CN109509187B - Efficient inspection algorithm for small defects in large-resolution cloth images - Google Patents
Efficient inspection algorithm for small defects in large-resolution cloth images Download PDFInfo
- Publication number
- CN109509187B CN109509187B CN201811336000.3A CN201811336000A CN109509187B CN 109509187 B CN109509187 B CN 109509187B CN 201811336000 A CN201811336000 A CN 201811336000A CN 109509187 B CN109509187 B CN 109509187B
- Authority
- CN
- China
- Prior art keywords
- image
- feature
- model
- training
- test
- 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
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
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
-
- 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
- G06T2207/10004—Still image; Photographic image
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- 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
- G06T2207/30108—Industrial image inspection
- G06T2207/30124—Fabrics; Textile; Paper
Abstract
The invention relates to an efficient inspection algorithm for small flaws in a large-resolution cloth image, which comprises the following steps: (1) Acquiring an image through a camera, and then labeling the image by using a labelImg tool; (2) Splitting the processed image into a training set and a test set, wherein the training set is used for training the test model, and the test set is used for evaluating the performance of the test model; (3) Inputting the training set images, the corresponding category information position information and the like into an improved se-next 101 inspection model, and training the inspection model; (4) And processing the images in the test set by using the trained inspection model, and acquiring the approximate positions and corresponding categories of the flaws. The method can realize the processing of the multi-scale characteristic image for the input image with single resolution, thereby processing the multi-scale image block, being suitable for various flaws with different sizes and greatly improving the detection precision and speed; meanwhile, the algorithm realizes the acquisition of the approximate positions of the defects on the image classification frame and the processing of the situations of various defects in the image.
Description
Technical Field
The invention relates to the field of image classification, in particular to an efficient inspection algorithm for small flaws in a large-resolution cloth image.
Background
Early image classification was mainly through traditional machine learning methods, generally divided into two parts: feature extraction based methods and template matching based methods, while feature extraction based methods mainly include statistical based methods, spectral based methods, texture model based methods, learning based methods and structure based methods. These methods all require artificial selection of features and are not very extensive.
With the development of deep learning, especially the application of the convolutional neural network in the aspects of image classification, image detection, image segmentation and the like, the obtained effect is incomparable with the traditional algorithm. However, the classification algorithm based on the deep learning has a good effect only if the proportion of the area of the image occupied by the target area in the trained data set is large, and if the image is a high-resolution image but the classified target is small, the classified target may occupy less than 1% of the total area of the image, so that the image classification is performed by directly using the traditional classification algorithm based on the deep learning, and the accuracy is low. Meanwhile, if multiple classes of objects exist in the image, the class information of the objects cannot be acquired.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides an efficient detection algorithm aiming at small flaws in a large-resolution cloth image, and the method can realize the processing of a multi-scale characteristic map on a single-resolution input image so as to process multi-scale image blocks, so that various flaws with different sizes can be effectively processed, and the detection precision and speed are greatly improved; meanwhile, the method can acquire the approximate area of the flaw on the image classification frame and process the condition that various flaws exist in the picture.
