Rice CT data generation method based on GAN network

Abstract

The invention belongs to the technical field of agricultural data processing, and particularly relates to a rice CT data generation method based on a GAN network, and the method comprises the following steps: S1, preprocessing rice seedlings; s2, two CT images of rice seedling stalks are collected; s3, carrying out standardization processing on the two acquired CT images; s4, constructing a rice CT data generation model in combination with a DCLB module and a 2DTo3D module; s5, training the generation model; and S6, inputting the two standardized CT images into the trained generation model, performing feature extraction and spatial information fusion on the input CT images, and generating rice CT image data of a middle section, thereby overcoming the defects in the prior art, effectively filling the blank between rice slice images, and improving the rice slice image quality. And reliable data support is provided for subsequent rice growth analysis and research.

Description

Technical Field

[0001] The present invention belongs to the technical field of agricultural data processing, and in particular relates to a method for generating rice CT data based on a GAN network. Background Art

[0002] Rice is one of the important crops in my country and the world. Especially in my country, the improvement of its yield and quality is related to the stable development of the economy. At present, the main method of rice planting in my country is machine transplanting, which is conducive to increasing yield and stabilizing output, while saving labor costs and alleviating the problem of labor shortage. However, the process of machine transplanting is prone to damage to the seedlings. For example, damage to the base of the stem may cause slow greening of the seedlings, prolong the growth period, and even cause seedling death, all of which seriously affect the yield and quality of rice.

[0003] In order to solve these problems, it is very important to study the biomechanical properties of rice seedlings. Generally, by establishing a biomechanical model that integrates multi-scale characteristic parameters of rice seedling stems, the mechanical behavior of rice stems can be accurately simulated, and the quantitative relationship between the degree of damage to rice stems and stress and strain can be obtained, breaking through the limitations of conventional experimental methods and contributing to the accurate quantitative evaluation of mechanical damage to seedling stems.

[0004] Obtaining a crop microstructure model is the basis for analyzing mechanical properties. Currently, scholars have used homogenization methods, representative volume unit methods, and multi-scale modeling based on X-ray computed tomography technology to reconstruct crop microstructure models and carry out research on the mechanical properties of biomaterials. The current research focuses on using modern technologies such as Micro-CT technology to obtain a three-dimensional microscopic physical model of rice seedling stems and reconstruct it through high-precision image segmentation technology.

[0005] This method can not only obtain the true geometric characteristics of the material without destroying the sample, but also help to accurately construct key technologies for growth, damage detection and variety selection, providing a scientific basis for rice cultivation. However, there are many varieties of rice in my country, and with the development of hybrid rice technology, the number of rice varieties has increased year by year. Each variety has multiple forms. If each form is scanned by Micro-CT technology to construct a three-dimensional microscopic physical model, the cost will be very high. Therefore, it is necessary to develop a method that can generate a three-dimensional microscopic physical model of a section of rice by analyzing the cross-section images at both ends of a small section of rice. With the continuous development of agricultural automation and intelligent technology, deep learning technology has gradually been applied to crop reconstruction, but most of them are reconstructions of the overall morphology of crops, and the microstructure cannot be accurately generated.

[0006] A network framework X2CT-GAN based on deep learning generative adversarial networks (GAN) has been used to reconstruct CT from lung X-ray images, but this method is applied in the medical field, and the conversion of two-dimensional features to three-dimensional features is not stable, resulting in the reconstructed CT having no clear boundaries and textures in the fine structure. Therefore, the present invention proposes a method for reconstructing CT using dual-plane cross-slice images to solve the problems existing in the prior art. Summary of the invention

[0007] The purpose of the present invention is to provide a rice CT data generation method based on a GAN network, which overcomes the shortcomings of the prior art, effectively fills the gaps between rice slice images, and provides reliable data support for subsequent rice growth analysis and research.

[0008] To solve the above problems, the technical solutions adopted by the present invention are as follows:

[0009] A method for generating rice CT data based on a GAN network comprises the following steps:

[0010] S1, pre-treating rice seedlings;

[0011] S2, collecting CT images of cross sections at both ends of the rice seedling stem;

[0012] S3, standardizing the two acquired CT images;

[0013] S4, build a rice CT data generation model with x2ct-gan as the backbone network and combine the DCLB module (Dilation Convolutional LayerBlock) and 2DTo3D module;

[0014] S5, training the generation model;

[0015] S6. Input the two standardized CT images into the trained generation model, perform feature extraction and spatial information fusion on the input CT images, and generate CT image data of the middle section of rice.

[0016] Furthermore, the rice seedlings described in S1 are rice that has achieved two-excellence in 40 days, and the pretreatment is to soak the rice seedlings in 15% cesium iodide solution.

[0017] Furthermore, the standardization process in S3 includes adjusting the resolution and grayscale value range of the image.

[0018] Furthermore, the generation model includes a 3D generator and a 3D discriminator, and the DCLB module is used in the encoder of the 3D generator to extract features; the 2DTo3D module is used to convert the two-dimensional features into three-dimensional features.

[0019] Furthermore, the DCLB module uses four dilated convolution kernels with different dilation rates (r=1, r=2, r=3, r=4) to process input features in parallel, capture spatial features through multiple scale receptive fields, and enhance the modeling capability of the fine structure in rice CT data. The module operates through the following steps:

[0020] (1) The input features are first reduced in size through a downsampling module, which halves the width and height of the original features, thereby achieving feature compression;

[0021] (2) Then, a convolutional layer is used to reduce the number of channels to 1 / 4 of the original number of channels to reduce the computational complexity;

[0022] (3) The reduced features are convolved in parallel through four dilated convolutional layers, using different dilation rates r = 1, r = 2, r = 3, and r = 4 to capture spatial features of different scales;

[0023] (4) The outputs of the four dilated convolutional layers are added element by element to fuse multi-scale information;

[0024] (5) Then, the fused features are restored to the original number of channels through a full-channel convolutional layer, and then the output feature map after the features before dimensionality reduction in step (2) is added.

[0025] (6) Repeat steps (2) to (5) N times and output the feature map.

[0026] (7) Finally, the feature map is concatenated with the downsampled features in step (1) in the dimension of the number of channels to generate the final feature map.

[0027] Furthermore, the 2DTo3D module is used to expand the two-dimensional features to three-dimensional features, and the specific steps are as follows:

[0028] (1) Two-dimensional feature broadcast expansion: The input two-dimensional features are expanded along the third dimension (depth direction) through a broadcast mechanism to generate a three-dimensional feature map with four layers of depth.

[0029] (2) Concatenation operation: Concatenate the expanded two-dimensional features with the three-dimensional features after downsampling so that the spatial information of the two-dimensional features and the three-dimensional features remain consistent during the information fusion process.

[0030] (3) Three-dimensional convolution processing: The spliced ​​features are processed through multi-layer three-dimensional convolution operations to further extract structural information in the three-dimensional space and improve the model's ability to model three-dimensional data.

[0031] Further, in S5, the generation model is trained, and the specific method includes:

[0032] S51, collecting the pre-processed CT images of the rice seedling stems;

[0033] S52, creating a CT image data set from the collected CT images of the rice seedling stems;

[0034] S53, dividing the CT image data set into a training set and a test set, using the training set to input the generative model for data modeling, and using the test set to test the generative model.

[0035] Further, the collected CT images of the rice seedling stems are converted into a CT image data set in S52, and the specific steps are as follows:

[0036] (1) Performing center cropping of different sizes on the collected rice seedling stem CT images, and grouping images cropped with the same size into the same batch;

[0037] (2) Scale the cropped image to 256*256;

[0038] (3) In the same batch, 256 images are stacked in sequence to form three-dimensional CT image data. After the images are rotated, a CT image dataset consisting of 1180 images is obtained.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] The present invention discloses a rice CT data generation method based on a GAN network, which constructs a rice CT data generation model by using an x2ct-gan network as the backbone network and combining a DCLB module and a 2DTo3D module.

[0041] Through the encoder-decoder structure and the DCLB module, the input slice image is subjected to feature extraction and spatial information fusion. The 2DTo3D module expands the two-dimensional features into three-dimensional features through the broadcast mechanism, and then concatenates them with the downsampled three-dimensional features. The spatial structure information is further extracted through multi-layer three-dimensional convolution, thereby enhancing the model's modeling ability for three-dimensional data and generating rice CT image data of the middle segment, which can accurately reflect the internal structure details between the two slice images. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is a CT image of rice seedling stems in a rice CT data generation method based on a GAN network.

[0043] Figure 2 This is the network model architecture of the generative model in a rice CT data generation method based on a GAN network.

[0044] Figure 3This is the rice CT image data generated by the model in a rice CT data generation method based on a GAN network. DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] The present invention discloses a method for generating rice CT data based on a GAN network. In this embodiment, the method works in an Ubuntu 22.04.2 operating system, and the server is configured with 2*Intel XRONGold 6133 CPU @ 2.5 GHz processors, 4*NVIDIA GeForce RTX 3090 24g TURBO graphics cards, 8*Samsung 32G DDR4 / 2933recc memory, Pycharm 2023 development tools, Pytorch development framework, and Python 3.6 programming language.

[0047] The specific steps are as follows:

[0048] S1. Pre-treating rice seedlings.

[0049] The rice seedlings are 40-day-old rice with excellent quality, and the pretreatment is to immerse the rice seedlings in 15% cesium iodide solution.

[0050] S2. Collect two CT images of rice seedling stems.

[0051] The images were acquired by SkyScan1272 scanner at an acceleration voltage of 50 kV, a current of 80 uA and a detector exposure time of 2800 ms.

[0052] Figure 1 These are CT images of rice seedling stems collected.

[0053] S3. Standardize the two acquired CT images.

[0054] The two rice slice images were normalized to ensure that their size and format were consistent with the model requirements. Normalization included adjusting the image resolution and grayscale value range to meet the model input specifications.

[0055] S4, build a rice CT data generation model with x2ct-gan as the backbone network and combine the DCLB module (Dilation Convolutional LayerBlock) and 2DTo3D module;

[0056] The generation model includes a 3D generator and a 3D discriminator. The DCLB module is used in the 3D generator for feature extraction; the 2DTo3D module is used to convert two-dimensional features into three-dimensional features.

[0057] The DCLB module uses four dilated convolution kernels with different dilation rates (r=1, r=2, r=3, r=4) to process input features in parallel, capture spatial features through multiple scale receptive fields, and enhance the modeling ability of fine structures in rice CT data. The module operates through the following steps:

[0058] (1) The input features are first reduced in size through a downsampling module, which halves the width and height of the original features, thereby achieving feature compression;

[0059] (2) Then, a convolutional layer is used to reduce the number of channels to 1 / 4 of the original number of channels to reduce the computational complexity;

[0060] (3) The reduced features are convolved in parallel through four dilated convolutional layers, using different dilation rates r = 1, r = 2, r = 3, and r = 4 to capture spatial features of different scales;

[0061] (4) The outputs of the four dilated convolutional layers are added element by element to fuse multi-scale information;

[0062] (5) Then, the fused features are restored to the original number of channels through a full-channel convolutional layer, and then the output feature map after the features before dimensionality reduction in step (2) is added.

[0063] (6) Repeat steps (2) to (5) N times and output the feature map.

[0064] (7) Finally, the feature map is concatenated with the downsampled features in step (1) in the dimension of the number of channels to generate the final feature map.

[0065] The 2DTo3D module is used to expand two-dimensional features to three-dimensional features. The specific steps are as follows

[0066] (1) Two-dimensional feature broadcast expansion: The input two-dimensional features are expanded along the third dimension (depth direction) through a broadcast mechanism to generate a three-dimensional feature map with four layers of depth.

[0067] (2) Concatenation operation: Concatenate the expanded two-dimensional features with the three-dimensional features after downsampling so that the spatial information of the two-dimensional features and the three-dimensional features remain consistent during the information fusion process.

[0068] (3) Three-dimensional convolution processing: The spliced ​​features are processed through multi-layer three-dimensional convolution operations to further extract structural information in the three-dimensional space and improve the model's ability to model three-dimensional data.

[0069] Figure 2 The network model architecture for the generative model.

[0070] S5. Training the generation model.

[0071] Specific methods include:

[0072] S51, collecting the pre-processed CT images of the rice seedling stems;

[0073] S52, creating a CT image data set from the collected CT images of the rice seedling stems;

[0074] S53, dividing the CT image data set into a training set and a test set, using the training set to input the generative model for data modeling, and using the test set to test the generative model.

[0075] The best effect is obtained by adjusting the parameters. The following table shows the experimental results of different models under the same parameters.

[0076] Model CosineSimilarity↑ PSNR↑ SSIM↑ x2ct-gan+S 0.9109536 49.2100563 0.6921427 x2ct-gan+B 0.9777877 53.7586789 0.7880505 DAGct-gan 0.9809820 54.7150993 0.8211958

[0077] Note: +s is single view; +B is dual view; DAGct-gan is the name of the improved model CosineSimilarity: cosine similarity, PSNR: peak signal-to-noise ratio, SSIM: structural similarity,

[0078] S6. Input the two standardized CT images into the trained generation model, perform feature extraction and spatial information fusion on the input CT images, and generate CT image data of the middle section of rice.

[0079] The two standardized rice slice images were input into the rice CT data generation model. The 3D generator of the model used an encoder-decoder structure and a DCLB module to extract features and fuse spatial information from the input slice images. The 2DTo3D module was used to convert the two-dimensional features into three-dimensional features, and finally the rice CT image data of the middle section was generated. The CT data generated by the model has high precision and quality, and can accurately reflect the internal structure details between the two slice images. During the generation process, the model continuously optimized parameters and compared experimental results, and finally achieved the best generation effect.

[0080] Figure 3 The rice CT image data generated by the model is shown, and it can be seen that the generated CT image data has good effects in both details and overall structure. In this way, the present invention can effectively fill the gaps between rice slice images and provide reliable data support for subsequent rice growth analysis and research.

[0081] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A method for generating rice CT data based on a GAN network, characterized in that: The following steps are involved: S1, pre-treating rice seedlings; S2, collecting two CT images of rice seedling stems; S3, standardizing the two acquired CT images; S4, construct a rice CT data generation model based on x2ct-gan as the backbone network, combined with the DCLB module and 2DTo3D module; S5, training the generation model; S6. Input the two standardized CT images into the trained generation model, perform feature extraction and spatial information fusion on the input CT images, and generate CT image data of the middle section of rice.

2. The method for generating rice CT data based on a GAN network according to claim 1, characterized in that: The rice seedlings described in S1 are 40-day-old rice with excellent quality, and the pretreatment is to soak the rice seedlings in 15% cesium iodide solution.

3. The method for generating rice CT data based on a GAN network according to claim 1, characterized in that: The standardization process in S3 includes adjusting the resolution and grayscale value range of the image.

4. The method for generating rice CT data based on a GAN network according to claim 1, characterized in that: The generation model includes a 3D generator and a 3D discriminator, and a DCLB module is used in the encoder of the 3D generator to extract features; a 2DTo3D module is used to convert two-dimensional features into three-dimensional features.

5. The method for generating rice CT data based on a GAN network according to claim 4, characterized in that: The DCLB module uses four dilated convolution kernels with different dilation rates, namely r=1, r=2, r=3, and r=4, to process input features in parallel, capture spatial features through multiple scale receptive fields, and enhance the modeling capability of fine structures in rice CT data; The specific operation steps of the DCLB module are: (1) The input features are first reduced in size through a downsampling module, which reduces the width and height of the original features by half for feature compression; (2) Perform dimensionality reduction through a convolutional layer to reduce the number of channels to 1 / 4 of the original number of channels to reduce computational complexity; (3) The reduced features are convolved in parallel through four dilated convolutional layers, using different dilation rates r = 1, r = 2, r = 3, and r = 4 to capture spatial features of different scales; (4) The outputs of the four dilated convolutional layers are added element by element to fuse multi-scale information; (5) Restore the fused features to the original number of channels through a full-channel convolution layer, and add the output feature map after the features before dimensionality reduction in step (2); (6) Repeat steps (2) to (5) N times and output the feature map; (7) The feature map is concatenated with the features downsampled in step (1) in the dimension of the number of channels to generate the final feature map.

6. The method for generating rice CT data based on a GAN network according to claim 1, characterized in that: The DCLB module can finely extract the texture, edges and complex internal structure in rice CT data through its multi-scale feature extraction mechanism, improve the generation model's ability to restore the microstructure of rice stems, and ensure that the generated CT images have high resolution and detail accuracy.

7. The method for generating rice CT data based on a GAN network according to claim 5, characterized in that: The 2DTo3D module is used to expand two-dimensional features to three-dimensional features. The specific steps are as follows: (1) 2D feature broadcast expansion: The input 2D features are expanded along the third dimension through a broadcast mechanism to generate a 3D feature map with four layers of depth; (2) Splicing operation: Splicing the expanded 2D features with the downsampled 3D features so that the spatial information of the 2D features and the 3D features remain consistent during the information fusion process; (3) Three-dimensional convolution processing: The spliced ​​features are processed through multi-layer three-dimensional convolution operations to further extract structural information in the three-dimensional space and improve the model's ability to model three-dimensional data.

8. The method for generating rice CT data based on a GAN network according to claim 1, characterized in that: In S5, the generation model is trained, and the specific method includes: S51, collecting the pre-processed CT images of the rice seedling stems; S52, creating a CT image data set from the collected CT images of the rice seedling stems; S53, dividing the CT image data set into a training set and a test set, using the training set to input the generative model for data modeling, and using the test set to test the generative model.

9. The method for generating rice CT data based on a GAN network according to claim 8, characterized in that: S52 is to prepare a CT image data set from the collected CT images of rice seedling stems, and the specific steps are as follows: (1) Performing center cropping of different sizes on the collected rice seedling stem CT images, and grouping images cropped with the same size into the same batch; (2) Scale the cropped image to 256*256; (3) In the same batch, 256 images are stacked in sequence to form three-dimensional CT image data. After the images are rotated, a CT image dataset consisting of 1180 images is obtained.

Images