-class loss function was used as the loss function for L oss function YO L O v3, as follows:
the derivation can be:
when the error is large, the weight is updated quickly, and when the error is small, the weight is updated slowly, so that the training speed can be greatly improved; when there is a situation where objects in the picture occlude each other, one box may belong to several objects, and it is more advantageous to use the two-class penalty function.
IOU threshold value:
obtaining a bounding box prediction result by adopting the following formula
bx=σ(tx)+cx
by=σ(ty)+cy
Wherein, the parameter that the network actually needs to learn is tx,ty,tw,thCalculating the coordinates of the center point and the width and the height b of the prediction frame according to the four formulasx,by,bw,bhWherein c isx,cyFor the number of grid offset of the current grid relative to the upper left corner grid, the function of sigma () is a logistic function, the coordinates are normalized to be between 0 and 1, and the finally obtained bx,byNormalized value against grid cell, pw,phThe width and height of the anchor box (prior box) with the largest coincidence with the grountruth. The model identification accuracy is improved by continuously adjusting the threshold value of the IOU.
5) Training to obtain an improved YO L Ov3 target detection model;
based on the improved YO L Ov3 convolutional neural network, a YO L Ov3 model is obtained through training.
6) The improved YO L Ov3 target detection model is adopted to realize the sorting of the non-ferrous metal crushed materials.
Comparing the identification results of the target detection models, and obtaining the following table:
therefore, on the basis of the original YO L Ov3 model, a series of improvements including data enhancement, improved Focal L oss and determination of the IOU optimal threshold are adopted, the aluminum recognition accuracy is finally enabled to reach 95.3%, the copper recognition accuracy reaches 91.4%, the detection model can achieve high accuracy only through small sample training, the training time and the recognition time are short, the real-time recognition effect can be achieved, and the industrial requirements are met.
It will be understood that modifications and variations can be made by persons skilled in the art in light of the above teachings and all such modifications and variations are intended to be included within the scope of the invention as defined in the appended claims.
Claims (4)
1. A non-ferrous metal broken material intelligent identification method based on machine vision is characterized by comprising the following steps:
1) data acquisition, namely acquiring images of copper and aluminum two non-ferrous metal types in an actual sorting scene, and establishing a multi-target image data set of the non-ferrous metal crushed material, wherein the multi-target image data set is divided into a training set, a verification set and a test set;
2) carrying out image enhancement processing on the training set to reduce the influence of illumination on the image quality;
3) establishing a neural network model and initializing parameters by adopting a YO L Ov3 model;
the construction of the YO L Ov3 model is completed by adopting a Darknet-53 network, the whole YO L O v3 structure has no pooling layer and a full connection layer, only a convolution layer is used, the whole YO L O v3 structure is a full convolution neural network, and in forward propagation, tensor size transformation is realized by changing the step of a convolution kernel;
initializing the model parameters according to the trained YO L Ov3 network model parameters of the COCO data set;
4) determining a loss function and an IOU threshold of the model;
5) training to obtain an improved YO L Ov3 target detection model;
6) the improved YO L Ov3 target detection model is adopted to realize the sorting of the non-ferrous metal crushed materials.
2. The intelligent non-ferrous metal broken material identification method based on machine vision according to claim 1, characterized in that the training set is subjected to image enhancement in the step 2), specifically as follows:
2.1) converting the image color space from RGB to HSV;
2.2) partitioning the image, and carrying out histogram equalization operation on the brightness component of each partition in the HSV color space, wherein the operation cuts and equally divides gray level pixels exceeding a threshold value in the histogram to each gray level;
2.3) splicing the V component after the adjustment and the H, S component to obtain an enhanced image.
3. The intelligent machine vision-based non-ferrous metal broken material identification method according to claim 1, wherein the loss function in the step 3) is as follows:
using leaky Re L U as the activation function of YO L Ov3, the specific formula is as follows:
the two-class loss function was used as the loss function for L oss function YO L O v3, as follows:
the derivation can be:
when the error is large, the weight is updated quickly, and when the error is small, the weight is updated slowly, so that the training speed can be greatly improved;
4. the intelligent machine vision-based non-ferrous metal broken material identification method according to claim 1, wherein the IOU threshold value in the step 3) is as follows:
obtaining a bounding box prediction result by adopting the following formula
bx=σ(tx)+cx
by=σ(ty)+cy
Wherein, the parameter that the network actually needs to learn is tx,ty,tw,thCalculating the coordinates of the center point and the width and the height b of the prediction frame according to the four formulasx,by,bw,bhWherein c isx,cyFor the number of grid offset of the current grid relative to the upper left corner grid, the function of sigma () is a logistic function, the coordinates are normalized to be between 0 and 1, and the finally obtained bx,byNormalized value against grid cell, pw,phThe width and height of the prior frame with the largest coincidence degree with the group route are shown.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010287169.5A CN111507391A (en) | 2020-04-13 | 2020-04-13 | Intelligent identification method for nonferrous metal broken materials based on machine vision |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010287169.5A CN111507391A (en) | 2020-04-13 | 2020-04-13 | Intelligent identification method for nonferrous metal broken materials based on machine vision |