In order to achieve the above purpose, the method provided by the invention comprises the following specific steps:
(1) Acquiring an image, shooting a cloth image by using a camera with the resolution of 2560-1920, acquiring a related data set, renaming the image, such as 1.jpg,2.jpg,3.jpg, 8230, M.jpg and the like, then zooming the image to 1024-768 size, and labeling the shot image by using a labeimg tool to acquire a label about a flaw in the image, wherein the label of the flaw comprises the coordinates (x 1, y 1) of the flaw at the upper left corner, the coordinates (x 2, y 2) of the lower right corner and the type deffectn of the flaw, wherein N represents a number, such as 1,2,3, 8230and the like, and particularly, if the shot image has no flaw, the label img cannot be used for processing, and only the type information norm is recorded;
(2) Dividing the image into a training set and a test set, wherein the two parts do not have the same image, the training set is used for training the test model, and the test set is used for evaluating the performance of the test model;
(3) Image preprocessing, including random up-down turning, random left-right turning, random illumination changing and the like, wherein the random up-down turning, the random left-right turning and the random illumination changing only aim at a training set, and particularly, when the random up-down turning and the random left-right turning are carried out, coordinate information of flaws also needs to be changed correspondingly;
(4) Training a test model, inputting images and label information in a training set after image preprocessing into the test model for training, wherein the test model is improved on the basis of se-next 101, so that a network can obtain a multi-scale characteristic diagram on the model aiming at an input image with single resolution, obtain the class probability value of each characteristic point on each characteristic diagram through forward propagation of the test model, calculate classification Loss through a Focal Loss function, and reversely propagate the test model by using a gradient descent algorithm with momentum;
(5) Performing cloth image inspection, namely inputting the images in the test set into a trained inspection model to extract features and acquiring the class probability value of each feature point on the multi-scale feature map; if all the feature points in two or more than three feature maps are judged to be norm, the image type is considered to be norm, and the defects exist in the image under other conditions; for an image judged to have a defect, each feature point corresponds to a certain image block in an original image, a related thermodynamic diagram is obtained by converting the pixel value of the corresponding image block through the prediction type of the feature point, the thermodynamic diagrams corresponding to a plurality of feature diagrams are superposed to obtain a final thermodynamic diagram, the final thermodynamic diagram is used for obtaining the approximate position of the defect, and the probability average value is taken for the image blocks near the defect to obtain the type of the defect.
The training in the step (4) comprises a training step based on an improved se-context 101 model, a transfer learning step, a two-stage learning rate adjusting step, a convolution network feature extraction step, a self-adaptive feature weight adjusting step, a multi-scale image block processing step, a Focal local calculating step and a model back propagation training step by utilizing a gradient descent algorithm with momentum.
As shown in fig. 1 and 2, the step (4) specifically includes:
(4.1) replacing the last global pooling layer of the original se-resnext101 model with 3 feature block pooling small modules consisting of parallel feature block global pooling layers and feature block maximum pooling layers, wherein each small module is in a parallel relation, the pooling layers in each small module have the same size but different sizes, in addition, the last fully-connected layer is replaced by 1 convolution operation with the size of 1 × 1 and the step length of 1, and the improved se-resnext101 model is used as an inspection model;
(4.2) initializing a modified version of the se-resnext101 model using the weights trained by the se-resnext101 model on the ImageNets image set, i.e. the check model, and we only retain the weights except all bias weights, the last global pooling layer, the last full-link layer and the softmax layer;
(4.3) adjusting the learning rate of the network by adopting a two-stage learning rate during model training, namely training the last three layers of the model including the feature block pooling module at a certain learning rate in the initial stage and keeping the weights of other layers of the model unchanged, using a larger learning rate for the last three layers of the model after training a plurality of iteration cycles (each iteration cycle traverses all images in a training set), using a smaller learning rate for other layers, and reducing the learning rate according to a certain rule;
(4.4) inputting the training image into an improved se-rescext 101 model, extracting features by using convolution operation, increasing the receptive field of a feature map, and enabling a network to adaptively adjust feature weight by using extrusion and excitation sub-modules contained in the original se-rescext 101 model, highlighting effective features, inhibiting ineffective features and improving two dimensions of a feature space and a feature channel;
(4.5) pooling sub-modules by using 3 parallel feature blocks for the feature map output by the last convolutional layer, as shown in fig. 3, the size of the pooling layer in each small module is the same, but the size of the pooling layer between different modules is different, so as to obtain 3 feature maps with different sizes; performing convolution operation with the size of 1 x1 and the step length of 1 on the obtained feature map, and then calculating a class probability value corresponding to each feature point on the feature maps with 3 different sizes by utilizing softmax;
(4.6) due to the existence of the receptive field, the feature points on the feature map correspond to image blocks of the original image, the real category of each image block can be obtained according to the positions of flaws in the image, so that the real category of the corresponding feature points is obtained, the classification Loss is calculated by using a Focal local function according to the predicted category probability value and the real category information, and finally, the gradient descent algorithm of the momentum is used for carrying out back propagation to update the detection model parameters.
