CN114782409B - Vehicle surface abnormality detection method based on side direction camera - Google Patents
Vehicle surface abnormality detection method based on side direction camera Download PDFInfo
- Publication number
- CN114782409B CN114782409B CN202210564080.8A CN202210564080A CN114782409B CN 114782409 B CN114782409 B CN 114782409B CN 202210564080 A CN202210564080 A CN 202210564080A CN 114782409 B CN114782409 B CN 114782409B
- Authority
- CN
- China
- Prior art keywords
- image
- vehicle
- convolutional
- images
- network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 17
- 230000005856 abnormality Effects 0.000 title claims abstract description 13
- 230000011218 segmentation Effects 0.000 claims description 21
- 238000012549 training Methods 0.000 claims description 18
- 230000002159 abnormal effect Effects 0.000 claims description 13
- 238000012545 processing Methods 0.000 claims description 13
- 238000000034 method Methods 0.000 claims description 9
- 238000013527 convolutional neural network Methods 0.000 claims description 8
- 238000003709 image segmentation Methods 0.000 claims description 7
- 230000000873 masking effect Effects 0.000 claims description 6
- 238000012795 verification Methods 0.000 claims description 6
- 238000002372 labelling Methods 0.000 claims description 4
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000012544 monitoring process Methods 0.000 claims description 3
- 238000006386 neutralization reaction Methods 0.000 claims description 3
- 238000012360 testing method Methods 0.000 claims description 3
- 238000004519 manufacturing process Methods 0.000 claims description 2
- 238000013461 design Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000011866 long-term treatment Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005498 polishing Methods 0.000 description 1
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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- 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
-
- G06T5/77—
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- 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
- G06T2207/10016—Video; Image sequence
-
- 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
- G06T2207/30248—Vehicle exterior or interior
- G06T2207/30252—Vehicle exterior; Vicinity of vehicle
-
- 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
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Abstract
The invention belongs to the technical field of vehicle surface detection, and relates to a vehicle surface abnormality detection method based on a side direction camera.
Description
Technical Field
The invention belongs to the technical field of vehicle surface detection, and particularly relates to a vehicle surface abnormality detection method based on a side direction camera.
Background
With the rapid development of technologies such as artificial intelligence and computer vision, intelligent traffic plays an important role in intelligent cities, and the intelligent traffic covers aspects of roads in cities, and one problem that is relatively easy to ignore in intelligent traffic is anomaly detection of vehicle surfaces, such as slight scratches, breakage and the like, which are often ignored or difficult to find by human eyes, and if the use of vehicles is not affected by long-term treatment, a vehicle surface flaw detection method and a vehicle surface flaw detection system are disclosed in CN110596116a at present; the method for detecting the surface flaws of the vehicle comprises the following steps: s1, uniformly polishing a surface to be tested of a vehicle, and then continuously acquiring images of the surface to be tested of the vehicle by using a camera to acquire sampling images; s2, performing image fusion on all the sampling images to obtain a spliced image; s3, performing image processing on the spliced images to display a flaw area, introducing new detection equipment besides a camera, for example, detecting the whole polished surface of the vehicle, wherein the technologies can only detect large scratches, large concave-convex surfaces and the like on the surface of the vehicle, small flaws and small scratches on the surface of the vehicle are difficult to detect, and the introduction of the new detection equipment improves the deployment difficulty for on-site deployment. Therefore, it is desirable to design a novel vehicle surface abnormality detection method that accurately detects a vehicle surface abnormality region on the basis of using existing equipment.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, and designs and provides a vehicle surface abnormality detection method based on a side azimuth camera.
