CN115880223A - Improved YOLOX-based high-reflectivity metal surface defect detection method - Google Patents
Improved YOLOX-based high-reflectivity metal surface defect detection method Download PDFInfo
- Publication number
- CN115880223A CN115880223A CN202211403533.5A CN202211403533A CN115880223A CN 115880223 A CN115880223 A CN 115880223A CN 202211403533 A CN202211403533 A CN 202211403533A CN 115880223 A CN115880223 A CN 115880223A
- Authority
- CN
- China
- Prior art keywords
- metal surface
- yolox
- improved
- image
- defect detection
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
Abstract
The invention discloses a high-reflectivity metal surface defect detection method based on improved YOLOX, which comprises the steps of constructing an image data set of metal surface defects, inputting the data set into an image feature enhancement module for preprocessing, and manufacturing corresponding labels aiming at each type of defects; constructing a high-reflectivity metal surface defect detection model of improved YOLOX, which consists of three parts, namely a main feature extraction network CSPDarknet-53, a BT-FPN bidirectional feature fusion network and a classification prediction network YOLO Head, reducing the loss of feature information and enhancing the feature fusion performance; acquiring a pre-training model by using transfer learning, and training a metal defect data set by combining a Mosaic and Mixup data enhancement method to obtain a weight w; and inputting the weight w into the improved YOLOX network for classification prediction. Compared with the prior art, the method and the device improve the detection effect and detection speed of the high-reflectivity metal defects, and have better real-time performance.
Description
Technical Field
The invention belongs to the technical field of image processing detection, and particularly relates to a high-reflectivity metal surface defect detection method based on improved YOLOX.
Background
The aluminum material has high recycling rate and strong practicability, can not be replaced by other materials in the application of chemical industry, building industry and automobile industry, but the surface quality of the aluminum profile is more and more important while the demand is increased. Extruded aluminum profiles are also frequently used in automotive parts, train frames, building doors and windows, support structures for skyscrapers, and the like. During the production process, defects such as surface defects, scratches, blisters and the like often occur due to wear of the appliances and changes in process conditions. Surface defects generated during the production process must be correctly identified to prevent delivery of defective material to the customer. The defects may be ranked by sorting the respective summary files appropriately according to the severity and location of each defect.
At present, the main mode for detecting the surface quality of the aluminum profile is still manual sampling inspection and visual inspection judgment, but because the manual detection has low efficiency and the detection precision depends on the experience of workers, the requirements of greatly promoting the capacity optimization and expansion of the aluminum profile in China are far from being met. The existing technology for detecting the surface flaws of the aluminum profile in China lags behind and is far from the international advanced technical level, and a detection method which is more suitable for modern enterprises is urgently needed. Meanwhile, the surface failure detection of the aluminum profile equipment is mainly realized by human eyes, and on the contrary, in the industrial production field of high-texture materials such as cork, textiles and the like, a machine vision system is widely adopted for quality detection. For example, gonzalez Adrados and Pereira, applied image analysis techniques and discriminant analysis techniques to classify certain defects in cork to an accuracy of over 90%. Lopes and Pereira determined the quality of the cork panels by automated visual inspection techniques. In addition, jordanov used an automatic vision detection system to classify four different types of cork tiles, first they extracted features and then classified using a feed-forward neural network with an accuracy of up to 95%. Therefore, the defect detection by using machine vision in the aluminum profile related field is very necessary. The aluminum profile surface flaw detection based on the machine vision is a non-contact detection mode, and the information of the aluminum profile surface can be quickly and effectively extracted by utilizing the non-contact detection mode, so that the detection of the aluminum profile surface flaw is realized.
The reason that image analysis techniques are not common in the metal working industry is that extruded aluminum has high reflectivity, which leads to problems with the photopic imaging of the image and therefore makes it more difficult to perform the classification task. However, the problems of high cost, time-intensive property and the like of quality control through human eyes force the metal processing industry to develop towards the direction of automatic defect detection, and meanwhile, strong power is provided for the development of automatic defect detection of aluminum profiles.
