CN112365478A - Motor commutator surface defect detection model based on semantic segmentation - Google Patents
Motor commutator surface defect detection model based on semantic segmentation Download PDFInfo
- Publication number
- CN112365478A CN112365478A CN202011269251.1A CN202011269251A CN112365478A CN 112365478 A CN112365478 A CN 112365478A CN 202011269251 A CN202011269251 A CN 202011269251A CN 112365478 A CN112365478 A CN 112365478A
- Authority
- CN
- China
- Prior art keywords
- network
- segmentation
- samples
- convolution
- detection model
- 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
- 230000011218 segmentation Effects 0.000 title claims abstract description 71
- 230000007547 defect Effects 0.000 title claims abstract description 47
- 238000001514 detection method Methods 0.000 title claims abstract description 35
- 238000012549 training Methods 0.000 claims description 17
- 210000004205 output neuron Anatomy 0.000 claims description 15
- 230000002950 deficient Effects 0.000 claims description 14
- 238000011176 pooling Methods 0.000 claims description 14
- 238000000034 method Methods 0.000 claims description 10
- 238000012360 testing method Methods 0.000 claims description 6
- 238000000605 extraction Methods 0.000 claims description 5
- 230000009286 beneficial effect Effects 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 4
- 230000009467 reduction Effects 0.000 claims description 4
- 238000002790 cross-validation Methods 0.000 claims description 3
- 238000013461 design Methods 0.000 claims description 3
- 230000008014 freezing Effects 0.000 claims description 3
- 238000007710 freezing Methods 0.000 claims description 3
- 238000004519 manufacturing process Methods 0.000 abstract description 6
- 238000012800 visualization Methods 0.000 abstract 1
- 238000010586 diagram Methods 0.000 description 6
- 230000008901 benefit Effects 0.000 description 3
- 238000013136 deep learning model Methods 0.000 description 3
- 230000000007 visual effect Effects 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 238000012952 Resampling Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a semantic segmentation-based motor commutator surface defect detection model. The model is divided into two stages, wherein the first stage is a segmentation network which is used for carrying out pixel-level positioning on the surface defects, providing visualization and providing interpretability for judging a classification network; and the second stage is a classification network, the output of the segmentation network is used as input to assist the classification network in making accurate judgment, the probability value between 0 and 1 is output, and the probability is higher, which represents that the probability of the existence of the defect is higher. Experiments show that the detection model provided by the invention can obtain higher detection precision, less false judgment rate and missing judgment rate aiming at a commutator data set of a small target and a small sample, has good generalization capability, and can meet the requirements of industrial automatic production on the precision, speed and generalization capability of commutator defect detection.
Description
Technical Field
The invention relates to the field of motor commutator surface defect detection, in particular to a semantic segmentation-based motor commutator surface defect detection model.
Technical Field
With the rise of national intelligent manufacturing, the traditional low-end manufacturing mode with low efficiency cannot meet the increasing demands of consumers. As an important component of intelligent manufacturing, the rapidity, accuracy and intelligence of product quality detection determine the development level of intelligent manufacturing.
The motor commutator plays a role in commutation when the direct current motor and the alternating current motor rotate, and is an important part of the armature of the motor. In industrial automation, motors are widely used, and therefore quality inspection of key components of the motors becomes very important. At present, an online visual detection technology based on a traditional machine vision method is available, and the online visual detection technology has high detection precision. However, the traditional machine vision detection method firstly needs to preprocess the image, then manually extract the features, and finally train a classifier to classify the defects. The method needs technicians to spend a large amount of time to extract features by using an image algorithm, has long project period, and the quality of feature extraction directly influences the precision and the robustness, and is far from meeting the requirements of industrial automatic production on the precision, the speed and the generalization capability of defect detection.
Compared with the traditional machine vision method, the deep learning can omit the preprocessing of data, and directly learn the abstract and essential characteristics from the original data, thereby replacing the manual characteristic extraction. In recent years, the method is widely applied to the field of defect detection. However, the existing model for detecting defects by utilizing deep learning cannot well solve the problem of small samples in the industry, the classical deep learning model can obtain higher precision only by a large number of training samples, and in order to avoid secret disclosure and external data disclosure in a factory in the industrial field, data is difficult to obtain, and the small samples are unavoidable problems. Secondly, with the popularization of high-performance image acquisition equipment, the threshold for acquiring high-quality images is reduced, so that the problem of detecting small targets in high-resolution images is caused, namely, defects only account for a small part of the whole high-resolution image, and the problem cannot be well solved in the conventional deep learning model.