In order to achieve the above purpose, the specific process of the invention for realizing the detection of the abnormal surface of the vehicle is as follows:
(1) Constructing a vehicle segmentation data set: collecting video images of a road side parking lot side azimuth camera, segmenting and labeling the vehicle surface in the images, classifying pixels in the images into vehicle surface classes and background classes to obtain segmented data sets, and dividing the segmented data sets into training sets, verification sets and test sets according to actual needs;
(2) Segmenting the vehicle surface image using a segmentation network: scaling the pixels of the vehicle surface image to 1333 x 800, selecting an image segmentation network to segment the vehicle surface image, and setting the energy loss of the vehicle surface image;
(3) Masking and repairing the vehicle surface image: scaling each vehicle surface image obtained in the step (2) to 512 x 256 resolution, carrying out mask processing on each vehicle in sequence under the condition that two vehicles are segmented in the image acquired by the fisheye camera, carrying out mask processing on the mask with the size of 64 x 32 pixels from the upper left corner to the lower right corner in sequence, and carrying out no processing on the background part, wherein the number of masks of each vehicle is at most 64; selecting 64 images from the images after mask processing, and inputting the images into a convolutional neural network for restoration to obtain the images after mask restoration;
(4) Comparing the original image with the image repaired by the mask: comparing the image repaired by the mask obtained in the step (3) with an original image, judging the similarity of each 64 positions repaired by the image, judging that the positions are not damaged if the original surface image is similar to the repaired image, and judging that the positions are damaged if the surface image is dissimilar to the repaired image;
(5) Training the network to obtain a trained parameter model: manufacturing a data set by utilizing the segmentation result in the step (2), scaling 1000 segmented images, masking each image from the upper left corner to the lower right corner in sequence, taking the images as input training samples of a network, outputting the images which are not masked, calculating loss errors by adopting absolute value loss, updating parameters by back propagation, and after training and iterating through a complete training set for a set number of times (500 times), storing model parameters with the best result on a verification set as parameters trained by a final model, thus obtaining trained image restoration network parameters;
(6) The result is obtained by reasoning: inputting the video image of the side direction monitoring camera into the image restoration network trained in the step (5) for forward reasoning to obtain a restored vehicle surface image, and judging whether the shot vehicle has surface abnormality or not through the step (4).
Further, the image segmentation network model selected in the step (2) is a dynamic and rapid semantic segmentation model (refer to SOLOv2: dynamic and Fast Instance Segmentation (2020 NeurIPS)), and the network structure and the segmentation energy loss are both structures and loss functions adopted in the literature.
Further, in the step (3), the convolutional neural network includes nine convolutional modules, namely, a convolutional module 1, a convolutional module 2, a convolutional module 3, a convolutional module 4, a convolutional module 5, a convolutional module 4, a convolutional module 3, a convolutional module 2 and a convolutional module 1 in order from left to right, wherein a convolutional kernel of a first layer of the convolutional layer in the leftmost convolutional module 1 is 192×3×3, 192 indicates that 64 color images are input into the convolutional neural network together as an image group, 3*3 indicates that 3*3 convolutional is adopted in an image coordinate plane, the layer has 64 convolutional kernels in total, the convolutional of a second layer in the convolutional module 1 is 64×3×3, the obtained size of a characteristic layer is 64×512×256, 64 is the thickness of the characteristic layer, 512×256 is the dimension of the plane of the characteristic layer, and the number of the convolutional kernels in the convolutional modules 2, 3, 4 and 5 is 128×256, 512 and 512 in order.
Further, in the step (4), the similarity determination is performed by using a structural similarity criterion (SSIM), and the SSIM calculation process is as follows:,
wherein the neutralization is respectively the original imageI g And repairing an imageI d And respectively the original imageI g And repairing an imageI d Is covariance, c 1 And c 2 Is constant, c 1 Take 6.5025, c 2 Taking 58.5225; SSIM valueThe more the range is 0-1, the more the image obtained by restoration is close to the original real image, if SSIM is more than or equal to 0.8, the original surface image is judged to be not abnormal, if SSIM is not less than 0.8<And 0.8, judging that the original vehicle surface is damaged, and giving an alarm prompt.
Further, the training sample picture input in step (5) has a size ofSequentially inputting the batch sizes B into a network, wherein the input of the whole network is that。
Compared with the prior art, the invention has the beneficial effects that: the method comprises the steps of training an image segmentation network through a data set, constructing an image restoration data set on the segmented image, constructing an image restoration network model, carrying out restoration processing on an original image after mask processing, calculating the similarity between the original image and the restored image through a structural similarity criterion, considering that the position is abnormal if the similarity is smaller than a threshold value, considering that the position is abnormal if the similarity is larger than or equal to the threshold value, traversing from the upper left corner to the lower right corner, considering that the visible part of the vehicle is not abnormal if all image blocks are similar to the original image after restoration, and considering that the surface of the vehicle is abnormal if the abnormal blocks are present, so that the abnormality of the surface of the vehicle can be rapidly and effectively detected.
Drawings
Fig. 1 is a diagram of an image restoration network according to the present invention.
Fig. 2 is a schematic diagram of one of the masks on the surface of the vehicle according to the present invention, wherein 6 is the vehicle and 7 is the mask.
Fig. 3 is a schematic of the workflow of the present invention.
Detailed Description
The invention is further described below by way of examples in connection with the accompanying drawings, but in no way limit the scope of the invention.