Disclosure of Invention
The invention aims to: aiming at the problems pointed out in the background technology, the invention provides a high-reflectivity metal surface defect detection method based on improved YOLOX, which is used for solving the technical problems of low accuracy and low detection speed in the detection of various defects of high-reflectivity metal.
The technical scheme is as follows: the invention discloses a high-reflectivity metal surface defect detection method based on improved YOLOX, which comprises the following steps:
s1: constructing an image data set of the metal surface defects, inputting the data set into an image characteristic enhancement module for preprocessing, and manufacturing a corresponding label aiming at each type of defects;
s2: constructing a high-reflectivity metal surface defect detection model of improved YOLOX, which consists of three parts, namely a trunk feature extraction network CSPDarknet-53, a BT-FPN bidirectional feature fusion network and a classification prediction network YOLO Head, reducing loss of feature information and enhancing feature fusion performance;
s3: obtaining a pre-training model by using transfer learning, and training a metal defect data set by combining a Mosaic and Mixup data enhancement method to obtain a weight w;
s4: and inputting the weight w into a high-reflectivity metal surface defect detection model of improved YOLOX for classification prediction.
Further, the step S1 is implemented as follows:
s1.1: obtaining defect data samples required by detecting the defects of the highly reflective metal surface from the disclosed GC10-DET data set, wherein the defect data samples comprise four types of defect data of crease, oil point, indentation and welding line;
s1.2: preprocessing a data sample, converting the format of the data sample into a standard VOC data format, and labeling metal defects in image data by using a Labelme labeling tool, wherein labels of the types of the metal defects comprise creases, oil spots, indentations and welding seams;
s1.3: the method comprises the following steps of sequentially carrying out Gaussian filtering and gamma correction image enhancement processing on an image data set with a defect mark, carrying out Gaussian filtering denoising on the image data set, carrying out weighted average processing on each metal defect image, wherein the value of each pixel point is obtained by carrying out weighted average on the pixel point and other pixel points in a neighborhood, and the Gaussian weight calculation formula of each pixel point is as follows:
wherein x, y are the coordinates of each pixel point, and σ is the variance of x;
carrying out gamma correction processing on the image data set after Gaussian filtering processing, correcting the gray value of the image, enhancing the contrast, and carrying out one-time gamma product operation on each pixel point on the image, wherein a gamma correction formula is as follows:
s＝cr γ
wherein s is the gray value of the output pixel point, r is the gray value of the input pixel point, c and gamma are constants, and the gray value of the image is adjusted by setting the values of c and gamma to reach the optimum value;
s1.4: and (3) dividing the metal defect data set subjected to the feature enhancement treatment into a training set, a verification set and a test set according to the following steps of 1.
Further, the implementation process of step S2 is as follows:
s2.1: constructing a trunk feature extraction network CSPDarknet-53 by utilizing Focus, CBS, resblock body and SPP modules, and establishing connection between the trunk feature extraction network CSPDarknet-53 and a BT-FPN bidirectional feature fusion network;
s2.2: constructing a CBS module by utilizing a common convolution Conv, a standardized BN and a SilU activation function, splitting a residual block into a main part and a large residual edge shortconv to construct a CSP structure, constructing a Resblock body structure by adopting the CBS and the CSP structure, wherein the SPP module is composed of CBS and 4 pooling cores with the sizes of 1 × 1, 5 × 5, 9 × 9 and 13 × 13;
s2.3: a BT-FPN bidirectional feature fusion network with emphasis on texture information is used, information flow from bottom to top is added in the network, and quick connection is added in a P4 layer and a P5 layer;
s2.4: in the BT-FPN network structure, three input layers P3, P4 and P5 correspond to the outputs of the backbone networks of the Dark3, dark4 and Dark5 of three Dark modules, wherein the three input layers comprise a CSP2_ X layer, a CBS block, up-sampling and a connection function; the CSP2_ X layer replaces the remaining small stack with a CBS module stack, and the three associated output equations are as follows:
wherein Pk out Representing the output of the k-th image, conv representing the convolution operation, w kj Represents the jth node weight of the kth layer,
the input of the nth node of the kth layer of the image is represented, the feature extraction network has three outputs with different scales, namely a small target, a medium target and a large target, because the bottom-up information flow is enhanced, and the texture information acquisition capability of P4 and P5 is further enhanced;
s2.5: 3 primary feature layers with different sizes are respectively led out from a Resblock body structure of a main feature extraction network CSPDarknet-53, input into a bidirectional feature fusion network BT-FPN for feature fusion, and input into a YOLO Head for classification prediction after the feature fusion of the 3 enhanced feature layers.
Further, the step S3 is implemented as follows:
s3.1: using CSPDarknet-53 trained on an ImageNet data set as a pre-training backbone network model through transfer learning, inputting a training set into a high-reflectivity metal surface defect detection model of improved YOLOX for training, and training for 100 rounds;
s3.2: freezing the trunk of the high-reflectivity metal surface defect detection model of the improved YOLOX at the first stage and performing iterative training for 50 times by using a Mosaic and Mixup data enhancement method, thawing the trunk of the high-reflectivity metal surface defect detection model of the improved YOLOX at the second stage, and performing iterative training for 50 times to obtain a training weight w.
Further, the step S4 is implemented as follows:
s4.1: inputting the metal surface defect data set into a high-reflectivity metal surface defect detection model of improved YOLOX for training for 100 times to obtain 100 groups of weights w, and inputting the weight w with the minimum loss value into the high-reflectivity metal surface defect detection model of improved YOLOX;
s4.2: inputting a metal surface defect data set to be tested into a high-reflection metal surface defect detection model of improved YOLOX with optimal training weight, and verifying the defect detection effect of the high-reflection metal surface defect detection model of improved YOLOX by using a test set image.
Has the advantages that: compared with the prior art, the invention has the following beneficial effects: according to the invention, the diversity of sample data is increased by performing data enhancement on the high-reflectivity metal defect image data set, the BT-FPN bidirectional feature fusion network module is added, the loss of feature information is reduced, the feature fusion performance is enhanced, the CSP _ X layer is formed by connecting two parallel CBS modules and an X residual error unit in series, the feature fusion and learning capacity of the trunk network for extracting depth features is enhanced, so that the classification and the positioning of metal defects are performed, the multi-scale target detection effect is achieved, the accuracy and the detection speed of high-reflectivity metal defect detection are improved, and the real-time performance is better.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a diagram of highly reflective metal defect types according to an embodiment of the present invention;
FIG. 3 is a flow chart of the structure of the improved YOLOX algorithm in the embodiment of the invention;
FIG. 4 is a diagram of the backbone network structure of the improved YOLOX algorithm in the embodiment of the present invention;
FIG. 5 is a detailed BT-FPN feature extraction network structure diagram in the embodiment of the present invention;
FIG. 6 is a diagram illustrating a simple BT-FPN feature extraction structure according to an embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings.
As shown in fig. 1, a method for detecting defects on a highly reflective metal surface based on improved YOLOX disclosed in the embodiments of the present invention mainly includes the following steps:
step 1: and constructing an image data set of the metal surface defects, inputting the data set into an image feature enhancement module for preprocessing, and manufacturing corresponding labels aiming at each type of defects.
The data set source of the embodiment is GC10-DET metal defect image data, the metal defect image is shown in FIG. 2, the image data are different types of defect images caused by some factors in the production or transportation process, the collected metal defect image data are classified, the metal defects are classified into 4 types, namely, creases, oil spots, indentations and welding seams, and the number of the defects is shown in Table 1.
TABLE 1 Defect types and amounts
| Kind of defect | Fold line | Oil point | Indentation(s) | Weld seam |
| Number of | 368 | 421 | 374 | 467 |
Preprocessing a data sample, converting the format of the data sample into a standard VOC data format, labeling metal defects in image data by using a Labelme labeling tool, wherein labels of the types of the metal defects comprise creases, oil spots, indentations and welding seams, and performing data enhancement processing by rotating, cutting and other operations on sample data to expand a sample data set.
The method comprises the following steps of sequentially carrying out Gaussian filtering and gamma correction image enhancement processing on an image data set with marked defects, firstly carrying out Gaussian filtering denoising on the image data set, carrying out weighted average processing on each metal defect image, wherein the value of each pixel point is obtained by carrying out weighted average on the pixel point and other pixel points in a neighborhood, and the Gaussian weight calculation formula of each pixel point is as follows:
wherein x and y are coordinates of each pixel point, and sigma is the variance of x;
and then, carrying out gamma correction processing on the image data set after the Gaussian filtering processing, correcting the gray value of the image, and enhancing the contrast, namely, carrying out gamma product operation on each pixel point on the image once, wherein the gamma correction formula is as follows:
s＝cr γ
wherein s is the gray value of the output pixel point, r is the gray value of the input pixel point, and c and gamma are constants, and the gray value of the image is adjusted by setting the values of c and gamma to reach the optimum value.
And finally, dividing the metal defect data set subjected to the feature enhancement treatment into a training set, a verification set and a test set according to the following steps of 1.
And 2, step: a high-reflectivity metal surface defect detection model of improved YOLOX, which is composed of three parts, namely a trunk feature extraction network CSPDarknet-53, a BT-FPN bidirectional feature fusion network and a classification prediction network YOLO Head, is constructed, loss of feature information is reduced, and feature fusion performance is enhanced.
As shown in fig. 3, an improved YOLOX network structure was constructed. Constructing a high-reflectivity metal surface defect detection model of an improved Yolox algorithm, which consists of three parts, namely a main feature extraction network CSPDarknet-53, a bidirectional feature fusion network BT-FPN and a classification prediction network Yolonead; constructing a trunk feature extraction network CSPDarknet-53 by using Focus, CBS, resblock body and SPP modules; establishing connection between a trunk feature extraction network CSPDarknet-53 and a bidirectional feature fusion network, respectively leading out 3 primary feature layers with different sizes from a Resblock body structure of the trunk feature extraction network CSPDarknet-53, and inputting the primary feature layers into a BT-FPN bidirectional feature fusion network for feature fusion; inputting the 3 enhanced feature layers after feature fusion into a YOLO Head for classification prediction, wherein the specific network structure comprises:
the method comprises the following steps of constructing a CBS module by utilizing a common convolution Conv, a standardized BN and a SilU activation function, dividing a residual block into a main part and a large residual edge shortconv to construct a CSP structure, constructing a Resblock body structure by adopting the CBS and the CSP structure, wherein an SPP module is composed of the CBS and 4 pooling cores with the sizes of 1 × 1, 5 × 5, 9 × 9 and 13 × 13, and the CSPDarknet-53 network structure is shown in a table 2:
TABLE 2 CSPDarknet-53 network architecture
The Focus structure is that a pixel value channel of each image is obtained, each channel image is regenerated into 4 new channel images and then overlapped, the width and height information of the image is concentrated into the channel information, and compared with an original image 3 channel, the image is changed into 12 channels, so that a backbone network can keep more complete sampling information; the CSP layer structure is mainly formed by connecting two parallel CBS modules and an X residual error unit in series, and can be divided into two parallel branches, wherein one branch comprises a CBL module followed by X residual error units, and the other branch comprises a CBL module;
the BT-FPN bidirectional feature fusion network is constructed, as shown in fig. 5, a BT-FPN bidirectional feature fusion network with emphasis on texture information is used, in which information flow from bottom to top is added, and a shortcut connection is added in the P4 layer and the P5 layer, in the network structure of the BT-FPN, three input layers (P3, P4, P5) correspond to the outputs of (Dark 3, dark4, dark 5) backbone networks of three Dark modules, including CSP2_ X layer, CBS block, up-sampling, connection function, where CSP2_ X layer uses CBS module stack to replace the rest of small part stack, and three related output formulas are as follows:
wherein Pk out Representing the output of the k-th image, conv representing the convolution operation, w kj Represents the jth node weight of the kth layer,
the input of the nth node of the k layer of the image is represented, the total output of the feature extraction network is 3 different scales, namely a small target, a medium target and a large target, because the information flow from bottom to top is enhanced, the information flow from bottom to top is further enhancedThe texture information acquisition capability of p4 and p5 is enhanced, as shown in fig. 6;
the YOLO Head structure is characterized in that fused feature graphs are respectively sent to a YOLOX anchor-frame-free decoupling detection Head, a target classification feature graph, a target frame position regression feature graph and a target frame confidence coefficient regression graph are obtained through the YOLOX anchor-frame-free decoupling detection Head, corresponding metal defect type information can be obtained according to the target classification feature graph, and corresponding position information is obtained according to the target frame position regression feature graph.
As shown in fig. 4, the size of the metal defect image is 640 × 640 × 3, the metal defect image is input into a Focus module, the metal defect feature is preliminarily extracted, at this time, the channel is expanded to 12, the size of the metal defect image is 320 × 320 × 12, the size of the metal defect image is then changed to 160 × 160 × 64 through a CBS structure, the feature is further extracted by using a Resblock structure 5 times, an SPP structure is added into the last Resblock structure, the feature layer sizes of the five convolutions are 160 × 160 × 64, 80 × 80 × 128, 40 × 40 × 256, 20 × 20 × 512 and 10 × 10 × 1024 respectively, 4 feature layers are extracted from the main feature extraction network, and the feature layer sizes are respectively: the method comprises the steps of obtaining 4 output features by 80 multiplied by 128, 40 multiplied by 256, 20 multiplied by 512 and 10 multiplied by 1024, taking the features of three feature layers of P3, P4 and P5 as the input of a BT-FPN bidirectional feature fusion network, and respectively sending fused feature maps into a YOLOX anchor-free frame decoupling detection head for classification prediction.
And step 3: and (3) obtaining a pre-training model by using transfer learning, and training a metal defect data set by combining a Mosaic and Mixup data enhancement method to obtain a weight w.
Using CSPDarknet-53 trained on an ImageNet data set as a pre-training backbone network model through transfer learning, inputting a metal defect training set subjected to image enhancement processing into a high-reflectivity metal surface defect detection model of improved YOLOX for training for 100 rounds in total; freezing the trunk of the improved YOLOX high-reflectivity metal surface defect detection model in the first stage, performing iterative training for 50 times by using a Mosaic and Mixup data enhancement method, thawing the trunk of the improved YOLOX high-reflectivity metal surface defect detection model in the second stage, and performing iterative training for 50 times to obtain a training weight w.
And 4, step 4: and inputting the weight w into a high-reflectivity metal surface defect detection model of improved YOLOX for classification prediction.
Inputting the metal surface defect data set into a high-reflectivity metal surface defect detection model of improved YOLOX for training 100 times to obtain 100 groups of weights w, inputting the weight w with the minimum loss value into the high-reflectivity metal surface defect detection model of improved YOLOX, inputting the metal surface defect test data set into the high-reflectivity metal surface defect detection model of improved YOLOX with the optimal training weight introduced, and verifying the defect detection effect of the high-reflectivity metal surface defect detection model of improved YOLOX by using a test set image.
