CN112990391A - Feature fusion based defect classification and identification system of convolutional neural network - Google Patents
Feature fusion based defect classification and identification system of convolutional neural network Download PDFInfo
- Publication number
- CN112990391A CN112990391A CN202110548602.0A CN202110548602A CN112990391A CN 112990391 A CN112990391 A CN 112990391A CN 202110548602 A CN202110548602 A CN 202110548602A CN 112990391 A CN112990391 A CN 112990391A
- Authority
- CN
- China
- Prior art keywords
- module
- convolutional
- feature
- convolutional layer
- attention
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- 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
Abstract
The invention discloses a defect classification and identification system of a convolutional neural network based on feature fusion, which comprises the following steps: the floor sample acquisition module is used for acquiring 3 pieces of image information of the new material floor at the same position and different angles; the defect identification module is used for identifying whether the new material floor has defects according to the acquired 3 pieces of image information, the defect identification module comprises a feature extraction module, a feature fusion module, a feature flattening module and a feature decision module, the feature extraction module is realized by adopting a 3-path parallel ResNet-34 network model, the invention provides a 3-branch feature fusion convolutional neural network model, and the network model has good network performance under the condition of considering precision, parameter quantity and memory occupation, takes ResNet-34 as a branch network basic framework, takes merging operation as a feature fusion mode and is embedded into a TFFCNN-CBAM attention module; compared with other traditional classification convolutional neural networks, the accuracy is improved by 1.28% -2.86%.
Description
Technical Field
The invention relates to the field of new material floor detection, in particular to a defect classification and identification system of a convolutional neural network based on feature fusion.
Background
Common defects of new material floors include impurities, scratches, broken buttons, broken corners, bubbles, and stains. The difficulty of identifying and detecting the defects of the new material floor is high, and the main reasons are as follows: (1) the method has the advantages that the types of surface textures are more, the trend of the textures is complex, partial textures are similar to defects, the trends of the defects are very similar to those of a color film, even partial defects are difficult to distinguish by human vision, the types of background textures of the floor are more, the characteristics are chaotic and irregular, the background information required to be learned by a network model is increased, and the training difficulty of a defect detection model is improved; (2) the glossiness of the surface film is high, the image quality is easily affected by light, the problems that the bubble defects are required to have single reflection characteristic or single light source direction and the like are solved, the top layer of the floor hierarchical structure is a UV (ultraviolet) coating layer, and the glossiness of the paint surface is high, so that the specific defect characteristics can not be seen from visual observation at an angle, and the difficulty of network training is improved. (3) The defects of different types have specific shapes, and the defects of the same type have various sizes and are different in size; due to the influences of foreign matters, more temporary storage and transportation times, more process flow steps and the like in the floor production process, the appearance shapes of the generated defects are various, and the processing difficulty of the sample data set is higher.
Disclosure of Invention
The invention aims to provide a defect classification and identification system of a convolutional neural network based on feature fusion, which provides a 3-branch feature fusion convolutional neural network model, and the network model has extremely high accuracy in identifying defects of a new material floor.
The defect classification and identification system of the convolutional neural network based on feature fusion comprises the following steps: the floor sample acquisition module is used for acquiring 3 pieces of image information of the new material floor at the same position and different angles;
the defect identification module is used for identifying whether the new material floor has defects according to the acquired 3 pieces of image information, and comprises a feature extraction module, a feature fusion module, a feature flattening module and a feature decision module; the characteristic extraction module is used for extracting defect characteristics from 3 pieces of image information and outputting a characteristic diagram, and is realized by adopting a 3-path parallel ResNet-34 network model; the ResNet-34 network model comprises an input layer and 5 convolutional layers, wherein the convolutional layers comprise a first convolutional layer, a maximum pooling layer, a second convolutional layer, a third convolutional layer, a fourth convolutional layer and a fifth convolutional layer which are sequentially arranged, the first convolutional layer, the second convolutional layer, the third convolutional layer, the fourth convolutional layer and the fifth convolutional layer comprise 33 convolutional blocks, and an attention mechanism block is connected behind each convolutional block; the characteristic fusion module is used for merging and fusing the defect characteristics of the characteristic graphs output by the 3-path parallel ResNet-34 network model, the characteristic flattening module is used for adjusting the dimensionality of the characteristic graphs merged and fused by the characteristic fusion module into one dimension, and the characteristic decision module is used for carrying out decision analysis on the one-dimensional characteristic graphs output by the characteristic flattening module and outputting the result whether the new material floor has defects or not.
Further, the first convolutional layer comprises 1 convolutional block with convolution kernel of 7 × 7, the second convolutional layer comprises 6 convolutional blocks with convolution kernel of 3 × 3, the third convolutional layer comprises 8 convolutional blocks with convolution kernel of 3 × 3, the fourth convolutional layer comprises 12 convolutional blocks with convolution kernel of 3 × 3, and the fifth convolutional layer comprises 6 convolutional blocks with convolution kernel of 3 × 3.
Further, the attention mechanism block comprises a channel attention module and a space attention module, wherein the channel attention module respectively performs maximum pooling operation and average pooling operation on a middle feature map output by the convolution block along the axial direction of the space of the middle feature map, a vector generated by the maximum pooling operation and a vector generated by the average pooling operation are transmitted backwards to enter a single-layer multilayer perceptron, the multilayer perceptron generates a maximum pooled attention vector and an average pooled attention vector, the maximum pooled attention vector and the average pooled attention vector are combined and added according to corresponding positions of elements, and finally the output feature vectors are input into a Sigmoid activation function to obtain channel attention vectors;
the spatial attention module carries out maximum pooling operation and average pooling operation on an output feature map of the channel attention module along the axial direction of a channel of the channel attention module, combines an attention moment array generated by the maximum pooling operation and an attention matrix generated by the average pooling operation to generate a dual-channel feature matrix, applies convolution operation to generate a single-channel matrix, and finally inputs a result into a Sigmoid activation function in the same way and carries out dot multiplication on the result and the feature map according to the corresponding position of a pixel.
The invention provides a 3-branch characteristic fusion convolutional neural network model, which has good network performance under the condition of considering precision, parameter quantity and memory occupation, takes ResNet-34 as a branch network infrastructure, takes merging (coordination) operation as a characteristic fusion mode and embeds a TFFCNN-CBAM network model of a CBAM attention module; compared with other traditional classification convolutional neural networks, the accuracy is improved by 1.28% -2.86%.
Drawings
FIG. 1 is a schematic structural diagram of the present invention.
FIG. 2 is a flow chart of a defect identifying module of the present invention.
Detailed Description
As shown in fig. 1, the system for identifying and classifying defects of a convolutional neural network based on feature fusion provided in this embodiment is configured to detect defects of a new material floor, and includes a new material floor sample obtaining module and a defect identifying module; the new material floor sample acquisition module is used for acquiring 3 pieces of image information of the new material floor at the same position and different angles, and comprises three cameras with MCU control, wherein the three cameras are connected with a computer, the cameras are arranged above a conveying device of the new material floor and can acquire the position of the image information of the new material floor, the three cameras have different shooting angles, the acquired image information is transmitted to the computer through a network or a cable after shooting, and the acquired image information is preprocessed, wherein the preprocessing is to divide the image information into input image information with 100 x 100 pixels.
The defect identification module is used for identifying whether the new material floor has defects according to the acquired 3 pieces of image information, and as shown in fig. 2, the defect identification module comprises a feature extraction module, a feature fusion module, a feature flattening module and a feature decision module.
The feature extraction module is implemented by adopting a 3-path parallel ResNet-34 network model, 3 pictures acquired by the floor sample acquisition module are respectively input into the ResNet-34 network model and corresponding feature maps are output, the ResNet-34 network model comprises an input layer and 5 convolutional layers, a network framework is shown in table 1, the input layer is used for inputting 3 pieces of image information acquired by the floor sample acquisition module, the convolutional layers comprise a first convolutional layer, a maximum pooling layer, a second convolutional layer, a third convolutional layer, a fourth convolutional layer and a fifth convolutional layer which are sequentially arranged, the first convolutional layer comprises 1 convolutional block with a convolutional core of 7 × 7, the second convolutional layer comprises 6 convolutional cores of 3 × 3 convolutional blocks, the third convolutional layer comprises 8 convolutional cores of 3 × 3 convolutional blocks, the fourth convolutional layer comprises 12 convolutional cores of 3 × 3 convolutional blocks, and the fifth convolutional layer comprises 6 convolutional blocks with a convolutional core of 3 × 3, the first, second, third, fourth, and fifth convolutional layers include 33 convolutional blocks.
TABLE 1 ResNet-34 network module architecture
An attention machine making block is connected behind each rolling block; the introduction of the attention mechanism block helps the ResNet-34 network module to learn different feature information more intensely. The attention mechanism block comprises a channel attention module and a space attention module, wherein the channel attention module respectively performs maximum pooling operation and average pooling operation on a middle feature map output by the convolution block along the axial direction of the space of the convolution block, a vector generated by the maximum pooling operation and a vector generated by the average pooling operation are transmitted backwards to enter a single-layer multilayer perceptron, the multilayer perceptron generates a maximum pooling attention vector and an average pooling attention vector, the maximum pooling attention vector and the average pooling attention vector are combined and added according to corresponding positions of elements, and finally the output feature vector is input into a Sigmoid activation function to obtain a channel attention vector.
The spatial attention module carries out maximum pooling operation and average pooling operation on an output feature map of the channel attention module along the axial direction of a channel of the channel attention module, combines an attention moment array generated by the maximum pooling operation and an attention matrix generated by the average pooling operation to generate a dual-channel feature matrix, applies convolution operation to generate a single-channel matrix, and finally inputs a result into a Sigmoid activation function in the same way and carries out dot multiplication on the result and the feature map according to the corresponding position of a pixel.
The feature fusion module is used for merging and fusing the feature graphs output by the 3-path parallel ResNet-34 network model, wherein the merging and fusing refers to superposing the feature graphs along the channel direction.
And the feature flattening module is used for adjusting the dimension of the feature graph output by the feature fusion module into one dimension.
The characteristic decision module is used for carrying out decision analysis on the one-dimensional characteristic diagram output by the characteristic flattening module to output a result, and the result is whether the new material floor has defects or not. The method indexes and normalizes the numerical value of the output vector, ensures the nonnegativity of the probability and simultaneously ensures the sum of the probability to be 1.
The defect classification and identification system based on the convolutional neural network with feature fusion provided by the embodiment is applied to a new material floor manufacturer to obtain 1959 floor samples, wherein 508 floor samples are normal samples, and 1451 floor samples are defect samples. Because the size of the image pixel after image scaling processing is 900 × 900, and one image data may contain a plurality of different types of defects or a plurality of same types of defects, when the area in the floor width direction is obtained by jointly shooting with three industrial cameras in the image acquisition system, each image data is divided into a plurality of small sample data with the pixel size of 100 × 100, the sample data is divided into a training set and a test set according to the ratio of 8:2, and table 2 shows the division result of the data set.
TABLE 2 sample data set partitioning results
To verify the accuracy of this embodiment, the network model of the feature extraction module is respectively changed to VGG-16, ResNet-18, ResNet-34 and their combinations, training and testing are performed according to the data partitioning results, and the accuracy is analyzed for the test results and the actual results, the results of which are shown in table 3. To represent the accuracy of the training result, the parameters of the network training are set as follows: the GPU display card is NVIDIA GeForce RTX 2080 Ti; the total number of iterations is 100; using a parameter optimizer SGD, weight attenuation 1 e-4; the batch size of the training data is 32 per time; the initial learning rate is 0.1, after which the learning rate decays to 1/10 for the original data for 30 iterations; the data set was initially normalized before each training.
TABLE 3 network architecture comparison experimental data of different branches
The defect identification modules of the ResNet-34 of the 3 branch networks have higher accuracy rate than the defect identification modules of other network architectures, and the accuracy rate is 99.12%; the network depth (33 convolutional layers) and the parameter number (65.42M) of the basic network architecture of ResNet-34 are larger than those of other network architectures, and the feature extraction capability and the model expression capability are also strongest.
Further, an attention mechanism block CBAM is arranged after the convolution blocks of the defect identification modules of the ResNet-34 in all the 3 branch networks, the defect identification modules of the ResNet-34 in all the 3 branch networks are trained and tested by adopting the sample data, the accuracy is analyzed by the test result and the actual result, and the result is shown in the table 4.
Table 4 network architecture comparison experimental data of different branches
The 3 branch networks are ResNet-34 and the defect identification module with the attention mechanism block has higher accuracy than the defect identification module with the 3 branch networks being ResNet-34, and the accuracy can reach as high as 99.23%.
The system provides a 3-branch feature fusion convolutional neural network model, and the TFFCNN-CBAM network model which takes ResNet-34 as a branch network infrastructure, a merging (merging) operation as a feature fusion mode, Conv5_ x level as a feature fusion position and is embedded into a CBAM attention module has good network performance under the condition of considering precision, parameter quantity and memory occupation. Compared with other traditional classification convolutional neural networks, the accuracy is improved by 1.28% -2.86%, and the effectiveness of the embodiment is proved.
The above description is only a preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any modification and replacement based on the technical solution and inventive concept provided by the present invention should be covered within the scope of the present invention.
Claims (3)
1. The defect classification and identification system of the convolutional neural network based on feature fusion is characterized by comprising the following steps: the floor sample acquisition module is used for acquiring 3 pieces of image information of the new material floor at the same position and different angles;
the defect identification module is used for identifying whether the new material floor has defects according to the acquired 3 pieces of image information, and comprises a feature extraction module, a feature fusion module, a feature flattening module and a feature decision module; the characteristic extraction module is used for extracting defect characteristics from 3 pieces of image information and outputting a characteristic diagram, and is realized by adopting a 3-path parallel ResNet-34 network model; the ResNet-34 network model comprises an input layer and 5 convolutional layers, wherein the convolutional layers comprise a first convolutional layer, a maximum pooling layer, a second convolutional layer, a third convolutional layer, a fourth convolutional layer and a fifth convolutional layer which are sequentially arranged, the first convolutional layer, the second convolutional layer, the third convolutional layer, the fourth convolutional layer and the fifth convolutional layer comprise 33 convolutional blocks, and an attention mechanism block is connected behind each convolutional block; the characteristic fusion module is used for merging and fusing the defect characteristics of the characteristic graphs output by the 3-path parallel ResNet-34 network model, the characteristic flattening module is used for adjusting the dimensionality of the characteristic graphs merged and fused by the characteristic fusion module into one dimension, and the characteristic decision module is used for carrying out decision analysis on the one-dimensional characteristic graphs output by the characteristic flattening module and outputting the result whether the new material floor has defects or not.
2. The system of claim 1, wherein the feature fusion based convolutional neural network defect classification recognition system comprises: the first convolutional layer comprises 1 convolutional block with 7 × 7 convolutional cores, the second convolutional layer comprises 6 convolutional blocks with 3 × 3 convolutional cores, the third convolutional layer comprises 8 convolutional blocks with 3 × 3 convolutional cores, the fourth convolutional layer comprises 12 convolutional blocks with 3 × 3 convolutional cores, and the fifth convolutional layer comprises 6 convolutional blocks with 3 × 3 convolutional cores.
3. The system for identifying defect classification of convolutional neural network based on feature fusion according to claim 1 or 2, wherein: the attention mechanism block comprises a channel attention module and a space attention module, wherein the channel attention module respectively performs maximum pooling operation and average pooling operation on a middle feature map output by the convolution block along the axial direction of the space of the convolution block, and transmits a vector generated by the maximum pooling operation and a vector generated by the average pooling operation backwards into a single-layer multilayer perceptron;
the spatial attention module carries out maximum pooling operation and average pooling operation on an output feature map of the channel attention module along the axial direction of a channel of the channel attention module, combines an attention moment array generated by the maximum pooling operation and an attention matrix generated by the average pooling operation to generate a dual-channel feature matrix, applies convolution operation to generate a single-channel matrix, and finally inputs a result into a Sigmoid activation function in the same way and carries out dot multiplication on the result and the feature map according to the corresponding position of a pixel.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110548602.0A CN112990391A (en) | 2021-05-20 | 2021-05-20 | Feature fusion based defect classification and identification system of convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110548602.0A CN112990391A (en) | 2021-05-20 | 2021-05-20 | Feature fusion based defect classification and identification system of convolutional neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112990391A true CN112990391A (en) | 2021-06-18 |