Publications (1)
Publication Number | Publication Date |
---|---|
CN111507391A true CN111507391A (en) | 2020-08-07 |
Family
ID=71875991
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010287169.5A Pending CN111507391A (en) | 2020-04-13 | 2020-04-13 | Intelligent identification method for nonferrous metal broken materials based on machine vision |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111507391A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111680689A (en) * | 2020-08-11 | 2020-09-18 | 武汉精立电子技术有限公司 | Target detection method, system and storage medium based on deep learning |
CN112348791A (en) * | 2020-11-04 | 2021-02-09 | 中冶赛迪重庆信息技术有限公司 | Intelligent scrap steel detecting and judging method, system, medium and terminal based on machine vision |
CN113343179A (en) * | 2021-06-02 | 2021-09-03 | 江苏邦鼎科技有限公司 | Striking and crushing method and system based on oblique shearing |
CN113700978A (en) * | 2021-09-15 | 2021-11-26 | 中航华东光电(上海)有限公司 | Pipeline foreign matter detection device and detection method |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106000913A (en) * | 2016-05-25 | 2016-10-12 | 武汉理工大学 | Photoelectric color separate method and device for nonferrous metals of decommissioned cars |
CN107876429A (en) * | 2017-12-05 | 2018-04-06 | 湖南机电职业技术学院 | A kind of waste non-ferrous metals automatic sorting system based on machine vision |
AU2019101142A4 (en) * | 2019-09-30 | 2019-10-31 | Dong, Qirui MR | A pedestrian detection method with lightweight backbone based on yolov3 network |
-
2020
- 2020-04-13 CN CN202010287169.5A patent/CN111507391A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106000913A (en) * | 2016-05-25 | 2016-10-12 | 武汉理工大学 | Photoelectric color separate method and device for nonferrous metals of decommissioned cars |
CN107876429A (en) * | 2017-12-05 | 2018-04-06 | 湖南机电职业技术学院 | A kind of waste non-ferrous metals automatic sorting system based on machine vision |
AU2019101142A4 (en) * | 2019-09-30 | 2019-10-31 | Dong, Qirui MR | A pedestrian detection method with lightweight backbone based on yolov3 network |
Non-Patent Citations (3)
Title |
---|
刘国成: "人群异常行为数字图像处理与分析", 西南交通大学出版社, pages: 65 * |
缪希仁等: "基于深度卷积神经网络的输电线路防鸟刺部件识别与故障检测", vol. 45, no. 1, pages 127 - 129 * |
陈洋: "被动毫米波序列影像的金属违禁品检测研究", no. 07, pages 36 - 46 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111680689A (en) * | 2020-08-11 | 2020-09-18 | 武汉精立电子技术有限公司 | Target detection method, system and storage medium based on deep learning |
CN111680689B (en) * | 2020-08-11 | 2021-03-23 | 武汉精立电子技术有限公司 | Target detection method, system and storage medium based on deep learning |
CN112348791A (en) * | 2020-11-04 | 2021-02-09 | 中冶赛迪重庆信息技术有限公司 | Intelligent scrap steel detecting and judging method, system, medium and terminal based on machine vision |
CN112348791B (en) * | 2020-11-04 | 2023-03-14 | 中冶赛迪信息技术(重庆)有限公司 | Intelligent scrap steel detecting and judging method, system, medium and terminal based on machine vision |
CN113343179A (en) * | 2021-06-02 | 2021-09-03 | 江苏邦鼎科技有限公司 | Striking and crushing method and system based on oblique shearing |
CN113700978A (en) * | 2021-09-15 | 2021-11-26 | 中航华东光电(上海)有限公司 | Pipeline foreign matter detection device and detection method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111507391A (en) | Intelligent identification method for nonferrous metal broken materials based on machine vision | |
CN108960245B (en) | Tire mold character detection and recognition method, device, equipment and storage medium | |
CN111695514B (en) | Vehicle detection method in foggy days based on deep learning | |
US20160334546A1 (en) | Weather recognition method and device based on image information detection | |
CN108765443A (en) | A kind of mark enhancing processing method of adaptive color Threshold segmentation | |
CN109918971B (en) | Method and device for detecting number of people in monitoring video | |
CN113177924A (en) | Industrial production line product flaw detection method | |
CN110060221B (en) | Bridge vehicle detection method based on unmanned aerial vehicle aerial image | |
CN111896540B (en) | Water quality on-line monitoring system based on block chain | |
CN111611907A (en) | Image-enhanced infrared target detection method | |
CN110706235A (en) | Far infrared pedestrian detection method based on two-stage cascade segmentation | |
CN115100174B (en) | Ship sheet metal part paint surface defect detection method | |
CN109509188B (en) | Power transmission line typical defect identification method based on HOG characteristics | |
CN111540203B (en) | Method for adjusting green light passing time based on fast-RCNN | |
CN109325426B (en) | Black smoke vehicle detection method based on three orthogonal planes time-space characteristics | |
CN105354547A (en) | Pedestrian detection method in combination of texture and color features | |
CN112785568A (en) | Magnetic shoe defect segmentation method based on deep learning | |
CN110334703B (en) | Ship detection and identification method in day and night image | |
CN112200007A (en) | License plate detection and identification method under community monitoring scene | |
CN115115893B (en) | Intelligent sorting method for waste metal recovery | |
CN113128308A (en) | Pedestrian detection method, device, equipment and medium in port scene | |
Wang et al. | Low-light traffic objects detection for automated vehicles | |
CN202650248U (en) | Intelligent transportation control device based on image processing | |
CN111950409B (en) | Intelligent identification method and system for road marking line | |
CN114494728A (en) | Small target detection method based on deep learning |
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 |