Examples:
the specific process of detecting the abnormal vehicle surface based on the side direction camera in the embodiment of the invention is shown in fig. 3, and specifically comprises the following steps:
(1) Vehicle segmentation dataset construction:
collecting video images of a road side parking lot side azimuth camera, segmenting and labeling the vehicle surface in the images, classifying pixels in the images into an automobile surface class and a background class, wherein the automobile surface is not labeled by tires when the segmentation and labeling are performed due to the fact that the abnormality of the vehicle surface is detected, and the abnormal image data of the vehicle surface do not need to be collected independently because the abnormal vehicle surface sample is relatively rare, and the collected segmentation data set is divided into a training set, a verification set and a test set;
(2) Segmentation of the vehicle surface image using an example segmentation network:
and (3) utilizing the vehicle segmentation data constructed in the step (1) to segment the vehicle surface area by selecting an image segmentation network, and setting the energy loss of the vehicle surface area, wherein the selected image segmentation network model is as follows: dynamic and fast semantic segmentation model (reference SOLOv2: dynamic and Fast Instance Segmentation (2020 NeurIPS)), network structure and segmentation energy loss are both the structure and loss function employed in this document; because the resolution of the image acquired by the original side-view camera is 1920×1080, the image is scaled to 1333×800 before being segmented, and the size of the output segmentation result image is 1333×800;
(3) Masking and repairing the vehicle surface image:
scaling each vehicle surface image obtained in the step (2) to 512 x 256 resolution, for a fisheye camera, if the image is segmented into two vehicles, sequentially performing mask processing on each vehicle, wherein the mask size is 64 x 32 pixels, sequentially performing mask processing from the upper left corner to the lower right corner, and the background part does not need to be processed, so that the number of masks of each vehicle is 64 at most;
constructing an image restoration network shown in fig. 1, inputting 64 images into the convolutional neural network together, unlike common image restoration, the common image restoration algorithm is to input one image with a mask and output the image with the mask restoration, while the algorithm is to input 64 images with the mask at one time and output the image with the mask restoration, in a leftmost convolution module 1 of the image restoration network, the convolution kernel of a first layer of convolution layer is 192 x 3, 192 represents 64 color images as an image group to be input into the convolutional neural network together, 3*3 represents 3*3 convolution in an image coordinate plane, the layer has 64 convolution kernels in total, the convolution of a second convolution layer in the convolution module is 64 x 3, the size of an obtained feature layer is 64 x 512 x 256, 64 is the thickness of the feature layer, and 512 x 256 is the dimension of the feature layer plane; the number of convolution kernels in the convolution modules 2, 3, 4 and 5 is 128, 256, 512 and 512 in sequence;
(4) Comparing the original image with the restored image result to detect the abnormal surface of the vehicle:
comparing the restoration result obtained in the step (3) with an original image, performing similarity judgment on each of 64 positions subjected to image restoration, judging that the position is not damaged if the original surface image is similar to the restored image, judging that the position is damaged if the surface image is dissimilar to the restored image, selecting the similarity judgment as a structural similarity criterion (SSIM),
the SSIM calculation method is as follows:,
wherein the neutralization is respectively the original imageI g And repairing an imageI d And respectively the original imageI g And repairing an imageI d Is covariance, c 1 And c 2 For a constant, c1 takes 6.5025 and c2 takes 58.5225; the value range of the SSIM is 0-1, the closer to the original real image, the more the restored image is, if the SSIM is more than or equal to 0.8, the original surface image is judged to be abnormal, and if the SSIM is more than or equal to 0.8<0.8, judging that the original vehicle surface is damaged, and carrying out alarm prompt;
(5) Training a network to obtain a trained parameter model:
making a data set by using the segmentation result in the step (2), scaling 1000 segmented images, masking each image sequentially from the upper left corner to the lower right corner, and outputting the images as input training samples of a network, wherein the network outputs the images which are not masked, and the input images have the size ofSequentially inputting into the network according to batch size (B), inputting into the whole networkCalculating a loss error by adopting absolute value loss, updating parameters through back propagation, and after 500 complete training set training iterations, storing the model parameters with the best results on the verification set as the parameters trained by the final model, thus obtaining the trained image restoration network parameters;
(6) The result is obtained by reasoning:
inputting the video image of the side direction monitoring camera into the trained network in the step (5) for forward reasoning to obtain a repaired vehicle surface image, and judging whether the photographed vehicle has surface abnormality or not through the step (4).
Techniques not described in detail herein are all generic or prior art in the art.