Family
ID=76337033
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110548602.0A Pending CN112990391A (en) | 2021-05-20 | 2021-05-20 | Feature fusion based defect classification and identification system of convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112990391A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113516661A (en) * | 2021-09-15 | 2021-10-19 | 深圳新视智科技术有限公司 | Defect detection method and device based on feature fusion |
| CN113570003A (en) * | 2021-09-23 | 2021-10-29 | 深圳新视智科技术有限公司 | Feature fusion defect detection method and device based on attention mechanism |
| CN113850242A (en) * | 2021-11-30 | 2021-12-28 | 北京中超伟业信息安全技术股份有限公司 | Storage abnormal target detection method and system based on deep learning algorithm |
| CN114067153A (en) * | 2021-11-02 | 2022-02-18 | 暨南大学 | Image classification method and system based on parallel double-attention light-weight residual error network |
| CN114387467A (en) * | 2021-12-09 | 2022-04-22 | 哈工大（张家口）工业技术研究院 | Medical image classification method based on multi-module convolution feature fusion |
| CN114418003A (en) * | 2022-01-20 | 2022-04-29 | 北京科技大学 | Double-image identification and classification method based on attention mechanism and multi-size information extraction |
| CN116258947A (en) * | 2023-03-07 | 2023-06-13 | 浙江研几网络科技股份有限公司 | Industrial automatic processing method and system suitable for home customization industry |
| EP4306943A1 (en) * | 2022-07-13 | 2024-01-17 | wiiDetect | Method for determining a state of at least one floor element, associated system and assembly |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN203705352U (en) * | 2014-01-23 | 2014-07-09 | 四川大学 | Magnetic shoe online detecting device based on machine vision |
| CN110223275A (en) * | 2019-05-28 | 2019-09-10 | 陕西师范大学 | A kind of cerebral white matter fiber depth clustering method of task-fMRI guidance |
| CN110826588A (en) * | 2019-08-29 | 2020-02-21 | 天津大学 | Drainage pipeline defect detection method based on attention mechanism |
| CN111325713A (en) * | 2020-01-21 | 2020-06-23 | 浙江省北大信息技术高等研究院 | Wood defect detection method, system and storage medium based on neural network |
| CN111339827A (en) * | 2020-01-18 | 2020-06-26 | 中国海洋大学 | SAR image change detection method based on multi-region convolutional neural network |
| CN112450881A (en) * | 2020-11-12 | 2021-03-09 | 武汉大学 | Multi-modal sleep staging method based on time sequence relevance driving |
| CN112819804A (en) * | 2021-02-23 | 2021-05-18 | 西北工业大学 | Insulator defect detection method based on improved YOLOv5 convolutional neural network |
-
2021
- 2021-05-20 CN CN202110548602.0A patent/CN112990391A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN203705352U (en) * | 2014-01-23 | 2014-07-09 | 四川大学 | Magnetic shoe online detecting device based on machine vision |
| CN110223275A (en) * | 2019-05-28 | 2019-09-10 | 陕西师范大学 | A kind of cerebral white matter fiber depth clustering method of task-fMRI guidance |
| CN110826588A (en) * | 2019-08-29 | 2020-02-21 | 天津大学 | Drainage pipeline defect detection method based on attention mechanism |
| CN111339827A (en) * | 2020-01-18 | 2020-06-26 | 中国海洋大学 | SAR image change detection method based on multi-region convolutional neural network |
| CN111325713A (en) * | 2020-01-21 | 2020-06-23 | 浙江省北大信息技术高等研究院 | Wood defect detection method, system and storage medium based on neural network |
| CN112450881A (en) * | 2020-11-12 | 2021-03-09 | 武汉大学 | Multi-modal sleep staging method based on time sequence relevance driving |
| CN112819804A (en) * | 2021-02-23 | 2021-05-18 | 西北工业大学 | Insulator defect detection method based on improved YOLOv5 convolutional neural network |
Non-Patent Citations (3)
| Title |
|---|
| ADITYA M. DESHPANDE等: "One-Shot Recognition of Manufacturing Defects in Steel Surfaces", 《NAMRC 48》 * |
| LUOFENG XIE等: ""FFCNN: A Deep Neural Network for Surface Defect Detection of Magnetic Tile", 《IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS》 * |
| 殷鹰等: "基于深度学习的磁瓦内部缺陷声振检测方法", 《中国测试》 * |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113516661B (en) * | 2021-09-15 | 2022-02-08 | 深圳新视智科技术有限公司 | Defect detection method and device based on feature fusion |
| CN113516661A (en) * | 2021-09-15 | 2021-10-19 | 深圳新视智科技术有限公司 | Defect detection method and device based on feature fusion |