Claims (5)
1. A high-reflectivity metal surface defect detection method based on improved YOLOX is characterized by comprising the following steps:
s1: constructing an image data set of the metal surface defects, inputting the data set into an image characteristic enhancement module for preprocessing, and manufacturing a corresponding label aiming at each type of defects;
s2: constructing a high-reflectivity metal surface defect detection model of improved YOLOX, which consists of three parts, namely a trunk feature extraction network CSPDarknet-53, a BT-FPN bidirectional feature fusion network and a classification prediction network YOLO Head, reducing loss of feature information and enhancing feature fusion performance;
s3: acquiring a pre-training model by using transfer learning, and training a metal defect data set by combining a Mosaic and Mixup data enhancement method to obtain a weight w;
s4: and inputting the weight w into a high-reflectivity metal surface defect detection model of improved YOLOX for classification prediction.
2. The improved YOLOX-based high-reflectivity metal surface defect detection method according to claim 1, wherein the step S1 is implemented as follows:
s1.1: obtaining defect data samples required by detecting the defects of the highly reflective metal surface from the disclosed GC10-DET data set, wherein the defect data samples comprise four types of defect data of crease, oil point, indentation and welding line;
s1.2: preprocessing the data sample, converting the format of the data sample into a standard VOC data format, and labeling metal defects in the image data by using a Labelme labeling tool, wherein labels of the types of the metal defects comprise creases, oil spots, indentations and welding seams;
s1.3: the method comprises the following steps of sequentially carrying out Gaussian filtering and gamma correction image enhancement processing on an image data set with a marked defect, carrying out Gaussian filtering denoising on the image data set, carrying out weighted average processing on each metal defect image, wherein the value of each pixel point is obtained by carrying out weighted average on the pixel point and other pixel points in a neighborhood, and the Gaussian weight calculation formula of each pixel point is as follows:
wherein x and y are coordinates of each pixel point, and sigma is the variance of x;
carrying out gamma correction processing on the image data set after Gaussian filtering processing, correcting the gray value of the image, enhancing the contrast, and carrying out one-time gamma product operation on each pixel point on the image, wherein a gamma correction formula is as follows:
s＝cr γ
wherein s is the gray value of the output pixel point, r is the gray value of the input pixel point, c and gamma are constants, and the gray value of the image is adjusted to be optimal by setting the values of c and gamma;
s1.4: and (3) dividing the metal defect data set subjected to the characteristic enhancement treatment into a training set, a verification set and a test set according to the following steps of 8.
3. The improved YOLOX-based method for detecting defects on a highly reflective metal surface according to claim 1, wherein the step S2 is performed as follows:
s2.1: constructing a trunk feature extraction network CSPDarknet-53 by utilizing Focus, CBS, resblockbody and SPP modules, and establishing connection between the trunk feature extraction network CSPDarknet-53 and a BT-FPN bidirectional feature fusion network;
s2.2: constructing a CBS module by utilizing a common convolution Conv, a standardized BN and a Silu activation function, dividing a residual block into a main part and a large residual edge shortconv to construct a CSP structure, constructing a Resblockbody structure by adopting the CBS and the CSP structure, and constructing an SPP module by utilizing CBS and 4 pooling kernels with the sizes of 1 × 1, 5 × 5, 9 × 9 and 13 × 13;
s2.3: a BT-FPN bidirectional feature fusion network with emphasis on texture information is used, information flow from bottom to top is added in the network, and quick connection is added in a P4 layer and a P5 layer;
s2.4: in the BT-FPN network structure, three input layers P3, P4 and P5 correspond to the outputs of the backbone networks of the Dark3, dark4 and Dark5 of three Dark modules, wherein the backbone networks comprise a CSP2_ X layer, a CBS block, up-sampling and a connection function; the CSP2_ X layer replaces the remaining small stack with a CBS module stack, and the three associated output equations are as follows:
wherein Pk is out Denotes the output of the k-th image, conv denotes the convolution operation, w kj Represents the jth node weight of the kth layer,
the input of the nth node of the k layer of the image is represented, the feature extraction network has three outputs with different scales, namely a small target, a medium target and a large target, because of the bottom-upThe information flow is enhanced, and further the texture information acquisition capacity of P4 and P5 is enhanced;
s2.5: 3 initial feature layers with different sizes are respectively led out from a Resblockbody structure of a trunk feature extraction network CSPDarknet-53, input into a bidirectional feature fusion network BT-FPN for feature fusion, and input into a YOLO Head for classification prediction after the feature fusion.
4. The improved YOLOX-based method for detecting defects on a highly reflective metal surface according to claim 1, wherein the step S3 is implemented as follows:
s3.1: using CSPDarknet-53 trained on an ImageNet data set as a pre-training backbone network model through transfer learning, inputting a training set into a high-reflectivity metal surface defect detection model of improved YOLOX for training, and training for 100 rounds;
s3.2: freezing the trunk of the improved YOLOX high-reflectivity metal surface defect detection model in the first stage, performing iterative training for 50 times by using a Mosaic and Mixup data enhancement method, thawing the trunk of the improved YOLOX high-reflectivity metal surface defect detection model in the second stage, and performing iterative training for 50 times to obtain a training weight w.
5. The improved YOLOX-based high-reflectivity metal surface defect detection method of claim 1, wherein the step S4 is implemented as follows:
s4.1: inputting the metal surface defect data set into a high-reflectivity metal surface defect detection model of improved YOLOX for training for 100 times to obtain 100 groups of weights w, and inputting the weight w with the minimum loss value into the high-reflectivity metal surface defect detection model of improved YOLOX;
s4.2: inputting a metal surface defect data set to be tested into a high-reflection metal surface defect detection model of improved YOLOX with optimal training weight, and verifying the defect detection effect of the high-reflection metal surface defect detection model of improved YOLOX by using a test set image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211403533.5A CN115880223A (en) | 2022-11-10 | 2022-11-10 | Improved YOLOX-based high-reflectivity metal surface defect detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211403533.5A CN115880223A (en) | 2022-11-10 | 2022-11-10 | Improved YOLOX-based high-reflectivity metal surface defect detection method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115880223A true CN115880223A (en) | 2023-03-31 |
Family
ID=85759577
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211403533.5A Pending CN115880223A (en) | 2022-11-10 | 2022-11-10 | Improved YOLOX-based high-reflectivity metal surface defect detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115880223A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116883393A (en) * | 2023-09-05 | 2023-10-13 | 青岛理工大学 | Metal surface defect detection method based on anchor frame-free target detection algorithm |
| CN116958086A (en) * | 2023-07-21 | 2023-10-27 | 盐城工学院 | Metal surface defect detection method and system with enhanced feature fusion capability |
| CN117078608A (en) * | 2023-08-06 | 2023-11-17 | 武汉纺织大学 | Double-mask guide-based high-reflection leather surface defect detection method |
| CN117333383A (en) * | 2023-09-07 | 2024-01-02 | 广东奥普特科技股份有限公司 | Surface defect detection method, device and equipment |
| CN117455923A (en) * | 2023-12-26 | 2024-01-26 | 通达电磁能股份有限公司 | Insulator defect detection method and system based on YOLO detector |
| CN117934466A (en) * | 2024-03-21 | 2024-04-26 | 上海渺知科技有限公司 | Three-dimensional model auxiliary electromagnetic-optical fusion imaging metal structure detection method |
| CN117934466B (en) * | 2024-03-21 | 2024-06-04 | 上海渺知科技有限公司 | Three-dimensional model auxiliary electromagnetic-optical fusion imaging metal structure detection method |
-
2022
- 2022-11-10 CN CN202211403533.5A patent/CN115880223A/en active Pending
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116958086A (en) * | 2023-07-21 | 2023-10-27 | 盐城工学院 | Metal surface defect detection method and system with enhanced feature fusion capability |
| CN116958086B (en) * | 2023-07-21 | 2024-04-19 | 盐城工学院 | Metal surface defect detection method and system with enhanced feature fusion capability |
| CN117078608A (en) * | 2023-08-06 | 2023-11-17 | 武汉纺织大学 | Double-mask guide-based high-reflection leather surface defect detection method |
| CN117078608B (en) * | 2023-08-06 | 2024-01-26 | 武汉纺织大学 | Double-mask guide-based high-reflection leather surface defect detection method |
| CN116883393A (en) * | 2023-09-05 | 2023-10-13 | 青岛理工大学 | Metal surface defect detection method based on anchor frame-free target detection algorithm |
| CN116883393B (en) * | 2023-09-05 | 2023-12-01 | 青岛理工大学 | Metal surface defect detection method based on anchor frame-free target detection algorithm |
| CN117333383A (en) * | 2023-09-07 | 2024-01-02 | 广东奥普特科技股份有限公司 | Surface defect detection method, device and equipment |
| CN117333383B (en) * | 2023-09-07 | 2024-05-24 | 广东奥普特科技股份有限公司 | Surface defect detection method, device and equipment |
| CN117455923A (en) * | 2023-12-26 | 2024-01-26 | 通达电磁能股份有限公司 | Insulator defect detection method and system based on YOLO detector |
| CN117455923B (en) * | 2023-12-26 | 2024-03-15 | 通达电磁能股份有限公司 | Insulator defect detection method and system based on YOLO detector |
| CN117934466A (en) * | 2024-03-21 | 2024-04-26 | 上海渺知科技有限公司 | Three-dimensional model auxiliary electromagnetic-optical fusion imaging metal structure detection method |
| CN117934466B (en) * | 2024-03-21 | 2024-06-04 | 上海渺知科技有限公司 | Three-dimensional model auxiliary electromagnetic-optical fusion imaging metal structure detection method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115880223A (en) | Improved YOLOX-based high-reflectivity metal surface defect detection method | |
| CN110806736B (en) | Method for detecting quality information of forge pieces of die forging forming intelligent manufacturing production line | |
| CN113674247B (en) | X-ray weld defect detection method based on convolutional neural network | |
| CN108918536B (en) | Tire mold surface character defect detection method, device, equipment and storage medium | |
| CN111861978B (en) | Bridge crack example segmentation method based on Faster R-CNN | |
| US10803573B2 (en) | Method for automated detection of defects in cast wheel products | |
| CN111080622B (en) | Neural network training method, workpiece surface defect classification and detection method and device | |
| CN109035233B (en) | Visual attention network system and workpiece surface defect detection method | |
| CN110490842B (en) | Strip steel surface defect detection method based on deep learning | |
| CN107328787A (en) | A kind of metal plate and belt surface defects detection system based on depth convolutional neural networks | |
| CN110992349A (en) | Underground pipeline abnormity automatic positioning and identification method based on deep learning | |
| CN111798447B (en) | Deep learning plasticized material defect detection method based on fast RCNN | |
| CN113643268A (en) | Industrial product defect quality inspection method and device based on deep learning and storage medium | |
| CN110503638B (en) | Spiral adhesive quality online detection method | |
| CN113469951B (en) | Hub defect detection method based on cascade region convolutional neural network | |
| CN112819748B (en) | Training method and device for strip steel surface defect recognition model | |
| CN111179250A (en) | Industrial product defect detection system based on multitask learning | |
| CN113393438B (en) | Resin lens defect detection method based on convolutional neural network | |
| CN115482220A (en) | High-reflectivity metal surface defect detection method based on improved fast RCNN | |
| CN113393426A (en) | Method for detecting surface defects of rolled steel plate | |
| CN114581782A (en) | Fine defect detection method based on coarse-to-fine detection strategy | |
| CN113111875A (en) | Seamless steel rail weld defect identification device and method based on deep learning | |
| CN111951218A (en) | Lipstick quality inspection system of mixed deep learning model | |
| CN111768380A (en) | Method for detecting surface defects of industrial spare and accessory parts | |
| CN114119505A (en) | Method and device for detecting chip adhesion area defects |
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 |