It should be noted that the present embodiment is intended to aid further understanding of the present invention, but those skilled in the art will understand that: various alternatives and modifications are possible without departing from the spirit and scope of the invention and the appended claims. Therefore, the invention should not be limited to the disclosed embodiments, but rather the scope of the invention is defined by the appended claims.
Claims (5)
1. The vehicle surface abnormality detection method based on the side direction camera is characterized by comprising the following specific steps of:
(1) Constructing a vehicle segmentation data set: collecting video images of a road side parking lot side azimuth camera, segmenting and labeling the vehicle surface in the images, classifying pixels in the images into vehicle surface classes and background classes to obtain segmented data sets, and dividing the segmented data sets into training sets, verification sets and test sets according to actual needs;
(2) Segmenting the vehicle surface image using a segmentation network: scaling the pixels of the vehicle surface image to 1333 x 800, selecting an image segmentation network to segment the vehicle surface image, and setting the energy loss of the vehicle surface image;
(3) Masking and repairing the vehicle surface image: scaling each vehicle surface image obtained in the step (2) to 512 x 256 resolution, carrying out mask processing on each vehicle in sequence under the condition that two vehicles are segmented in the image acquired by the fisheye camera, carrying out mask processing on the mask with the size of 64 x 32 pixels from the upper left corner to the lower right corner in sequence, and carrying out no processing on the background part, wherein the number of masks of each vehicle is at most 64; selecting 64 images from the images after mask processing, and inputting the images into a convolutional neural network for restoration to obtain the images after mask restoration;
(4) Comparing the original image with the image repaired by the mask: comparing the image repaired by the mask obtained in the step (3) with an original image, judging the similarity of each 64 positions repaired by the image, judging that the positions are not damaged if the original surface image is similar to the repaired image, and judging that the positions are damaged if the surface image is dissimilar to the repaired image;
(5) Training the network to obtain a trained parameter model: manufacturing a data set by utilizing the segmentation result in the step (2), scaling 1000 segmented images, masking each image from the upper left corner to the lower right corner in sequence, taking the images as input training samples of a network, outputting the images which are not masked, calculating loss errors by adopting absolute value loss, updating parameters by back propagation, and storing model parameters with the best results on a verification set after complete training set training iteration for a set number of times as parameters trained by a final model, so as to obtain trained image restoration network parameters;
(6) The result is obtained by reasoning: inputting the video image of the side direction monitoring camera into the image restoration network trained in the step (5) for forward reasoning to obtain a restored vehicle surface image, and judging whether the shot vehicle has surface abnormality or not through the step (4).
2. The method for detecting surface anomalies of a vehicle based on a side-view camera of claim 1, wherein the image segmentation network model selected in step (2) is a dynamic and fast semantic segmentation model.
3. The vehicle surface anomaly detection method based on the side-view camera according to claim 2, wherein the convolutional neural network in the step (3) includes nine convolutional modules, namely, a convolutional module 1, a convolutional module 2, a convolutional module 3, a convolutional module 4, a convolutional module 5, a convolutional module 4, a convolutional module 3, a convolutional module 2 and a convolutional module 1 in sequence from left to right, a convolutional kernel of a first layer of the convolutional modules 1 is 192×3×3, 192 indicates that 64 color images are input to the convolutional neural network together as an image group, 3*3 indicates that 3*3 convolutions are adopted in an image coordinate plane, the layer has 64 convolutions, a convolution of a second convolution layer of the convolutional modules 1 is 64×3×512, 64 indicates a thickness of a characteristic layer, 512×256 indicates a dimension of a plane of the characteristic layer, and the number of the convolutions is 128, 512×256 and 512×512 in sequence in the convolutional modules 2, 3, 4 and 5.
4. The vehicle surface anomaly detection method based on the side-view camera according to claim 3, wherein the similarity determination in the step (4) is performed by using a structural similarity criterion SSIM, and the SSIM calculation process is as follows:
,
wherein the neutralization is respectively the original imageI g And repairing an imageI d And respectively the original imageI g And repairing an imageI d Is covariance, c 1 And c 2 Is constant, c 1 Take 6.5025, c 2 Taking 58.5225; the value range of the SSIM is 0-1, the closer to the original real image, the more the restored image is, if the SSIM is more than or equal to 0.8, the original surface image is judged to be abnormal, and if the SSIM is more than or equal to 0.8<And 0.8, judging that the original vehicle surface is damaged, and giving an alarm prompt.