| CN113570003A (en) * | 2021-09-23 | 2021-10-29 | 深圳新视智科技术有限公司 | Feature fusion defect detection method and device based on attention mechanism |
| CN114067153A (en) * | 2021-11-02 | 2022-02-18 | 暨南大学 | Image classification method and system based on parallel double-attention light-weight residual error network |
| CN114067153B (en) * | 2021-11-02 | 2022-07-12 | 暨南大学 | Image classification method and system based on parallel double-attention light-weight residual error network |
| CN113850242A (en) * | 2021-11-30 | 2021-12-28 | 北京中超伟业信息安全技术股份有限公司 | Storage abnormal target detection method and system based on deep learning algorithm |
| CN114387467B (en) * | 2021-12-09 | 2022-07-29 | 哈工大（张家口）工业技术研究院 | Medical image classification method based on multi-module convolution feature fusion |
| CN114387467A (en) * | 2021-12-09 | 2022-04-22 | 哈工大（张家口）工业技术研究院 | Medical image classification method based on multi-module convolution feature fusion |
| CN114418003A (en) * | 2022-01-20 | 2022-04-29 | 北京科技大学 | Double-image identification and classification method based on attention mechanism and multi-size information extraction |
| CN114418003B (en) * | 2022-01-20 | 2022-09-16 | 北京科技大学 | Double-image recognition and classification method based on attention mechanism and multi-size information extraction |
| EP4306943A1 (en) * | 2022-07-13 | 2024-01-17 | wiiDetect | Method for determining a state of at least one floor element, associated system and assembly |
| FR3137965A1 (en) * | 2022-07-13 | 2024-01-19 | WiiCare | Method for determining a state of at least one ground element, associated system and assembly |
| CN116258947A (en) * | 2023-03-07 | 2023-06-13 | 浙江研几网络科技股份有限公司 | Industrial automatic processing method and system suitable for home customization industry |
| CN116258947B (en) * | 2023-03-07 | 2023-08-18 | 浙江研几网络科技股份有限公司 | Industrial automatic processing method and system suitable for home customization industry |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112990391A (en) | Feature fusion based defect classification and identification system of convolutional neural network | |
| CN111179251B (en) | Defect detection system and method based on twin neural network and by utilizing template comparison | |
| CN111598861B (en) | Improved Faster R-CNN model-based non-uniform texture small defect detection method | |
| US10692209B2 (en) | Image processing apparatus, image processing method, and computer-readable non-transitory recording medium storing image processing program | |
| Mery | Aluminum casting inspection using deep learning: a method based on convolutional neural networks | |
| CN111462120B (en) | Defect detection method, device, medium and equipment based on semantic segmentation model | |
| CN111223093A (en) | AOI defect detection method | |
| CN111257341B (en) | Underwater building crack detection method based on multi-scale features and stacked full convolution network | |
| CN110060237A (en) | A kind of fault detection method, device, equipment and system | |
| CN113643268B (en) | Industrial product defect quality inspection method and device based on deep learning and storage medium | |
| CN104200478B (en) | Low-resolution touch screen image defect detection method based on sparse representation | |
| CN109064459A (en) | A kind of Fabric Defect detection method based on deep learning | |
| CN106355579A (en) | Defect detecting method of cigarette carton surface wrinkles | |
| CN112990392A (en) | New material floor defect target detection system based on improved YOLOv5 algorithm | |
| KR102268909B1 (en) | Inspection method based on edge field and deep learning | |
| CN114463759A (en) | Lightweight character detection method and device based on anchor-frame-free algorithm | |
| CN111127454A (en) | Method and system for generating industrial defect sample based on deep learning | |
| CN113240647A (en) | Mobile phone shell rear cover defect detection method and system based on deep learning | |
| CN112819748A (en) | Training method and device for strip steel surface defect recognition model | |
| CN111145145A (en) | Image surface defect detection method based on MobileNet | |
| CN115457026A (en) | Paper defect detection method based on improved YOLOv5 | |
| CN110633739B (en) | Polarizer defect image real-time classification method based on parallel module deep learning | |
| CN116071315A (en) | Product visual defect detection method and system based on machine vision | |
| CN117392042A (en) | Defect detection method, defect detection apparatus, and storage medium | |
| CN117474863A (en) | Chip surface defect detection method for compressed multi-head self-attention neural 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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210618 |
|
| RJ01 | Rejection of invention patent application after publication |