The step (5) is specifically as follows: inputting a test image into a test model, acquiring feature maps with continuously increased receptive field and continuously reduced resolution ratio through forward propagation, processing the feature maps output by the last convolutional layer by using 3 parallel feature block pooling small modules, acquiring 3 feature maps with different sizes, performing convolution operation with the size of 1 x1 and the step length of 1 on the 3 feature maps with different sizes, and finally acquiring the category probability value corresponding to each feature point on each feature map by using softmax operation. If all the feature points in two or more than three feature maps are judged to be norm, the image type is considered to be norm, and in other cases, the image is considered to have defects. For an image with a defect, the pixel value of an image block of which a feature point distinguished as norm in a feature map is mapped back to an original image is assigned 0, the pixel value of an image block of which the other feature point in the feature map is mapped back to the original image is assigned 1, 3 thermodynamic diagrams are obtained in this way, the 3 thermodynamic diagrams are superposed to obtain a final thermodynamic diagram, the approximate position of the defect is obtained from the final thermodynamic diagram, and the type of the defect is determined by the type probability mean value of a plurality of feature points distinguished as defects.
Compared with the prior art, the invention has the beneficial effects that:
the invention realizes the processing of the multi-scale characteristic diagram of the input image with single resolution based on the improved se-resnext101 model, thereby processing the multi-scale image block, being suitable for various flaws with different sizes and greatly improving the detection precision and speed; based on the frame of the classification model, whether the predicted picture has defects or not can be given, and the images with various types of defects can be processed by giving the approximate positions and the corresponding types of the defects aiming at the images with the defects; the model has stronger generalization performance by utilizing transfer learning and two-stage learning rate adjustment, and the inspection performance of a new sample is improved; the Focal local function is used for replacing the traditional cross entropy Loss function to calculate the classification Loss, the problem of unbalanced samples is effectively solved, the attention of the test model to difficultly-classified samples is improved, and the performance of the test model is improved.
Drawings
FIG. 1 is a schematic diagram of se-resnext101 model
FIG. 2 is a schematic diagram of an improved se-resnext101 model
FIG. 3 is a block diagram of a feature block pooling module
FIG. 4 is a graph comparing the accuracy of two classifications of se-resnext model and se-resnext101 modified version on a test set
FIG. 5 is a histogram of flaw species identification accuracy in the test set of the improved se-resnext101 model
Detailed Description
The present invention is further explained below.
The implementation process and the embodiment of the invention are as follows:
(1) And (3) image acquisition, wherein a camera with the resolution of 2560-1920 is adopted to shoot the cloth images, 5000 cloth images are obtained in total and are renamed to be 1.jpg,2.jpg, \8230;, 5000.jpg, then the pictures are scaled to be 1024-768, and the shot images are marked by a labelImg tool to obtain the labels about the flaws in the images. The flaw label comprises the coordinates (x 1, y 1) of the upper left corner of the flaw in the image, the coordinates (x 2, y 2) of the lower right corner of the image and the flaw class deffectN, wherein N belongs to {1,2,3, \8230;, 9}, and represents a total of 9 types of flaws in the data set, the flaws are oil stains, jumping flowers, lacuna, hanging warps, thinning, holes, erasing holes, spots and pricking holes respectively, and the flaw sequence corresponds to the number of N in the deffectN one by one. For convenience of processing, information in an xml file obtained by processing the labelImg is converted into a txt file, only the types and the corresponding positions of defects in pictures are stored, each picture corresponds to one txt file, and the names of the images and the txt files are the same. Particularly, if the shot image has no flaws, the image cannot be processed by using labelImg, only the category information norm is recorded and stored in a txt file;
(2) Splitting the image, namely dividing the image into a training set and a test set, wherein the training set is used for training a test model, and the test set is used for evaluating the performance of the test model, the training set is the front 4500 images in the data set, and the test set is the rear 500 images in the data set;
(3) Image preprocessing, taking an image I in a training set as an example, inputting the image I and contents in a corresponding txt file into an image preprocessing module for online data enhancement and content transformation, wherein the contents of the txt file are stored in a list; for images with defects in the image, the list contents are as follows:
for images where no flaws exist in the image, the list contents are as follows:
[norm]
the on-line data enhancement comprises random up-down turning, random left-right turning, random illumination change and the like, and particularly, when the random up-down turning and the random left-right turning are carried out, the coordinate information of the flaw also needs to be changed correspondingly;
(4) Constructing an inspection model, as shown in fig. 1 and fig. 2, replacing the last global pooling layer of an original se-resnext101 model with 3 feature block pooling small modules consisting of parallel feature block global pooling layers and feature block maximum pooling layers, wherein the structure of the feature block pooling small modules is as shown in fig. 3, each small module is in a parallel relationship, the pooling layers in each small module have the same size, but the pooling layers between different modules have different sizes, in addition, the last fully-connected layer is replaced by 1 convolution operation with the size of 1 × 1 and the step length of 1, and the improved se-resnext101 model is used as the inspection model;
the parameter settings of each feature block pooling module in the specific implementation are shown in table 1:
TABLE 1 parameter settings for feature Block pooling Module
After an image I with a resolution of 1024 × 768 is subjected to forward propagation of an improved se-rescext 101 model, the resolution of a feature map output by an improved se-rescext 101 module 5 is 32 × 24, the resolution of the feature map processed by 3 parallel feature block pooling modules is 10 × 6, 12 × 8 and 14 × 10, respectively, wherein the size of an image block in an original image corresponding to a feature point on the feature map with the resolution of 10 × 6 is 448, and the sliding step size is 64; the size of the feature point on the feature map with the resolution of 12 × 8 corresponding to the image block in the original image is 320 × 320, and the sliding step size is 64; the feature points on the feature map with the resolution of 14 × 6 correspond to the image block size in the original image of 192 × 192, and the sliding step size is 64. By the method, 3 image blocks with different sizes can be obtained, so that the problem caused by the fact that small flaws account for too small in an original image is effectively solved, flaws with various scales can be processed, and the inspection performance of the model is stronger.
(5) Training a test model, inputting images in a training set and corresponding label information into the test model, obtaining a class probability value corresponding to each feature point on each feature map through forward propagation, calculating classification Loss through a Focal Loss function, training the test model by utilizing a gradient descent algorithm with momentum, and obtaining network parameters of the test model;
in specific implementation, the power is set to be 0.9, 1 image is input each time, 4500 steps are 1 iteration cycle, 50 iteration cycles are set, the last three layers of the model are set to have the learning rate of 0.001 in the first 10 iteration cycles, and the learning rates of other layers are set to be 0; setting the learning rate of the last three layers of the model to be 0.0005 and the learning rate of other layers to be 0.00005 in the subsequent iteration period, and simultaneously changing the learning rate to be 0.94 of the original learning rate in every 4 iteration periods; in the Focal local function, α is set to 0.25, and β is set to 2. And after the training is finished, saving the parameters of the test model.
(6) Performing cloth image inspection, namely inputting the images in the test set into a trained inspection model to extract features and acquiring the class probability value of each feature point on the multi-scale feature map; if all the feature points in two or more feature maps in the three feature maps are judged to be norm, the image type is considered to be norm, and the defects exist in the image under other conditions; for an image judged to have a defect, each feature point corresponds to a certain image block in an original image, a related thermodynamic diagram is obtained by converting the pixel value of the corresponding image block through the prediction type of the feature point, the thermodynamic diagrams corresponding to a plurality of feature diagrams are superposed to obtain a final thermodynamic diagram, the final thermodynamic diagram is used for obtaining the approximate position of the defect, and the probability average value is taken for the image blocks near the defect to obtain the type of the defect.
In this embodiment, a test is finally performed on a cloth test set, and fig. 4 shows the accuracy of the se-rescext 101 model and the improved se-rescext 101 model in the result of the two-classification test, that is, the accuracy of the image discrimination as a normal sample and a flaw sample, and it can be seen that the accuracy of the improved se-rescext 101 model is significantly higher than that of the se-rescext 101 model, and it can be seen from the test result that the method can perform efficient test on small flaws in a high-resolution image. Fig. 5 shows the accuracy of the flaw classification of the modified version of the rescext 101 model on the test set, and it can be seen that the method of the present patent can detect the flaw types existing in the image sub-modules for the images with various flaws.