Therefore, a motor commutator surface defect detection model based on semantic segmentation is provided to solve the problem.
Disclosure of Invention
The invention provides a motor commutator surface defect detection model based on semantic segmentation, and aims to solve the problems that an existing deep learning model cannot well solve small samples and detect small targets in high-resolution images.
A motor commutator surface defect detection model based on semantic segmentation is characterized by comprising the following steps:
step 2, constructing a segmentation network: inputting training data into a segmentation network, training the segmentation network by using pixel segmentation labels, and outputting a semantic segmentation result;
step 3, constructing a classification network: inputting the result of the segmentation network into a classification network, freezing the weight parameters of the trained segmentation network, and then training the classification network according to the marked labels, wherein the defective sample label is 1, and the non-defective sample label is 0;
and 4, inputting the test data into the trained segmentation network and classification network, and outputting a probability value between 0 and 1, wherein the higher the probability is, the probability that the defect exists is. 0.9 is selected as a threshold value for judging positive and negative samples, and samples greater than or equal to 0.9 are defective samples (positive samples) and samples less than 0.9 are non-defective samples (negative samples).
Further, the semantic segmentation-based motor commutator surface defect detection model is characterized in that the step 2 of constructing a segmentation network comprises the following steps:
step 21, adopting an encoder and a decoder structure in the segmentation network, fusing high-level semantic information and low-level semantic information, and effectively improving the segmentation precision of the small target;
step 22, an encoder in the segmentation network adopts an improved lightweight network MobileNet V3, so that global information is easier to capture, and the overfitting problem caused by small samples is effectively relieved;
in step 23, the decoder in the segmentation network only upsamples 2 times, i.e. the final output is only 1/8 times the resolution of the original image. The original image size is not upsampled to take into account the actual task requirements.
3. The motor commutator surface defect detection model based on semantic segmentation as claimed in claim 1, characterized in that the detection model adopts an improved lightweight network MobileNet V3 for feature extraction, specifically, the original MobileNet V3 bottleneck structure is modified, and a large convolution kernel and a separate convolution are adopted to obtain a more lightweight bottleneck structure, further reduce the number of parameters of the segmented network, and increase the receptive field, and improve the capability of the network to capture small targets;
the improved lightweight bottleneck structural parameters are as follows:
the parameters of the MobileNet V3 bottleneck structure are:
Cinis in the range of 16, 24, 48 and 96.
The method for detecting the surface defects of the motor commutator based on the semantic segmentation as claimed in claim 1, wherein the step 3 of constructing the classification network comprises the following steps:
step 31, splicing the output result (1 channel) of the segmentation network and the feature map (144 channels) before the 1 × 1 convolution is carried out channel reduction to be used as the input of the classification network;
and step 33, respectively performing global maximum pooling and global average pooling on the feature maps (32 channels) output by the classification network to generate 64 output neurons, performing global maximum pooling and global average pooling on the final output map (1 channel) of the segmentation network to obtain 2 output neurons, connecting the 64 output neurons and the 2 output neurons together to serve as input of a full connection layer, finally outputting the probability of 0-1, and judging whether the output neurons are defects according to a set threshold value.
Compared with the prior art, the invention has the following beneficial effects:
1. the detection model provided by the invention has good generalization capability on a small number of defect samples, can well learn the optimal characteristics from the small number of defect samples, and the constructed segmentation network can effectively relieve the over-fitting problem caused by the small samples and solve the problem of insufficient defect samples in the industrial field.
2. The detection model provided by the invention can effectively detect small defect targets in a high-resolution image, can accurately and reliably judge the existence of the surface crack defects of the motor commutator, and can provide visual segmentation positioning in the image.
