Deep learning-based textile jacquard detection method and system

Abstract

The invention discloses a deep learning-based textile jacquard detection method and system, and relates to the field of image data processing, and the method comprises the steps: setting a defect total category set, collecting textile images which comprise various jacquard patterns and contain various possible defects, marking defect images, and enhancing the defect images through a data enhancement technology, so that the defect detection accuracy is improved. Dividing the enhanced defect image set into a defect image training set and a defect image test set, and setting a defect image training label set and a defect image test label set; a CNN convolutional neural network is combined with an RNN recurrent neural network, an initial CRNN convolutional recurrent neural network model is constructed, the initial CRNN convolutional recurrent neural network model is trained and optimized by using the defect image training set, the defect image training label set, the defect image test set and the defect image test label set, and a final CRNN convolutional recurrent neural network model is obtained; and analyzing the images of the textiles collected in real time, and carrying out defect classification on the images of the textiles with defects.

Description

Technical Field

[0001] The present invention belongs to the technical field of image data processing, and in particular, relates to a textile jacquard detection method and system based on deep learning. Background Art

[0002] In the textile industry, jacquard pattern detection is an important part of ensuring product quality. Traditionally, jacquard detection mostly uses manual visual inspection or automatic detection systems based on simple image processing; although manual inspection has high accuracy, it is inefficient, labor-intensive and easily affected by subjective factors; and systems based on simple image processing often have problems with insufficient recognition accuracy and high false alarm rates when facing complex and changeable jacquard patterns; especially when the jacquard pattern becomes more refined and the color changes are rich, the existing technical solutions are difficult to effectively respond, resulting in low production efficiency and increased costs. Summary of the invention

[0003] In response to the problems in the related art, the present invention proposes a textile jacquard detection method and system based on deep learning to overcome the above-mentioned technical problems existing in the existing related technology.

[0004] In order to solve the above technical problems, the present invention is achieved through the following technical solutions: S1. Set a total defect type set; collect textile images containing various jacquard patterns and covering various possible defects to obtain a defect image set; annotate the defect image set to obtain a defect annotated image set; enhance the defect image set through data enhancement technology to obtain an enhanced defect image set; S2. Divide the enhanced defect image set into a defect image training set and a defect image test set, and set a defect image training label set and a defect image test label set accordingly; combine the CNN convolutional neural network with the RNN recurrent neural network to construct an initial CRNN convolutional recurrent neural network model; use the defect image training set and the defect image training set, the defect image test set and the defect image test label set to train and optimize the initial CRNN convolutional recurrent neural network model to obtain the final CRNN convolutional recurrent neural network model; S3. Analyze the images of textiles collected in real time, and classify the images of textiles with defects.

[0005] Preferably, the S1 comprises the following steps: S11. Set the total defect type set a ={ a 1， a 2，...， a i，...， a r},ina i Defect type set a The i defects, r Represents the total set of defect types a The Chinese Communist Party r types of defects; S12. Collect textile images containing various jacquard patterns and covering various possible defects to obtain a defect image set b ={ b 1, b 2,..., b i,..., b m},in b i Defect image set b The first i Defective images, m Defect image set b The Chinese Communist Party m defect images; S13. Check defect image set b For each image in the image, identify the defects in the image and classify them according to the total defect type set a , for the defect image set b Defect annotation is performed on each image in the image set to obtain the defect annotation image set b' ={ b' 1, b' 2,..., b' i,..., b' m},in b' i Defect image set b Middle i annotated defect images; S14. Defect labeling image set b' ={ b' 1, b' 2,..., b' i,..., b' m}, such as rotation, scaling, cropping, and color transformation, to create more training samples and obtain an enhanced defect image set c ={ c 1, c 2,..., ci,..., c o},in c i Represents the enhanced defect image set c The i defect images, o Represents the enhanced defect image set c The Chinese Communist Party o defect images; By setting the total defect category set, all possible defect types are defined, providing a standardized defect classification framework for subsequent image annotation and model training; collecting a diverse image dataset containing textile images with various jacquard patterns and all possible defect situations helps the model learn the manifestations of different defects and improve the generalization ability of the model; accurately matching the defects in the image with the defect types in the total defect category set improves the accuracy of model training; data enhancement obtains defect characteristics under different perspectives and changes.

[0006] Preferably, S2 comprises the following steps: S21. Divide the enhanced defect image set into a defect image training set d ={ d 1, d 2,..., d i,..., d p} and defect image test set e ={ e 1, e 2,..., e i..., e q}, according to the defect image training set and the defect image test set, a corresponding defect image training label set is set d' ={ d' 1, d' 2,..., d' i,..., d' p}、Defect image test label set e' ={ e' 1, e' 2,..., e' i..., e' q};in di Defect image training set d Middle i defect images, p Defect image training set d The Chinese Communist Party p defect images, e i Denotes defect image test set e Middle i defect images, q Denotes defect image test set e The Chinese Communist Party q defect images, d' i Defect image training set d Middle i The defect types of defect images, e' i Denotes defect image test set e Middle i Defect types of defect images; S22. Adopt the architecture combining CNN convolutional neural network and RNN recurrent neural network to build a special initial CRNN convolutional recurrent neural network model for jacquard detection. This model automatically extracts the hierarchical features of the jacquard pattern through multiple convolutional layers, and uses RNN to process the sequence information of the pattern to enhance the recognition ability of complex textures and color changes. S23. Using defect image training set d and defect image training label set d' Train the initial CRNN convolutional recurrent neural network model; after the training is completed, the trained CRNN convolutional recurrent neural network model is obtained, and then the defect image test set is e and defect image test label set e' Input into the trained CRNN convolutional recurrent neural network model for testing, optimize the trained CRNN convolutional recurrent neural network model according to the test results, and obtain the final CRNN convolutional recurrent neural network model; By dividing the enhanced defect image set into a training set and a test set, data is provided for the training model. By combining the CNN convolutional neural network with the RNN recurrent neural network, a CRNN convolutional recurrent neural network model is constructed to better handle the complex texture and color changes of the jacquard pattern. By adjusting the model parameters, the model can learn from the training data. The test model evaluates the generalization ability of the model to ensure that the model can perform well in practical applications. The optimization model is to adjust the model according to the test results to improve its performance.

[0007] Preferably, the S23 comprises the following steps: S231, constructing an initial CRNN convolutional recurrent neural network model, setting the number of iterations of the initial CRNN convolutional recurrent neural network model training to α, the batch size to β, the optimizer to adam optimizer, and the initial learning rate to ζ; S232, set the current training iteration number to z and the training error threshold is δ; according to the batch size β, the adam optimizer and the initial learning rate ζ, the defect image training set d and defect image training label set d′ Input into the initial CRNN convolutional recurrent neural network model for training, when the training error of the initial CRNN convolutional recurrent neural network model is <δ or when z When ≥α, stop training and obtain the trained CRNN convolutional recurrent neural network model; S233, setting an accuracy threshold λ, testing the defect image e and defect image test label set e′ Input into the trained CRNN convolutional recurrent neural network model for testing to obtain a test accuracy θ; when θ≥λ, use the trained CRNN convolutional recurrent neural network model as the final CRNN convolutional recurrent neural network model; when θ<λ, return to S232 to continue training the trained CRNN convolutional recurrent neural network model until θ≥λ, and obtain the final CRNN convolutional recurrent neural network model; By setting the relevant parameters of the CRNN convolutional recurrent neural network model, and then inputting the defect image training set into the CRNN convolutional recurrent neural network model for training, a trained CRNN convolutional recurrent neural network model is obtained. The defect image test set is used to test the trained CRNN convolutional recurrent neural network model. According to the test results, the trained CRNN convolutional recurrent neural network model is optimized to make the CRNN convolutional recurrent neural network model more accurate.

[0008] Preferably, the S232 comprises the following steps: S2321, the training error of the initial CRNN convolutional recurrent neural network model is calculated using a missing function, and the function expression is as follows: ; in, n is the sample size; y i It is i The true label of each sample, which takes the value 0 or 1; It is i The predicted probability of a sample indicates the probability that the model predicts that the sample belongs to category 1; The training error of the initial CRNN convolutional recurrent neural network model is calculated by applying the missing function, and the result is used as an important indicator in the model training process to more accurately evaluate the performance of the model and guide the optimization and adjustment of the model.