Claims (2)
1. An efficient inspection algorithm for small defects in large resolution cloth images, comprising the steps of:
(1) Acquiring an image, shooting a cloth image by using a camera with the resolution of 2560 × 1920, acquiring a related data set, renaming the image, zooming the image to 1024 × 768, labeling the shot image by using a label Img tool, and acquiring a label about a flaw in the image, wherein the label about the flaw comprises coordinates (x 1, y 1) of the flaw at the upper left corner, coordinates (x 2, y 2) of the lower right corner and a category deffectN of the flaw, wherein N represents a number, and if the shot image has no flaw, only recording category information norm of the shot image without processing by using the label Img;
(2) Dividing the image into a training set and a test set, wherein the two parts do not have the same image, the training set is used for training the test model, and the test set is used for evaluating the performance of the test model;
(3) Image preprocessing, including random up-down turning, random left-right turning and random illumination change, wherein the random up-down turning, the random left-right turning and the random illumination change only aim at a training set, and when the random up-down turning and the random left-right turning are carried out, coordinate information of flaws also needs to be changed correspondingly;
(4) Training a test model, inputting images and label information in a training set after image preprocessing into the test model for training, wherein the test model is improved on the basis of se-next 101, so that a network can obtain a multi-scale feature map on the model aiming at an input image with single resolution, obtain the class probability value of each feature point on each feature map through forward propagation of the test model, calculate classification Loss through a Focal local function, and reversely propagate the test model by utilizing a gradient descent algorithm with momentum;
(5) Performing cloth image inspection, inputting the images in the test set into a trained inspection model to extract features and acquiring the class probability value of each feature point on the multi-scale feature map; if all the feature points in two or more feature maps in the three feature maps are judged to be norm, the image type is considered to be norm, and the defects exist in the image under other conditions; for the image judged to have the defect, each feature point corresponds to a certain image block in an original image, a related thermodynamic diagram is obtained by converting the pixel value of the corresponding image block through the prediction type of the feature point, the thermodynamic diagrams corresponding to a plurality of feature diagrams are superposed to obtain a final thermodynamic diagram, the final thermodynamic diagram is used for obtaining the approximate position of the defect, the probability mean value is taken for the image blocks near the defect to obtain the type of the defect, and the algorithm can process the condition that the image has a plurality of defects and obtain the type and the approximate position of each defect;
the training in the step (4) comprises a training step based on an improved se-next 101 model, a transfer learning step, a two-stage learning rate adjusting step, a convolution network feature extracting step, a self-adaptive feature weight adjusting step, a multi-scale feature map processing step, a Focal local calculating step and a model back propagation training step by utilizing a gradient descent algorithm with momentum;
the step (4) is specifically as follows:
(4.1) replacing the last global pooling layer of the original se-resnext101 model with 3 feature block pooling small modules consisting of parallel feature block global pooling layers and feature block maximum pooling layers, wherein each small module is in a parallel relation, the pooling layers in each small module have the same size but different sizes, in addition, the last fully-connected layer is replaced by 1 convolution operation with the size of 1 × 1 and the step length of 1, and the improved se-resnext101 model is used as an inspection model;
(4.2) initializing the improved se-resnext101 model using the weights trained by the se-resnext101 model on the ImageNets image set, i.e. verifying the model, only preserving the weights except all bias weights, the last global pooling layer, the last full-link layer and the softmax layer;
(4.3) adjusting the learning rate of the network by adopting two-stage learning rate during model training, namely training the last three layers of the model including the feature block pooling module at a certain learning rate in an initial stage and keeping the weights of other layers of the model unchanged, using a larger learning rate for the last three layers of the model after training a plurality of iteration cycles, using a smaller learning rate for other layers, reducing the learning rate according to a certain rule, and traversing all images in a training set in each iteration cycle;
(4.4) inputting the training image into an improved se-rescext 101 model, extracting features by using convolution operation, increasing the receptive field of a feature map, and enabling a network to adaptively adjust feature weight by using extrusion and excitation sub-modules contained in the original se-rescext 101 model, highlighting effective features, inhibiting ineffective features and improving two dimensions of a feature space and a feature channel;
(4.5) utilizing 3 parallel feature block pooling sub-modules for the feature map output by the last convolutional layer, wherein the pooling layers in each small module have the same size but different sizes, so as to obtain 3 feature maps with different sizes; performing convolution operation with the size of 1 x1 and the step length of 1 on the obtained feature maps, and then calculating category probability values corresponding to each feature point on the 3 feature maps with different sizes by using softmax;
(4.6) due to the existence of the receptive field, the feature points on the feature map correspond to image blocks of the original image, the real category of each image block can be obtained according to the positions of flaws in the image, so that the real category of the corresponding feature points is obtained, the classification Loss is calculated by using a Focal Loss function according to the predicted category probability value and the real category information, and finally, the detection model parameters are updated by using a gradient descent algorithm with momentum.