Drawings
FIG. 1 is a schematic flow chart of a semantic segmentation-based motor commutator surface defect detection model provided by the invention;
FIG. 2 is a schematic overall structure diagram of a semantic segmentation-based motor commutator surface defect detection model provided by the invention;
FIG. 3 is a schematic structural diagram of a segmentation network in a semantic segmentation-based motor commutator surface defect detection model according to the present invention;
FIG. 4 is a schematic diagram of an improved bottleneck structure in a semantic segmentation-based motor commutator surface defect detection model according to the present invention;
fig. 5 is a schematic diagram of a part of data samples in a motor commutator surface defect detection model based on semantic segmentation according to the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the following detailed description of the present invention will be made for clarity and completeness. It should be understood that the described embodiments are only some, not all, and not all embodiments of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments of the present invention without inventive step, are within the scope of the present invention.
Referring to fig. 2, a semantic segmentation-based motor commutator surface defect detection model includes the following specific implementation steps:
referring to fig. 5, a motor commutator surface defect data set used in this example is provided and labeled by the company Kolektor, and pictures are taken under controlled conditions (uniform illumination, etc.). Consists of 50 motor commutators with defects, which are tiny cracks embedded on the plastic surface of the motor commutators. There were a total of 399 images, of which 52 had visible defects and the remaining 347 had no defects.
step 2, constructing a segmentation network: inputting training data into a segmentation network, training the segmentation network by using pixel segmentation labels, and outputting a semantic segmentation result;
constructing a segmented network according to step 2, with reference to fig. 3, comprises the following steps:
step 21, adopting an encoder and a decoder structure in the segmentation network, fusing high-level semantic information and low-level semantic information, and effectively improving the segmentation precision of the small target;
step 22, an encoder in the segmentation network adopts an improved lightweight network MobileNet V3, so that global information is easier to capture, and the overfitting problem caused by small samples is effectively relieved;
more specifically, referring to fig. 4, the detection model uses an improved lightweight network MobileNet V3 for feature extraction, specifically, the original MobileNet V3 bottleneck structure is modified, a large convolution kernel and a separate convolution are used, and the size of the convolution kernel is 3 × 3, 5 × 5, 1 × 7, 7 × 1, 1 × 9, 9 × 1, 1 × 11, 11 × 1. Each convolution kernel generates a deep convolution path, then the feature maps generated by each convolution kernel are spliced together, and finally point convolution is used for reducing the number of channels. The bottleneck structure with lighter weight is obtained, the parameters of the segmented network are further reduced, the receptive field is increased, and the capability of the network for capturing small targets is improved;
the improved lightweight bottleneck structural parameters are as follows:
the parameters of the MobileNet V3 bottleneck structure are:
Cinranges of values of (a) are 16, 24, 48 and 96;
referring to the following table, which describes in detail the structure of a segmented network encoder, block is the basic convolution unit, which can be either a standard convolution or a bottleneck structure. In the table, N represents the number of repetitions, and S is the step size. When the bottleneck block is repeated for multiple times, S is only used for the first bottleneck, and the step length of other bottlenecks is 1. The present invention uses step sizes in deep convolution rather than pooling to increase the field and decrease the feature dimension because the use of step sizes is less computationally expensive than pooling, while having no impact on accuracy. The input of the encoder is a gray scale map, which is down sampled 4 times, the resolution is reduced to 1/2 each time, each convolution is followed by a BN layer and a ReLU6 layer, and each convolution layer changes the data distribution in the network training process. If the data is at the edge of the activation function, the gradient will disappear and the parameters will not be updated. Therefore, a batch normalization layer and a nonlinear activation layer are arranged behind each convolution layer, batch normalization normalizes output after each convolution into zero-mean distribution with unit variance, and aims to cancel order of magnitude difference among dimensional data, so that data distribution is dense, the convergence speed of a model can be accelerated, and the disappearance of generated gradients is prevented. Meanwhile, the ReLU is replaced by the ReLU6, so that the ReLU6 does not cause precision loss in low-precision calculation, and the method is more robust.
The MobileNet V3 uses hard-Swish and hard-Sigmoid to replace the Sigmoid layer in the ReLU6 and SE modules, but only replaces the ReLU6 with h-Swish in the second half of the network, which considers that the advantages can be realized only when the network is used in deeper network layers, so the invention only uses h-Swish in 8 th to 10 th blocks, and uses the ReLU6 in the rest. Similarly, the SE attention mechanism can only play an advantage in a deeper network layer, so that the SE mechanism is only used in 5 th to 10 th blocks according to the MobileNet V3 model architecture;
partitioned network encoder
And step 23, upsampling the output of the 10 th block to the size of the output of the 7 th block by 2 times in the segmentation network decoder, splicing the result with the output characteristic diagram of the 7 th block, and finally performing channel reduction through 1 × 1 convolution to obtain a final single-channel segmentation output diagram. In consideration of the actual task requirements, the decoder only up-samples by a factor of 2, i.e. the final output is only 1/8 times the resolution of the original image and is not up-sampled to the original image size.
Step 3, constructing a classification network: inputting the result of the segmentation network into a classification network, freezing the weight parameters of the trained segmentation network, and then training the classification network according to the marked labels, wherein the defective sample label is 1, and the non-defective sample label is 0;
constructing a classification network according to step 3, comprising the steps of:
step 31, splicing the output result (1 channel) of the segmentation network and the feature map (144 channels) before the 1 × 1 convolution is carried out channel reduction to be used as the input of the classification network;
and step 33, respectively performing global maximum pooling and global average pooling on the feature maps (32 channels) output by the classification network to generate 64 output neurons, performing global maximum pooling and global average pooling on the final output map (1 channel) of the segmentation network to obtain 2 output neurons, connecting the 64 output neurons and the 2 output neurons together to serve as input of a full connection layer, finally outputting the probability of 0-1, and judging whether the output neurons are defects according to a set threshold value.
And 4, inputting the test data into the trained segmentation network and classification network, and outputting a probability value between 0 and 1, wherein the higher the probability is, the probability that the defect exists is. 0.9 is selected as the threshold for determining positive and negative samples. The samples with defects (positive samples) of 0.9 or more and the samples with defects (negative samples) of 0.9 or less are used.
The method selects F-Measure (F1), FN (number of missed judgment), FP (number of false judgment), precision and recall as the evaluation indexes of the model. The result of the model is that F-Measure is 0.981, precision is 0.981, recall is 0.981, FN is 1, FP is 1, so that the model provided by the invention has high detection precision and less false and missed judgment numbers.
The model is realized by using a pyrrch framework, Adam is used for training in classification and segmentation tasks, loss functions are all cross entropy losses, learning rates are all 0.01, 100 epochs are trained, only one image is used in each iteration in consideration of high image resolution and GPU memory limitation, and the batch processing size is set to be 1. In the training process, the model adopts a resampling strategy to alternately train defective samples and non-defective samples so as to ensure that the network observes the same number of defective images and non-defective images and prevent poor training effect of the model due to imbalance of positive samples and negative samples.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art should be considered to be within the technical scope of the present invention, and the technical solutions and the inventive concepts thereof according to the present invention should be equivalent or changed within the scope of the present invention.
Claims (4)
1. A motor commutator surface defect detection model based on semantic segmentation is characterized by comprising the following steps:
step 1, data preprocessing: dividing the data set into three parts, optionally selecting two parts as a training set, taking the other part as a test set, repeating the steps for three times, and executing triple cross validation;
step 2, constructing a segmentation network: inputting training data into a segmentation network, training the segmentation network by using pixel segmentation labels, and outputting a semantic segmentation result;
step 3, constructing a classification network: inputting the result of the segmentation network into a classification network, freezing the weight parameters of the trained segmentation network, and then training the classification network according to the marked labels, wherein the defective sample label is 1, and the non-defective sample label is 0;
and 4, inputting the test data into the trained segmentation network and classification network, and outputting a probability value between 0 and 1, wherein the higher the probability is, the probability that the defect exists is. 0.9 is selected as a threshold value for judging positive and negative samples, and samples greater than or equal to 0.9 are defective samples (positive samples) and samples less than 0.9 are non-defective samples (negative samples).
2. The semantic segmentation based surface defect detection model for the motor commutator of claim 1, wherein the step 2 of constructing the segmentation network comprises the following steps:
step 21, adopting an encoder and a decoder structure in the segmentation network, fusing high-level semantic information and low-level semantic information, and effectively improving the segmentation precision of the small target;
step 22, an encoder in the segmentation network adopts an improved lightweight network MobileNet V3, so that global information is easier to capture, and the overfitting problem caused by small samples is effectively relieved;
in step 23, the decoder in the segmentation network only upsamples 2 times, i.e. the final output is only 1/8 times the resolution of the original image. The original image size is not upsampled to take into account the actual task requirements.