[0009] Preferably, the S233 comprises the following steps: S2331, Randomly initialize the position of the particle swarm h ={ h 1, h 2,..., h i,..., h NP},in h i Indicates i The learning rate value, Np is the number of particles; the speed of initializing particles v ={ v 1, v 2,..., v i,..., v NP},in v i Indicates i Initialize the optimal position of each particle f ={ f 1, f 2,..., f i,..., f NP},in f i Indicates i The best learning rate achieved by each particle; Initialize the global best position l for f Any position in; Set the maximum number of iterations to f' , the performance of the global best position reaches the target value f'' ; S2332. For each particle, update the speed v i t+1= j × v i t+ c 1 × r 1 ×( fi t- h i )+ c 2 × r 2 ×( l t- h i t ),in t represents the current time step, t +1 indicates the next time step, j is the inertia weight, c 1 and c 2 is the learning factor, r 1 and r 2 is a random number in the range of 0 to 1; Update Location h i t+1= h i t+ v i t+1 , use the updated learning rate to train the CRNN convolutional recurrent neural network model and evaluate its performance. If the new position h i t+1 The performance is better than f i t , then update f i , if the new position h i t+1 The performance is not better than f i t , keep the original f i ; At time step t +1, traverse the historical best positions of all particles f i , and select the position with the best performance as the new global best position l t+1 ; S2333, repeat S2332, when the performance of the global best position reaches the target value f′′ Or the number of iterations reaches f' When , the iteration process is stopped, and the learning rate value corresponding to the global best position is output to obtain the optimal learning ratef' ; S2334, using the optimal learning rate as the learning rate of the trained CRNN convolutional recurrent neural network model to obtain an optimized CRNN convolutional recurrent neural network model; The particle swarm optimization algorithm is used to optimize the learning rate parameters in the trained CRNN convolutional recurrent neural network model. During the optimization, the performance of the CRNN convolutional recurrent neural network model is used as the optimization index, so as to continuously improve the performance of the CRNN convolutional recurrent neural network model, making the CRNN convolutional recurrent neural network model more accurate in identifying defects in textile images.

[0010] Preferably, S3 comprises the following steps: S31, deploy the final CRNN convolutional recurrent neural network model to dedicated image processing hardware or cloud server, and build a real-time jacquard detection system in combination with high-speed camera and image transmission system; S32, transmitting the images of the textiles captured by the system to the final CRNN convolutional recurrent neural network model to classify the images of the defective textiles; Through real-time image acquisition and processing, the system can detect and handle defective products in a timely manner, thereby improving the quality and efficiency of textile production.

[0011] A system for a textile jacquard detection method based on deep learning, comprising an image set construction and enhancement module, a CRNN convolutional recurrent neural network model training and optimization module, and a CRNN convolutional recurrent neural network model application module; The image set construction and enhancement module is used to collect textile images containing various jacquard patterns and covering various possible defects, and obtain an enhanced defect image set through annotation enhancement, which plays a role in expanding the data set and improving the generalization ability of subsequent models; The training and optimization module of the CRNN convolutional recurrent neural network model is used to train, test and optimize the initial CRNN convolutional recurrent neural network model to obtain the final CRNN convolutional recurrent neural network model; The CRNN convolutional recurrent neural network model application module is used to classify the real-time collected images containing jacquard patterns and textiles.

[0012] The present invention has the following beneficial effects: By defining all possible defect types, we ensure that the model can identify and classify all relevant defects. Collecting diverse image data helps the model learn the characteristics of different jacquard patterns and defect types. Image annotation provides the real labels required for model training, which helps the model understand the characteristics of different defects. Through data enhancement, the diversity of training samples is increased, which helps to improve the generalization and robustness of the model and reduce overfitting.

[0013] By dividing the dataset into a training set and a test set, the performance of the model can be objectively evaluated to ensure that the model performs well on unseen data; the CRNN model combines the feature extraction capabilities of CNN and the sequence modeling capabilities of RNN, and can more effectively process complex image data and improve detection accuracy; through training and optimization, the model can learn the characteristics of defect images and verify its performance on the test set, thereby improving the accuracy and reliability of the model.

[0014] Real-time image acquisition and analysis can detect problems in a timely manner, reduce the production of defective products, and improve production efficiency; defect classification helps to quickly identify and handle problems, reduce the workload of manual inspection, reduce labor costs, and at the same time improve the objectivity and consistency of detection.

[0015] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the invention, the drawings required for describing the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For ordinary technicians in this field, they can also obtain drawings based on these drawings without paying creative work.

[0017] Figure 1 The present invention is a flow chart of the textile jacquard detection method based on deep learning. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the invention to clearly and completely describe the technical solutions in the embodiments of the invention. Obviously, the described embodiments are only part of the embodiments of the invention, not all of the embodiments. Based on the embodiments in the invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the invention.

[0019] In the description of the present invention, it is necessary to understand that the terms "opening", "upper", "lower", "top", "middle", "inside" and the like indicating orientation or positional relationship are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the components or elements referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as limiting the invention.

[0020] The present invention is a textile jacquard detection method based on deep learning, comprising the following steps: S1. Set a total defect type set; collect textile images containing various jacquard patterns and covering various possible defects to obtain a defect image set; annotate the defect image set to obtain a defect annotated image set; enhance the defect image set through data enhancement technology to obtain an enhanced defect image set; The S1 comprises the following steps: S11. Set the total defect type set a ={ a 1， a 2，...， a i，...， a r},in a i Represents the total set of defect types a The i defects, r Represents the total set of defect types a The Chinese Communist Party r types of defects; S12. Collect textile images containing various jacquard patterns and covering various possible defects to obtain a defect image set b ={ b 1, b 2,..., b i,..., b m},in b i Defect image set b The first i Defective images, m Defect image set b The Chinese Communist Party m defect images; S13. Check defect image set b For each image in the image, identify the defects in the image and classify them according to the total defect type set a , for the defect image set b Defect annotation is performed on each image in the image set to obtain the defect annotation image set b' ={ b' 1, b' 2,..., b' i,..., b' m},in b' i Defect image set b Middle iannotated defect images; S14. Defect labeling image set b' ={ b' 1, b' 2,..., b' i,..., b' m} to obtain the enhanced defect image set c ={ c 1, c 2,..., c i,..., c o},in c i Represents the enhanced defect image set c The i defect images, o Represents the enhanced defect image set c The Chinese Communist Party o defect images; S2. Divide the enhanced defect image set into a defect image training set and a defect image test set, and set a defect image training label set and a defect image test label set accordingly; combine the CNN convolutional neural network with the RNN recurrent neural network to construct an initial CRNN convolutional recurrent neural network model; use the defect image training set and the defect image training label set, the defect image test set and the defect image test label set to train and optimize the initial CRNN convolutional recurrent neural network model to obtain the final CRNN convolutional recurrent neural network model; The S2 comprises the following steps: S21. Divide the enhanced defect image set into a defect image training set d ={ d 1, d 2,..., d i,..., d p} and defect image test set e ={ e 1, e 2,..., e i..., e q}, according to the defect image training set and the defect image test set, a corresponding defect image training label set is set d' ={ d' 1,d' 2,..., d' i,..., d' p}、Defect image test label set e' ={ e' 1, e' 2,..., e' i..., e' q};in d i Defect image training set d Middle i defect images, p Defect image training set d The Chinese Communist Party p defect images, e i Denotes defect image test set e Middle i defect images, q Denotes defect image test set e The Chinese Communist Party q defect images, d' i Defect image training set d Middle i The defect types of defect images, e' i Denotes defect image test set e Middle i Defect types of defect images; S22, using the architecture combining CNN convolutional neural network and RNN recurrent neural network to build the initial CRNN convolutional recurrent neural network model; S23. Using defect image training set d and defect image training label set d′ Train the initial CRNN convolutional recurrent neural network model; after the training is completed, the trained CRNN convolutional recurrent neural network model is obtained, and then the defect image test set is e and defect image test label set e′ Input into the trained CRNN convolutional recurrent neural network model for testing, optimize the trained CRNN convolutional recurrent neural network model according to the test results, and obtain the final CRNN convolutional recurrent neural network model; The S23 comprises the following steps: S231, constructing an initial CRNN convolutional recurrent neural network model, setting the number of iterations of the initial CRNN convolutional recurrent neural network model training to α, the batch size to β, the optimizer to adam optimizer, and the initial learning rate to ζ; S232, set the current training iteration number to z And the training error threshold is δ; according to The defect image training set is converted into d and defect image training label set d′ Input into the initial CRNN convolutional recurrent neural network model for training, when the training error of the initial CRNN convolutional recurrent neural network model is <δ or when z When ≥α, stop training and obtain the trained CRNN convolutional recurrent neural network model; The S232 comprises the following steps: S2321, the training error of the initial CRNN convolutional recurrent neural network model is calculated using a missing function, and the function expression is as follows: ; in, n is the sample size; y i It is i The true label of each sample, which takes the value 0 or 1; It is i The predicted probability of a sample indicates the probability that the model predicts that the sample belongs to category 1; S233, setting an accuracy threshold λ, testing the defect image e and defect image test label set e′ Input into the trained CRNN convolutional recurrent neural network model for testing to obtain a test accuracy θ; when θ≥λ, use the trained CRNN convolutional recurrent neural network model as the final CRNN convolutional recurrent neural network model; when θ<λ, return to S232 to continue training the trained CRNN convolutional recurrent neural network model until θ≥λ, and obtain the final CRNN convolutional recurrent neural network model; The S233 comprises the following steps: S2331, Randomly initialize the position of the particle swarm h ={ h 1, h 2,..., h i,..., h NP},in h iIndicates i The learning rate value, Np is the number of particles; the speed of initializing particles v ={ v 1, v 2,..., v i,..., v NP},in v i Indicates i Initialize the optimal position of each particle f ={ f 1, f 2,..., f i,..., f NP},in f i Indicates i The best learning rate achieved by each particle; Initialize the global best position l for f Any position in; Set the maximum number of iterations to f' , the performance of the global best position reaches the target value f'' ; S2332. For each particle, update the speed v i t+1= j × v i t+ c 1 × r 1 ×( f i t- h i )+ c 2 × r 2 ×( l t- h i t ),in t represents the current time step, t +1 indicates the next time step, j is the inertia weight, c 1 and c 2 is the learning factor,r 1 and r 2 is a random number in the range of 0 to 1; Update Location h i t+1= h i t+ v i t+1 , use the updated learning rate to train the CRNN convolutional recurrent neural network model and evaluate its performance. If the new position h i t+1 The performance is better than f i t , then update f i , if the new position h i t+1 The performance is not better than f i t , keep the original f i ; At time step t +1, traverse the historical best positions of all particles f i , and select the position with the best performance as the new global best position l t+1 ; S2333, repeat S2332, when the performance of the global best position reaches the target value f′′ Or the number of iterations reaches f' When , the iteration process is stopped, and the learning rate value corresponding to the global best position is output to obtain the optimal learning rate f' ; S2334, using the optimal learning rate as the learning rate of the trained CRNN convolutional recurrent neural network model to obtain an optimized CRNN convolutional recurrent neural network model; S3, analyzing the images of textiles collected in real time, and classifying the images of textiles with defects; The S3 comprises the following steps: S31, deploy the final CRNN convolutional recurrent neural network model to dedicated image processing hardware or cloud server, and build a real-time jacquard detection system in combination with high-speed camera and image transmission system; S32, transmitting the images of the textiles captured by the system to the final CRNN convolutional recurrent neural network model to classify the images of the defective textiles; A system for a textile jacquard detection method based on deep learning, comprising an image set construction and enhancement module, a CRNN convolutional recurrent neural network model training and optimization module, and a CRNN convolutional recurrent neural network model application module; The image set construction and enhancement module is used to collect textile images containing various jacquard patterns and covering various possible defects, and obtain an enhanced defect image set through annotation enhancement, which plays a role in expanding the data set and improving the generalization ability of subsequent models; The training and optimization module of the CRNN convolutional recurrent neural network model is used to train, test and optimize the initial CRNN convolutional recurrent neural network model to obtain the final CRNN convolutional recurrent neural network model; The CRNN convolutional recurrent neural network model application module is used to classify the real-time collected images containing jacquard patterns and textiles.

[0021] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0022] The preferred embodiments of the invention disclosed above are only used to help explain the invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the invention, so that those skilled in the art can understand and use the invention well.

Claims

1. A textile jacquard detection method based on deep learning, characterized in that: The following steps are involved: S1. Setting a total defect type set; collecting textile images containing various jacquard patterns and covering various possible defects to obtain a defect image set; Annotating the defect image set to obtain a defect annotated image set; By using data enhancement technology, the defect image set is enhanced to obtain an enhanced defect image set; S2. Divide the enhanced defect image set into a defect image training set and a defect image test set, and set a defect image training label set and a defect image test label set accordingly; combine the CNN convolutional neural network with the RNN recurrent neural network to construct an initial CRNN convolutional recurrent neural network model; The initial CRNN convolutional recurrent neural network model is trained and optimized using the defect image training set and defect image training label set, the defect image test set and the defect image test label set to obtain the final CRNN convolutional recurrent neural network model; S3. Analyze the images of textiles collected in real time, and classify the images of textiles with defects.

2. The textile jacquard detection method based on deep learning according to claim 1, characterized in that: The S1 comprises the following steps: S11. Set the total defect type set a ={ a 1， a 2，...， a i，...， a r },in a i Defect type set a The i defects, r Defect type set a The Chinese Communist Party r types of defects; S12. Collect textile images containing various jacquard patterns and covering various possible defects to obtain a defect image set b ={ b 1, b 2,..., b i,..., b m },in b i Defect image set b The first i Defective images, m Defect image set b The Chinese Communist Party m defect images; S13. Check defect image set b For each image in the image, identify the defects in the image and classify them according to the total defect type set a , for the defect image set b Defect annotation is performed on each image in the image set to obtain the defect annotation image set b' ={ b' 1, b' 2,..., b' i,..., b' m },in b' i Defect image set b Middle i annotated defect images; S14. Defect labeling image set b' ={ b' 1, b' 2,..., b' i,..., b' m } to obtain the enhanced defect image set c ={ c 1, c 2,..., c i,..., c o },in c i Represents the enhanced defect image set c The i defect images, o Represents the enhanced defect image set c The Chinese Communist Party o defect image.

3. The textile jacquard detection method based on deep learning according to claim 1, characterized in that: The S2 comprises the following steps: S21. Divide the enhanced defect image set into a defect image training set d ={ d 1, d 2,..., d i,..., d p } and defect image test set e ={ e 1, e 2,..., e i..., e q }, according to the defect image training set and the defect image test set, a corresponding defect image training label set is set d' ={ d' 1, d' 2,..., d' i,..., d' p }、Defect image test label set e' ={ e' 1, e' 2,..., e' i..., e' q };in d i Defect image training set d Middle i defect images, p Defect image training set d The Chinese Communist Party p defect images, e i Denotes defect image test set e Middle i defect images, q Denotes defect image test set e The Chinese Communist Party q defect images, d' i Defect image training set d Middle i The defect types of defect images, e' i Denotes defect image test set e Middle i Defect types of defect images; S22, using the architecture combining CNN convolutional neural network and RNN recurrent neural network to build the initial CRNN convolutional recurrent neural network model; S23. Using defect image training set d and defect image training label set d′ Train the initial CRNN convolutional recurrent neural network model; after the training is completed, the trained CRNN convolutional recurrent neural network model is obtained, and then the defect image test set is e and defect image test label set e′ The trained CRNN convolutional recurrent neural network model is input for testing, and the trained CRNN convolutional recurrent neural network model is optimized according to the test results to obtain the final CRNN convolutional recurrent neural network model.