5. The method for detecting surface anomalies of a vehicle based on a side-view camera as recited in claim 4, wherein the training sample picture input in step (5) is of a size ofSequentially inputting the batch sizes B into a network, wherein the input of the whole network is that。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210564080.8A CN114782409B (en) | 2022-05-23 | 2022-05-23 | Vehicle surface abnormality detection method based on side direction camera |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210564080.8A CN114782409B (en) | 2022-05-23 | 2022-05-23 | Vehicle surface abnormality detection method based on side direction camera |
Publications (2)
Publication Number | Publication Date |
---|---|
CN114782409A CN114782409A (en) | 2022-07-22 |
CN114782409B true CN114782409B (en) | 2024-03-29 |
Family
ID=82408793
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210564080.8A Active CN114782409B (en) | 2022-05-23 | 2022-05-23 | Vehicle surface abnormality detection method based on side direction camera |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114782409B (en) |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
AU2020103901A4 (en) * | 2020-12-04 | 2021-02-11 | Chongqing Normal University | Image Semantic Segmentation Method Based on Deep Full Convolutional Network and Conditional Random Field |
CN113095277A (en) * | 2021-04-27 | 2021-07-09 | 东南大学 | Unmanned aerial vehicle aerial photography vehicle detection method based on target space distribution characteristics |
CN113658115A (en) * | 2021-07-30 | 2021-11-16 | 华南理工大学 | Image anomaly detection method for generating countermeasure network based on deep convolution |
-
2022
- 2022-05-23 CN CN202210564080.8A patent/CN114782409B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
AU2020103901A4 (en) * | 2020-12-04 | 2021-02-11 | Chongqing Normal University | Image Semantic Segmentation Method Based on Deep Full Convolutional Network and Conditional Random Field |
CN113095277A (en) * | 2021-04-27 | 2021-07-09 | 东南大学 | Unmanned aerial vehicle aerial photography vehicle detection method based on target space distribution characteristics |
CN113658115A (en) * | 2021-07-30 | 2021-11-16 | 华南理工大学 | Image anomaly detection method for generating countermeasure network based on deep convolution |
Non-Patent Citations (1)
Title |
---|
阮航 ; 王立春 ; .基于特征图的车辆检测和分类.计算机技术与发展.2018,(第11期),第39-43页. * |
Also Published As
Publication number | Publication date |
---|---|
CN114782409A (en) | 2022-07-22 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109671071B (en) | Underground pipeline defect positioning and grade judging method based on deep learning | |
CN111709416B (en) | License plate positioning method, device, system and storage medium | |
CN111325721A (en) | Gas leakage detection method and system based on infrared thermal imaging | |
WO2021238420A1 (en) | Image defogging method, terminal, and computer storage medium | |
CN114898352A (en) | Method for simultaneously realizing image defogging and license plate detection | |
Wu et al. | Surface crack detection based on image stitching and transfer learning with pretrained convolutional neural network | |
Zhu et al. | Object detection in complex road scenarios: improved YOLOv4-tiny algorithm | |
CN116612106A (en) | Method for detecting surface defects of optical element based on YOLOX algorithm | |
WO2024021461A1 (en) | Defect detection method and apparatus, device, and storage medium | |
Peng et al. | Non-uniform illumination image enhancement for surface damage detection of wind turbine blades | |
CN114972202A (en) | Ki67 pathological cell rapid detection and counting method based on lightweight neural network | |
Li et al. | Fabric defect segmentation system based on a lightweight GAN for industrial Internet of Things | |
Saleh et al. | Tire defect detection model using machine learning | |
CN114612803A (en) | Transmission line insulator defect detection method for improving CenterNet | |
CN111325724B (en) | Tunnel crack region detection method and device | |
CN114782409B (en) | Vehicle surface abnormality detection method based on side direction camera | |
CN111612803B (en) | Vehicle image semantic segmentation method based on image definition | |
CN117252817A (en) | Transparent conductive film glass surface defect detection method and system | |
CN114596244A (en) | Infrared image identification method and system based on visual processing and multi-feature fusion | |
CN115797314B (en) | Method, system, equipment and storage medium for detecting surface defects of parts | |
Park et al. | Patch-based crack detection in black box road images using deep learning | |
CN111160282A (en) | Traffic light detection method based on binary Yolov3 network | |
CN115082650A (en) | Implementation method of automatic pipeline defect labeling tool based on convolutional neural network | |
CN113763261A (en) | Real-time detection method for far and small targets under sea fog meteorological condition | |
CN113298857A (en) | Bearing defect detection method based on neural network fusion strategy |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |