Texture image classification method based on prior knowledge

Abstract

The invention provides a priori knowledge-based texture image classification method. The method comprises the following steps of: preprocessing an image; dividing the image data set into a training set and a test set; training the priori knowledge-based texture image classification network to obtain a texture image classification network; inputting the obtained original texture image into a trained texture image classification network based on priori knowledge to output a classification result; wherein the prior knowledge-based texture image classification network comprises a residual Gabor convolution layer module, a Gabor direction filter module, a Gabor direction filter module and a Gabor direction filter module, and the residual Gabor convolution layer module is used for extracting and enhancing shallow layer features and comprises a layer of convolution and a layer of Gabor direction filter; the residual module is used for performing depth feature extraction on the enhanced shallow features; the DASM module is used for extracting a group of enhanced semantic information features; the Concat operation is used for fusing the enhanced semantic information features and the deep features; and the Softmax classifier is used for carrying out classification output.

Description

Technical Field

[0001] The invention belongs to the field of image processing, and in particular relates to a texture image classification method based on prior knowledge. Background Art

[0002] Texture is a visual feature that responds to homogeneous phenomena independently of color or brightness changes. It is a basic property of the surface of an object and is widely present in nature. Texture features contain important information about the structural organization and arrangement of the surface of an object and their connection with the surrounding environment. It is an important visual clue that conforms to human visual characteristics in describing and identifying objects, and has always been a focus of computer vision research.

[0003] Texture analysis has important applications in many fields, such as healthcare, aerospace manufacturing, and military. In image recognition and target detection, complex texture features can help algorithms better understand the content of images. These features can be used for tasks such as image classification, object recognition, and face recognition. In geological research, complex texture feature analysis can help identify geological structures and stratigraphic features. By analyzing the texture features in geological images, rock types, the arrangement and changes of stratigraphic layers can be determined, which is of great significance for geological surveys and resource exploration. In materials science, complex texture feature analysis can help researchers understand the structure and properties of materials. By analyzing the texture features of materials, the mechanical properties, thermal conductivity, electrical conductivity and other properties of materials can be predicted, thereby guiding material design and optimization. In the military field, complex texture image analysis can be used to automatically detect and identify enemy targets or places of interest in military intelligence collection, reconnaissance, and surveillance. By analyzing the texture features in images, such as the texture of terrain, buildings, vehicles, personnel, etc., it can help military personnel quickly detect and identify potential threats or targets. In the medical field, complex texture feature analysis is very important for tasks such as lesion detection, tissue classification, and disease diagnosis. For example, by analyzing the texture features of tissues in medical images, it can help doctors diagnose diseases such as cancer and tumors.

[0004] Promoting innovative research in the fields of image graphics, learning, reasoning, and decision-making in the new generation of artificial intelligence technology is a development goal for a long time to come. However, the current analysis of texture features based on deep learning algorithms (DCNN) has not yet broken away from the statistical learning model driven by massive data, and lacks a reasonable explanation and effective prediction of processing results, which limits the development space in the field.

[0005] Specifically, when DCNN is applied to finger vein recognition, since each category in the finger vein contains only a small number of samples, and DCNN requires a large amount of data to learn more abstract semantic information for classification, overfitting is very likely to occur. The reason is that the size of the receptive field is fixed at each position in the convolutional layer, and there is a lack of intrinsic mechanism for processing geometric transformations. Therefore, due to the limitation of insufficient finger vein training samples, DCNN feature learning is insufficient and the model generalization ability is weak.

[0006] In order to solve the above-mentioned technical difficulties, a texture image classification method based on small sample learning is urgently needed. Summary of the invention

[0007] In order to solve the above problems, it is necessary to provide a texture image classification method based on prior knowledge.

[0008] The first aspect of the present invention provides a texture image classification method based on prior knowledge, comprising the following method steps: Training a texture image classification network based on prior knowledge S1, preprocessing the texture image; S2, dividing the texture image dataset into a training set and a test set; S3, constructing a texture image classification network based on prior knowledge; using a training set and a test set to train the texture image classification network based on prior knowledge to obtain a texture image classification network; Wherein, the texture image classification network based on prior knowledge includes: A residual Gabor convolutional layer module for extracting and enhancing shallow features. This module includes a layer of convolution and a layer of Gabor directional filter. A residual module for deep feature extraction from enhanced shallow features; Dense Semantic Analysis (DASM) module for extracting a set of features that enhance semantic information; Concat operation used to fuse enhanced semantic information features with deep features; And a Softmax classifier for classification output; Texture Image Classification S4. Input the acquired original texture image into the trained texture image classification network based on prior knowledge to output the classification result.

[0009] A second aspect of the present invention provides a system based on the texture image classification method based on prior knowledge, the system comprising: Model training module, used to train texture image classification network based on prior knowledge; The step of training a texture image classification network based on prior knowledge includes: S1, preprocessing the texture image; S2, dividing the texture image dataset into a training set and a test set; S3, constructing a texture image classification network based on prior knowledge; using a training set and a test set to train the texture image classification network based on prior knowledge to obtain a texture image classification network; Wherein, the texture image classification network based on prior knowledge includes: A residual Gabor convolutional layer module for extracting and enhancing shallow features. This module includes a layer of convolution and a layer of Gabor directional filter. A residual module for deep feature extraction from enhanced shallow features; Dense Semantic Analysis (DASM) module for extracting a set of features that enhance semantic information; Concat operation used to fuse enhanced semantic information features with deep features; And a Softmax classifier for classification output; The texture image classification module is used to input the acquired original texture image into the trained texture image classification network based on prior knowledge, and output the recognition result.

[0010] The third aspect of the present invention proposes a terminal, including a processor, a memory, and a texture image classification method program based on prior knowledge stored in the memory. When the texture image classification method program based on prior knowledge is executed by the processor, the steps of the texture image classification method based on prior knowledge are implemented.

[0011] Compared with the prior art, the present invention has outstanding substantive features and significant progress, specifically: 1. The present invention trains a texture image classification network based on prior knowledge, and only requires simple image preprocessing operations, without the need to design a complex feature extraction algorithm, so that the image can be input into the network for classification and the classification result can be output; 2. The present invention adopts a residual mechanism to ensure that image features are not lost while being transmitted to the deep network, thereby obtaining higher-dimensional and abstract features of the image. At the same time, the residual Gabor convolutional layer module in the texture image classification network based on prior knowledge effectively enhances the angle and scale information in the shallow features of the network, and the dense semantic analysis module further fully extracts and analyzes the enhanced deep feature information, enhances the feature extraction and representation capabilities of the network, and can accelerate network training and improve classification accuracy.

[0012] 3. The present invention proposes a feature offline enhancement strategy FV-MIX, which linearly fuses the preprocessed features of a single individual in proportion, and finally obtains 2^n-1 features for training, thereby expanding the number of samples.

[0013] Additional aspects and advantages of the present invention will become apparent from the following description or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which: Figure 1 This is a flow chart of a texture image classification method based on prior knowledge provided by an embodiment of the present invention.

[0015] Figure 2 This is an original texture image sample provided by an embodiment of the present invention.

[0016] Figure 3 This is an image preprocessing effect diagram provided by an embodiment of the present invention.

[0017] Figure 4 It is a network structure diagram provided by an embodiment of the present invention.

[0018] Figure 5 This is a classification result diagram provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0019] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0020] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0021] Example 1 This embodiment provides a texture image classification method based on prior knowledge, such as Figure 1 As shown, the method comprises the following steps: Training a texture image classification network based on prior knowledge S1. Preprocess the texture image to obtain a texture image data set.

[0022] S2. Divide the texture image dataset into a training set, a test set, and a validation set; preferably, the images in each category of the texture image dataset are divided into a training set and a test set in a ratio of 2:1. In some embodiments, the training set can be further divided into a training set and a validation set in a ratio of 9:1.

[0023] S31, constructing a texture image classification network based on prior knowledge; specifically, the texture image classification network based on prior knowledge includes: A residual Gabor convolutional layer module for extracting and enhancing shallow features. This module includes a layer of convolution and a layer of Gabor directional filter. A residual module for deep feature extraction from enhanced shallow features; Dense Semantic Analysis (DASM) module for extracting a set of features that enhance semantic information; Concat operation used to fuse enhanced semantic information features with deep features; And a Softmax classifier for categorical output.

[0024] Specifically, the processing steps of the residual Gabor convolutional layer module include: Step 1: Send the texture image into the residual network to extract shallow features; Step 2, using a convolutional Gabor directional filter to enhance the shallow features, wherein the convolutional Gabor directional filter is obtained by modulating the original convolutional filter with a group of Gabor filters in different directions; Step 3: Use channel attention to fuse the shallow features obtained by the residual network with the shallow features enhanced by the convolutional Gabor directional filter; Step 4: Repeat steps 1 to 3 for U times; Step 5: Connect the U groups of fusion features obtained above to obtain the final enhanced features.

[0025] Specifically, the processing steps of the dense semantic analysis DASM module are: Step 6: Use global adaptive average pooling to process the deep features; Step 7: Use channel attention to process the deep features obtained in step 6 to obtain enhanced semantic information features; Step 8. Repeat steps 6 to 7 n times; Step nine: concatenate the above-obtained features to obtain a set of enhanced semantic information features.

[0026] S32, using a training set and a test set to train the texture image classification network based on prior knowledge to obtain a texture image classification network; specifically, using the training set to repeatedly iteratively train the above network until the network converges, and using the test set to verify the recognition accuracy of the final network.

[0027] Texture Image Classification S4. Input the acquired original texture image into the trained texture image classification network based on prior knowledge to output the classification result.

[0028] It can be seen that the present invention proposes a feature offline enhancement strategy FV-MIX, which linearly fuses the preprocessed features of a single individual in proportion, and finally obtains 2^n-1 features for training, thereby expanding the number of samples.

[0029] Moreover, the present invention can ensure that the image features will not be lost while being transmitted to the deep network by adopting the residual mechanism, thereby obtaining higher-dimensional and abstract features of the image. At the same time, the residual Gabor convolutional layer module in the texture image classification network based on prior knowledge effectively enhances the angle and scale information in the shallow features of the network, and the dense semantic analysis module further fully extracts and analyzes the enhanced deep feature information, enhances the feature extraction and representation capabilities of the network, and can accelerate network training and improve classification accuracy.

[0030] There is no need to design complex feature extraction algorithms. Only simple image preprocessing is required and then the image is input into a trained texture image classification network based on prior knowledge. The image can then be input into the network for classification and the classification results can be output. It is simple and easy to operate.

[0031] Example 2 In order to verify the effectiveness of a texture image classification method based on prior knowledge described in Example 1, this example conducts experiments on the finger vein database FV-USM created by the Bakhtiar team of Universiti Sains Malaysia, and compares the classification performance of the method of the present invention (rgcl) with seven existing texture image classification methods: Baseline, FV-MIX, DSAM, DSAM+ FV-MIX, RGCL, RGCL+ FV-MIX, RGCL+ DSAM. The evaluation indicators are training loss and verification loss. The higher the training loss and verification loss, the faster the convergence speed. The finger vein database FV-USM contains a total of 492 finger vein images of different types. Six images of each finger are collected in two stages. The original image size is , "jpg" format, such as Figure 2 shown.

[0032] The steps of a texture image classification method based on prior knowledge described in Example 1 are as follows: S1. Figure 2 The texture image samples in each category of the dataset shown in the figure are preprocessed, that is, the ROI area is determined, the ROI of each image is obtained and saved, and the results are shown in Figure 3 shown.

[0033] S2. Divide the processed data set into a training set and a test set, where the 6 images in each category of the data set are divided into a training set and a test set in a ratio of 2:1. The network is continuously iterated through the training set until the network converges to the optimal recognition result, and the performance of the final network is evaluated through the test set.

[0034] S3. Construct a texture image classification network based on prior knowledge, the texture image classification network based on prior knowledge includes: a residual Gabor convolution layer module (RGCL) for extracting and enhancing shallow features, the module includes a layer of convolution and a layer of Gabor directional filters (GoFs); a residual module for extracting deep features from the enhanced shallow features; a dense semantic analysis DASM module for extracting a set of enhanced semantic information features; a Concat operation (FV-MIX) for fusing enhanced semantic information features with deep features; and a Softmax classifier for classification output. The texture image classification network based on prior knowledge is trained using a training set and a test set to obtain a texture image classification network (RGCL+ DSAM+ FV-MIX).

[0035] like Figure 4 As shown in Figure 2, in order to ensure the adaptability of the convolution kernels of each layer of the network to the image and control the parameter scale of the network, the images in the training set and the test set are first resized to a unified input size of ; Then, the texture image in the training set or test set is sent to the residual Gabor convolution layer module in the proposed texture image classification network based on prior knowledge, the shallow features in the network are extracted and the angle and scale information in the shallow layer features are enhanced using a convolutional Gabor directional filter; After that, the enhanced shallow features are successively sent to the subsequent residual modules to extract the deep features in the image; Among them, in the third Bottleneck structure, the features are simultaneously sent to the dense semantic analysis DASM module to obtain a set of features rich in enhanced semantic information; Finally, the obtained set of features rich in enhanced semantic information is connected to the final fully connected layer to assist the final model in classifying the texture image.

[0036] Back propagation is set to use the Adam optimizer, with an initial learning rate of 0.0001. When the number of training times increases by 20% of the total number of times, the learning rate is reduced to 10% of the original. Finally, the Softmax classifier is used to output the classification results. Figure 5As shown, it can be seen that the final classification accuracy of the RGCL+DSAM+FV-MIX network model adopted in this application is 100%.

[0037] Example 3 Based on the same inventive concept as the method described in Example 1, this embodiment proposes a system based on the texture image classification method based on prior knowledge, the system comprising: Model training module, used to train texture image classification network based on prior knowledge The texture image classification module is used to input the acquired original texture image into the trained texture image classification network based on prior knowledge, and output the recognition result.

[0038] Specifically, the steps of training the texture image classification network based on prior knowledge include: S1, preprocessing the texture image; S2, dividing the texture image dataset into a training set and a test set; S3, constructing a texture image classification network based on prior knowledge; using a training set and a test set to train the texture image classification network based on prior knowledge to obtain a texture image classification network; Wherein, the texture image classification network based on prior knowledge includes: A residual Gabor convolutional layer module for extracting and enhancing shallow features. This module includes a layer of convolution and a layer of Gabor directional filter. A residual module for deep feature extraction from enhanced shallow features; Dense Semantic Analysis (DASM) module for extracting a set of features that enhance semantic information; Concat operation used to fuse enhanced semantic information features with deep features; And a Softmax classifier for categorical output.

[0039] In a specific implementation, the processing steps of the residual Gabor convolutional layer module for extracting and enhancing shallow features include: Step 1: Send the texture image into the residual network to extract shallow features; Step 2, using a convolutional Gabor directional filter to enhance the shallow features, wherein the convolutional Gabor directional filter is obtained by modulating the original convolutional filter with a group of Gabor filters in different directions; Step 3: Use channel attention to fuse the shallow features obtained by the residual network with the shallow features enhanced by the convolutional Gabor directional filter; Step 4: Repeat steps 1 to 3 for U times; Step 5: Connect the U groups of fusion features obtained above to obtain the final enhanced features.

[0040] In a specific implementation, the processing steps of the dense semantic analysis DASM module for extracting a set of enhanced semantic information features include: Step 6: Use global adaptive average pooling to process the deep features; Step 7: Use channel attention to process the deep features obtained in step 6 to obtain enhanced semantic information features; Step 8. Repeat steps 6 to 7 n times; Step nine: concatenate the above-obtained features to obtain a set of enhanced semantic information features.

[0041] Example 4 Based on the same inventive concept as the method described in Example 1, this embodiment proposes a terminal, including a processor, a memory, and a texture image classification method program based on prior knowledge stored in the memory. When the texture image classification method program based on prior knowledge is executed by the processor, the steps of the texture image classification method based on prior knowledge as described in Example 1 are implemented.

[0042] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0043] Each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0044] If the above-mentioned integrated module is implemented in the form of a software functional unit and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The above-mentioned computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the above-mentioned computer program includes computer program code, and the above-mentioned computer program code can be in source code form, object code form, executable file or some intermediate form, etc.

[0045] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A texture image classification method based on prior knowledge, characterized in that: The method comprises the following steps: Training a texture image classification network based on prior knowledge S1, preprocessing the texture image to obtain a texture image data set; S2, dividing the texture image dataset into training set, test set and validation set; S3, constructing a texture image classification network based on prior knowledge; training the texture image classification network based on prior knowledge using a training set, a test set, and a validation set to obtain a texture image classification network; Wherein, the texture image classification network based on prior knowledge includes: A residual Gabor convolutional layer module for extracting and enhancing shallow features. This module includes a layer of convolution and a layer of Gabor directional filter. A residual module for deep feature extraction from enhanced shallow features; Dense Semantic Analysis (DASM) module for extracting a set of features that enhance semantic information; Concat operation used to fuse enhanced semantic information features with deep features; And a Softmax classifier for classification output; Texture Image Classification S4. Input the acquired original texture image into the trained texture image classification network based on prior knowledge to output the classification result.

2. A texture image classification method based on prior knowledge according to claim 1, characterized in that: The processing steps of the residual Gabor convolutional layer module for extracting and enhancing shallow features include: Step 1: Send the texture image into the residual network to extract shallow features; Step 2, using a convolutional Gabor directional filter to enhance the shallow features, wherein the convolutional Gabor directional filter is obtained by modulating the original convolutional filter with a group of Gabor filters in different directions; Step 3: Use channel attention to fuse the shallow features obtained by the residual network with the shallow features enhanced by the convolutional Gabor directional filter; Step 4: Repeat steps 1 to 3 for U times; Step 5: Connect the U groups of fusion features obtained above to obtain the final enhanced features.

3. The texture image classification method based on prior knowledge according to claim 2 is characterized in that: In step 32, the original convolution filter is modulated with a group of Gabor filters in different directions to obtain a convolution Gabor directional filter.

4. The texture image classification method based on prior knowledge according to claim 1 is characterized in that: The processing steps of the dense semantic analysis DASM module for extracting a set of features that enhance semantic information include: Step 6: Use global adaptive average pooling to process the deep features; Step 7: Use channel attention to process the deep features obtained in step 6 to obtain enhanced semantic information features; Step 8. Repeat steps 6 to 7 n times; Step nine: concatenate the above-obtained features to obtain a set of enhanced semantic information features.

5. The texture image classification method based on prior knowledge according to claim 1, characterized in that: In step S2, after the images in each category of the data set are divided into a training set and a test set in a ratio of 2:1, the training set is further divided into a training set and a validation set in a ratio of 9:

1.

6. A texture image classification system based on prior knowledge, characterized in that: include: Model training module, used to train texture image classification network based on prior knowledge; The step of training a texture image classification network based on prior knowledge includes: S1, preprocessing the texture image; S2, dividing the texture image dataset into a training set and a test set; S3, constructing a texture image classification network based on prior knowledge; using a training set and a test set to train the texture image classification network based on prior knowledge to obtain a texture image classification network; Wherein, the texture image classification network based on prior knowledge includes: A residual Gabor convolutional layer module for extracting and enhancing shallow features of texture images, which includes a layer of convolution and a layer of Gabor directional filter; A residual module for deep feature extraction from enhanced shallow features; Dense semantic analysis (DASM) module for extracting enhanced semantic information features from deep features; Concat operation used to fuse enhanced semantic information features with deep features; And a Softmax classifier for classification output; The texture image classification module is used to input the acquired original texture image into the trained texture image classification network based on prior knowledge, and output the recognition result.

7. The texture image classification system based on prior knowledge according to claim 6, characterized in that: The processing steps of the residual Gabor convolutional layer module for extracting and enhancing shallow features include: Step 1: Send the texture image into the residual network to extract shallow features; Step 2, using a convolutional Gabor directional filter to enhance the shallow features, wherein the convolutional Gabor directional filter is obtained by modulating the original convolutional filter with a group of Gabor filters in different directions; Step 3: Use channel attention to fuse the shallow features obtained by the residual network with the shallow features enhanced by the convolutional Gabor directional filter; Step 4: Repeat steps 1 to 3 for U times; Step 5: Connect the U groups of fusion features obtained above to obtain the final enhanced features.

8. The texture image classification system based on prior knowledge according to claim 7, characterized in that: In step 32, the original convolution filter is modulated with a group of Gabor filters in different directions to obtain a convolution Gabor directional filter.

9. The texture image classification system based on prior knowledge according to claim 6, characterized in that: The processing steps of the dense semantic analysis DASM module for extracting a set of features that enhance semantic information include: Step 6: Use global adaptive average pooling to process the deep features; Step 7: Use channel attention to process the deep features obtained in step 6 to obtain enhanced semantic information features; Step 8. Repeat steps 6 to 7 n times; Step nine: concatenate the above-obtained features to obtain a set of enhanced semantic information features.

10. A terminal, characterized in that: The invention comprises a processor, a memory and a texture image classification method program based on prior knowledge stored in the memory. When the texture image classification method program based on prior knowledge is executed by the processor, the steps of the texture image classification method based on prior knowledge as described in any one of claims 1 to 5 are implemented.

Images