2. An efficient inspection algorithm for small defects in large-resolution cloth images as claimed in claim 1, wherein said step (5) is specifically: inputting a test image into an inspection model, acquiring a feature map with an increasing receptive field and a decreasing resolution ratio through forward propagation, processing the feature map output by the last convolutional layer by using 3 parallel feature block pooling small modules, acquiring 3 feature maps with different sizes, performing convolution operation with the size of 1 × 1 and the step length of 1 on the 3 feature maps with different sizes, and finally acquiring a category probability value corresponding to each feature point on each feature map by using softmax operation; if all the feature points in two or more feature maps in the three feature maps are judged to be norm, the image type is considered to be norm, and the defects exist in the image under other conditions; the method comprises the steps of obtaining 3 thermodynamic diagrams by assigning the pixel value of an image block which is obtained by mapping a feature point which is judged to be norm back to the original image in a feature diagram to 0 and assigning the pixel value of an image block which is obtained by mapping other feature points in the feature diagram back to the original image to 1, superposing the 3 thermodynamic diagrams to obtain a final thermodynamic diagram, and obtaining the approximate position of a defect from the final thermodynamic diagram, wherein the type of the defect is determined by the probability mean value of the types of a plurality of feature points which are judged to be defects.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811336000.3A CN109509187B (en) | 2018-11-05 | 2018-11-05 | Efficient inspection algorithm for small defects in large-resolution cloth images |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811336000.3A CN109509187B (en) | 2018-11-05 | 2018-11-05 | Efficient inspection algorithm for small defects in large-resolution cloth images |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109509187A CN109509187A (en) | 2019-03-22 |
| CN109509187B true CN109509187B (en) | 2022-12-13 |
Family
ID=65748135
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811336000.3A Active CN109509187B (en) | 2018-11-05 | 2018-11-05 | Efficient inspection algorithm for small defects in large-resolution cloth images |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109509187B (en) |
Families Citing this family (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110009629A (en) * | 2019-04-12 | 2019-07-12 | 北京天明创新数据科技有限公司 | A kind of pneumoconiosis screening system and its data training method |
| CN110321805B (en) * | 2019-06-12 | 2021-08-10 | 华中科技大学 | Dynamic expression recognition method based on time sequence relation reasoning |
| CN110610482A (en) * | 2019-08-12 | 2019-12-24 | 浙江工业大学 | Workpiece flaw detection method based on resnet |
| CN110504029B (en) * | 2019-08-29 | 2022-08-19 | 腾讯医疗健康(深圳)有限公司 | Medical image processing method, medical image identification method and medical image identification device |
| CN111028250A (en) * | 2019-12-27 | 2020-04-17 | 创新奇智(广州)科技有限公司 | Real-time intelligent cloth inspecting method and system |
| CN111340783A (en) * | 2020-02-27 | 2020-06-26 | 创新奇智(广州)科技有限公司 | Real-time cloth defect detection method |
| CN111325744A (en) * | 2020-03-06 | 2020-06-23 | 成都数之联科技有限公司 | Sample processing method in panel defect detection |
| CN112053357A (en) * | 2020-09-27 | 2020-12-08 | 同济大学 | FPN-based steel surface flaw detection method |
| CN112348776A (en) * | 2020-10-16 | 2021-02-09 | 上海布眼人工智能科技有限公司 | Fabric flaw detection method based on EfficientDet |
| CN113689381B (en) * | 2021-07-21 | 2024-02-27 | 航天晨光股份有限公司 | Corrugated pipe inner wall flaw detection model and detection method |
| CN115797349B (en) * | 2023-02-07 | 2023-07-07 | 广东奥普特科技股份有限公司 | Defect detection method, device and equipment |
| CN116304811B (en) * | 2023-02-28 | 2024-01-16 | 王宇轩 | Dynamic sample weight adjustment method and system based on focus loss function |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102331425A (en) * | 2011-06-28 | 2012-01-25 | 合肥工业大学 | Textile defect detection method based on defect enhancement |
| CN103456021A (en) * | 2013-09-24 | 2013-12-18 | 苏州大学 | Piece goods blemish detecting method based on morphological analysis |