3. The motor commutator surface defect detection model based on semantic segmentation as claimed in claim 1, characterized in that the detection model adopts an improved lightweight network MobileNet V3 for feature extraction, specifically, the original MobileNet V3 bottleneck structure is modified, and a large convolution kernel and a separate convolution are adopted to obtain a more lightweight bottleneck structure, further reduce the number of parameters of the segmented network, and increase the receptive field, and improve the capability of the network to capture small targets;
the improved lightweight bottleneck structural parameters are as follows:
the parameters of the MobileNet V3 bottleneck structure are:
Cinis in the range of 16, 24, 48 and 96.
4. The method for detecting the surface defects of the motor commutator based on the semantic segmentation as claimed in claim 1, wherein the step 3 of constructing the classification network comprises the following steps:
step 31, splicing the output result (1 channel) of the segmentation network and the feature map (144 channels) before the 1 × 1 convolution is carried out channel reduction to be used as the input of the classification network;
step 32, performing 5 × 5 convolution twice and two downsampling operations on the input, wherein the convolution operation adopts depth separable convolution, and the downsampling operation adopts step length of 2 to perform downsampling; therefore, the network can fully utilize beneficial characteristics, simplify the structural design of the classification network and improve the prediction performance of the classification network;
and step 33, respectively performing global maximum pooling and global average pooling on the feature maps (32 channels) output by the classification network to generate 64 output neurons, performing global maximum pooling and global average pooling on the final output map (1 channel) of the segmentation network to obtain 2 output neurons, connecting the 64 output neurons and the 2 output neurons together to serve as input of a full connection layer, finally outputting the probability of 0-1, and judging whether the output neurons are defects according to a set threshold value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011269251.1A CN112365478A (en) | 2020-11-13 | 2020-11-13 | Motor commutator surface defect detection model based on semantic segmentation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011269251.1A CN112365478A (en) | 2020-11-13 | 2020-11-13 | Motor commutator surface defect detection model based on semantic segmentation |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112365478A true CN112365478A (en) | 2021-02-12 |
Family
ID=74514711
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011269251.1A Pending CN112365478A (en) | 2020-11-13 | 2020-11-13 | Motor commutator surface defect detection model based on semantic segmentation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112365478A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113112456A (en) * | 2021-03-25 | 2021-07-13 | 湖南工业大学 | Thick food filling finished product defect detection method based on target detection algorithm |
| CN113240671A (en) * | 2021-06-16 | 2021-08-10 | 重庆科技学院 | Water turbine runner blade defect detection method based on YoloV4-Lite network |
| CN113850791A (en) * | 2021-09-28 | 2021-12-28 | 哈尔滨工业大学 | Bathroom ceramic defect detection method based on two-stage MobileNet |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019104767A1 (en) * | 2017-11-28 | 2019-06-06 | 河海大学常州校区 | Fabric defect detection method based on deep convolutional neural network and visual saliency |
| CN110910368A (en) * | 2019-11-20 | 2020-03-24 | 佛山市南海区广工大数控装备协同创新研究院 | Injector defect detection method based on semantic segmentation |
| CN110992317A (en) * | 2019-11-19 | 2020-04-10 | 佛山市南海区广工大数控装备协同创新研究院 | PCB defect detection method based on semantic segmentation |
| CN111402203A (en) * | 2020-02-24 | 2020-07-10 | 杭州电子科技大学 | Fabric surface defect detection method based on convolutional neural network |
| CN111768388A (en) * | 2020-07-01 | 2020-10-13 | 哈尔滨工业大学（深圳） | Product surface defect detection method and system based on positive sample reference |
-
2020
- 2020-11-13 CN CN202011269251.1A patent/CN112365478A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019104767A1 (en) * | 2017-11-28 | 2019-06-06 | 河海大学常州校区 | Fabric defect detection method based on deep convolutional neural network and visual saliency |
| CN110992317A (en) * | 2019-11-19 | 2020-04-10 | 佛山市南海区广工大数控装备协同创新研究院 | PCB defect detection method based on semantic segmentation |
| CN110910368A (en) * | 2019-11-20 | 2020-03-24 | 佛山市南海区广工大数控装备协同创新研究院 | Injector defect detection method based on semantic segmentation |
| CN111402203A (en) * | 2020-02-24 | 2020-07-10 | 杭州电子科技大学 | Fabric surface defect detection method based on convolutional neural network |
| CN111768388A (en) * | 2020-07-01 | 2020-10-13 | 哈尔滨工业大学（深圳） | Product surface defect detection method and system based on positive sample reference |
Non-Patent Citations (1)
| Title |
|---|
| 柳锋;李俊峰;戴文战;: "基于深度学习语义分割的导光板缺陷检测方法", 计算机系统应用, no. 06 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113112456A (en) * | 2021-03-25 | 2021-07-13 | 湖南工业大学 | Thick food filling finished product defect detection method based on target detection algorithm |
| CN113112456B (en) * | 2021-03-25 | 2022-05-13 | 湖南工业大学 | Thick food filling finished product defect detection method based on target detection algorithm |
| CN113240671A (en) * | 2021-06-16 | 2021-08-10 | 重庆科技学院 | Water turbine runner blade defect detection method based on YoloV4-Lite network |
| CN113240671B (en) * | 2021-06-16 | 2022-06-24 | 重庆科技学院 | Water turbine runner blade defect detection method based on YoloV4-Lite network |
| CN113850791A (en) * | 2021-09-28 | 2021-12-28 | 哈尔滨工业大学 | Bathroom ceramic defect detection method based on two-stage MobileNet |
| CN113850791B (en) * | 2021-09-28 | 2022-07-05 | 哈尔滨工业大学 | Bathroom ceramic defect detection method based on two-stage MobileNet |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109829891B (en) | Magnetic shoe surface defect detection method based on dense generation of antagonistic neural network | |
| CN112365478A (en) | Motor commutator surface defect detection model based on semantic segmentation | |
| CN110570396A (en) | industrial product defect detection method based on deep learning | |
| CN114627383B (en) | Small sample defect detection method based on metric learning | |
| CN114841972B (en) | Transmission line defect identification method based on saliency map and semantic embedded feature pyramid | |
| CN113436169B (en) | Industrial equipment surface crack detection method and system based on semi-supervised semantic segmentation | |
| CN111768388B (en) | Product surface defect detection method and system based on positive sample reference | |
| CN112465790A (en) | Surface defect detection method based on multi-scale convolution and trilinear global attention | |
| CN113888547A (en) | Non-supervision domain self-adaptive remote sensing road semantic segmentation method based on GAN network | |
| CN110020658B (en) | Salient object detection method based on multitask deep learning | |
| CN112967271B (en) | Casting surface defect identification method based on improved DeepLabv3+ network model | |
| CN110009622B (en) | Display panel appearance defect detection network and defect detection method thereof | |
| CN113139544A (en) | Saliency target detection method based on multi-scale feature dynamic fusion | |
| CN115731400A (en) | X-ray image foreign matter detection method based on self-supervision learning | |
| CN117252815A (en) | Industrial part defect detection method, system, equipment and storage medium based on 2D-3D multi-mode image | |
| CN114037684B (en) | Defect detection method based on yolov and attention mechanism model | |
| CN114331869A (en) | Dam face crack semantic segmentation method | |
| CN114758125B (en) | Gear surface defect detection method and system based on deep learning | |
| CN111210417A (en) | Cloth defect detection method based on convolutional neural network | |
| CN111612803B (en) | Vehicle image semantic segmentation method based on image definition | |
| CN117079237A (en) | Self-supervision monocular vehicle distance detection method | |
| CN116862883A (en) | Concrete slump detection method based on image semantic segmentation | |
| CN112418229A (en) | Unmanned ship marine scene image real-time segmentation method based on deep learning | |
| CN116596866A (en) | Defect detection method based on high-resolution image and storage medium | |
| CN115100409B (en) | Video portrait segmentation algorithm based on twin network |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20210212 |
|
| WD01 | Invention patent application deemed withdrawn after publication |