4. The textile jacquard detection method based on deep learning according to claim 3 is characterized in that: The S23 comprises the following steps: S231, constructing an initial CRNN convolutional recurrent neural network model, setting the number of iterations of the initial CRNN convolutional recurrent neural network model training to α, the batch size to β, the optimizer to adam optimizer, and the initial learning rate to ζ; S232, set the current training iteration number to z and the training error threshold is δ; according to the batch size β, the adam optimizer and the initial learning rate ζ, the defect image training set d and defect image training label set d′ Input into the initial CRNN convolutional recurrent neural network model for training, when the training error of the initial CRNN convolutional recurrent neural network model is <δ or when z When ≥α, stop training and obtain the trained CRNN convolutional recurrent neural network model; S233, setting an accuracy threshold λ, testing the defect image e and defect image test label set e′ Input into the trained CRNN convolutional recurrent neural network model for testing to obtain the test accuracy θ; when θ≥λ, use the trained CRNN convolutional recurrent neural network model as the final CRNN convolutional recurrent neural network model; when θ<λ, return to S232 to continue training the trained CRNN convolutional recurrent neural network model until θ≥λ, and obtain the final CRNN convolutional recurrent neural network model.

5. The textile jacquard detection method based on deep learning according to claim 4, characterized in that: The S232 comprises the following steps: S2321, the training error of the initial CRNN convolutional recurrent neural network model is calculated using a missing function, and the function expression is as follows: ； in, n is the sample size; y i It is i The true label of each sample, which takes the value 0 or 1; It is i The predicted probability of a sample indicates the probability that the model predicts that the sample belongs to category 1.

6. The textile jacquard detection method based on deep learning according to claim 4, characterized in that: The S233 comprises the following steps: S2331, Randomly initialize the position of the particle swarm h ={ h 1, h 2,..., h i,..., h NP },in h i Indicates i The learning rate value, Np is the number of particles; the speed of initializing particles v ={ v 1, v 2,..., v i,..., v NP },in v i Indicates i Initialize the optimal position of each particle f ={ f 1, f 2,..., f i,..., f NP },in f i Indicates i The best learning rate achieved by each particle; Initialize the global best position l for f Any position in; Set the maximum number of iterations to f' , the performance of the global best position reaches the target value f'' ; S2332. For each particle, update the speed v i t+1= j × v i t+ c 1× r 1×( f i t- h i )+ c 2× r 2×( l t- h i t ),in t represents the current time step, t +1 indicates the next time step, j is the inertia weight, c 1 and c 2 is the learning factor, r 1 and r 2 is a random number in the range of 0 to 1; Update Location h i t+1= h i t+ v i t+1 , use the updated learning rate to train the CRNN convolutional recurrent neural network model and evaluate its performance. If the new position h i t+1 The performance is better than f i t , then update f i , if the new position h i t+1 The performance is not better than f i t , keep the original f i ; At time step t +1, traverse the historical best positions of all particles f i , and select the position with the best performance as the new global best position l t+1 ; S2333, repeat S2332, when the performance of the global best position reaches the target value f′′ Or the number of iterations reaches f' When , the iteration process is stopped, and the learning rate value corresponding to the global best position is output to obtain the optimal learning rate f' ; S2334. Use the optimal learning rate as the learning rate of the trained CRNN convolutional recurrent neural network model to obtain an optimized CRNN convolutional recurrent neural network model.

7. The textile jacquard detection method based on deep learning according to claim 1, characterized in that: The S3 comprises the following steps: S31, deploy the final CRNN convolutional recurrent neural network model to dedicated image processing hardware or cloud server, and build a real-time jacquard detection system in combination with high-speed camera and image transmission system; S32. The images of the textiles captured by the system are transmitted to the final CRNN convolutional recurrent neural network model to classify the images of the defective textiles.

8. A system for a textile jacquard detection method based on deep learning according to any one of claims 1 to 7, characterized in that: Image set construction and enhancement module, CRNN convolutional recurrent neural network model training and optimization module, CRNN convolutional recurrent neural network model application module; The image set construction and enhancement module is used to collect textile images containing various jacquard patterns and covering various possible defects, and obtain an enhanced defect image set through annotation enhancement, which plays a role in expanding the data set and improving the generalization ability of subsequent models; The training and optimization module of the CRNN convolutional recurrent neural network model is used to train, test and optimize the initial CRNN convolutional recurrent neural network model to obtain the final CRNN convolutional recurrent neural network model; The CRNN convolutional recurrent neural network model application module is used to classify the real-time collected images containing jacquard patterns and textiles.

Images