| CN107316295A (en) * | 2017-07-02 | 2017-11-03 | 苏州大学 | A kind of fabric defects detection method based on deep neural network |
| CN108647723A (en) * | 2018-05-11 | 2018-10-12 | 湖北工业大学 | A kind of image classification method based on deep learning network |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20010012381A1 (en) * | 1997-09-26 | 2001-08-09 | Hamed Sari-Sarraf | Vision-based, on-loom fabric inspection system |
| GB201704373D0 (en) * | 2017-03-20 | 2017-05-03 | Rolls-Royce Ltd | Surface defect detection |
-
2018
- 2018-11-05 CN CN201811336000.3A patent/CN109509187B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102331425A (en) * | 2011-06-28 | 2012-01-25 | 合肥工业大学 | Textile defect detection method based on defect enhancement |
| CN103456021A (en) * | 2013-09-24 | 2013-12-18 | 苏州大学 | Piece goods blemish detecting method based on morphological analysis |
| CN107316295A (en) * | 2017-07-02 | 2017-11-03 | 苏州大学 | A kind of fabric defects detection method based on deep neural network |
| CN108647723A (en) * | 2018-05-11 | 2018-10-12 | 湖北工业大学 | A kind of image classification method based on deep learning network |
Non-Patent Citations (1)
| Title |
|---|
| 基于稀疏编码的双尺度布匹瑕疵检测;张龙剑 等;《计算机应用》;20141010;第34卷(第10期);第3009-3013页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109509187A (en) | 2019-03-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109509187B (en) | Efficient inspection algorithm for small defects in large-resolution cloth images | |
| CN111179251B (en) | Defect detection system and method based on twin neural network and by utilizing template comparison | |
| CN108562589B (en) | Method for detecting surface defects of magnetic circuit material | |
| CN108765412B (en) | Strip steel surface defect classification method | |
| CN108960245B (en) | Tire mold character detection and recognition method, device, equipment and storage medium | |
| CN109684922B (en) | Multi-model finished dish identification method based on convolutional neural network | |
| CN110033473B (en) | Moving target tracking method based on template matching and depth classification network | |
| CN111257341B (en) | Underwater building crack detection method based on multi-scale features and stacked full convolution network | |
| CN109002755B (en) | Age estimation model construction method and estimation method based on face image | |
| CN110532946B (en) | Method for identifying axle type of green-traffic vehicle based on convolutional neural network | |
| CN109598287A (en) | The apparent flaws detection method that confrontation network sample generates is generated based on depth convolution | |
| CN109284779A (en) | Object detecting method based on the full convolutional network of depth | |
| CN110245697B (en) | Surface contamination detection method, terminal device and storage medium | |
| CN112926652B (en) | Fish fine granularity image recognition method based on deep learning | |
| CN110008899B (en) | Method for extracting and classifying candidate targets of visible light remote sensing image | |
| CN112365497A (en) | High-speed target detection method and system based on Trident Net and Cascade-RCNN structures | |
| CN109977834B (en) | Method and device for segmenting human hand and interactive object from depth image | |
| CN110827312A (en) | Learning method based on cooperative visual attention neural network | |
| CN111539957A (en) | Image sample generation method, system and detection method for target detection | |
| CN111382766A (en) | Equipment fault detection method based on fast R-CNN | |
| CN112001901A (en) | Apple defect detection method and system based on convolutional neural network | |
| CN111340019A (en) | Grain bin pest detection method based on Faster R-CNN | |
| CN111695640B (en) | Foundation cloud picture identification model training method and foundation cloud picture identification method | |
| CN115409797A (en) | PCB defect image detection method based on improved deep learning algorithm | |
| CN111914902A (en) | Traditional Chinese medicine identification and surface defect detection method based on deep 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| DD01 | Delivery of document by public notice |
Addressee: Chen Chucheng Document name: Notification of Qualified Procedures |
|
| DD01 | Delivery of document by public notice |