A durian CT image processing method, system, storage medium and computer
By constructing a super-resolution convolutional neural network to reconstruct durian CT images, the problem of inaccurate CT image detection was solved, and high-precision detection of internal defects in durians was achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ZHEJIANG BOSHI NETWORK TECHNOLOGY CO LTD
- Filing Date
- 2026-03-23
- Publication Date
- 2026-07-31
AI Technical Summary
In existing technologies, CT images are difficult to accurately identify internal defects in durians, leading to inaccurate detection.
A super-resolution convolutional neural network was used to reconstruct images of durian from low-resolution and high-resolution CT images. High-resolution CT images were generated by constructing a training dataset and training the super-resolution convolutional neural network.
It improves the accuracy of detecting internal defects in durians, obtains higher-resolution CT images, and can more accurately identify internal defects in durians.
Smart Images
Figure CN122492452A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of fruit and vegetable quality testing technology, specifically to a durian CT image processing method, system, storage medium, and computer. Background Technology
[0002] The appearance quality of fruits and vegetables such as durian is usually inspected manually, while the detection of internal defects often uses computed tomography (CT) scans to obtain CT images, which are then used to detect internal defects in durians. However, the clarity of CT images is difficult to compare with that of high-definition images captured by cameras, resulting in inaccurate identification of internal defects in durians when using CT images for detection. Summary of the Invention
[0003] To address the technical problem in existing technologies where CT images cannot accurately identify internal defects in durians, this invention provides a durian CT image processing method, system, storage medium, and computer.
[0004] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: A method for processing durian CT images includes the following steps: Acquired low-resolution and high-resolution CT images of durian; A training dataset was constructed using the low-resolution CT images and the high-resolution CT images; Constructing a super-resolution convolutional neural network; The super-resolution convolutional neural network is trained using the training dataset to obtain the image reconstruction network; Acquire CT images of durians to be processed; The CT image to be processed is input into the image reconstruction network to obtain a high-resolution CT image.
[0005] The beneficial effects of this invention are: by using a super-resolution convolutional neural network to reconstruct the CT image of durian, a high-resolution CT image can be obtained, which facilitates the detection of internal defects in durian using the high-resolution CT image, thereby improving the accuracy of identifying internal defects in durian.
[0006] Based on the above technical solution, the present invention can be further improved as follows.
[0007] Furthermore, low-resolution and high-resolution CT images of durian were acquired, including the following steps: Use a camera to take a cross-sectional image of the durian; The cross-sectional image is processed to obtain the high-resolution CT image; The high-resolution CT image is downsampled to obtain the low-resolution CT image corresponding to the high-resolution CT image.
[0008] Furthermore, the downsampling process is a bicubic interpolation method.
[0009] Furthermore, a training dataset is constructed using the low-resolution CT images and the high-resolution CT images, including the following steps: The high-resolution CT image is paired with the corresponding low-resolution CT image to obtain an image pair; Image enhancement processing is performed on two pairs of images in the image pair to obtain enhanced image pairs; wherein, the training dataset includes the enhanced image pairs.
[0010] Furthermore, image enhancement processing includes random cropping, horizontal flipping, vertical flipping, and rotation.
[0011] Furthermore, the super-resolution convolutional neural network includes three convolutional layers: a feature extraction layer, a nonlinear mapping layer, and a reconstruction layer.
[0012] Furthermore, the loss function of the super-resolution convolutional neural network adopts mean squared error.
[0013] To address the aforementioned technical problems, the present invention also provides a durian CT image processing system, the specific technical content of which is as follows: A durian CT image processing system, comprising: The image acquisition module is used to acquire low-resolution and high-resolution CT images of durian; The data construction module is used to construct a training dataset using the low-resolution CT images and the high-resolution CT images; The model training module is used to construct a super-resolution convolutional neural network; the super-resolution convolutional neural network is trained using the training dataset to obtain an image reconstruction network; The image acquisition module is also used to acquire CT images of durians to be processed; The image reconstruction module is used to input the CT image to be processed into the image reconstruction network to obtain a high-resolution CT image.
[0014] To address the aforementioned technical problems, the present invention also provides a storage medium, the specific technical content of which is as follows: A storage medium storing a computer program or computer instructions, which, when executed by a computer processor, implement the steps of the durian CT image processing method described above.
[0015] To address the aforementioned technical problems, the present invention also provides a computer, the specific technical content of which is as follows: A computer includes a memory and one or more processors, wherein executable code is stored in the memory, and when the one or more processors execute the executable code, the steps of the durian CT image processing method described above are implemented. Attached Figure Description
[0016] Figure 1 This is a flowchart of a durian CT image processing method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a durian CT image processing system according to an embodiment of the present invention. Detailed Implementation
[0017] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.
[0018] like Figure 1 As shown, this embodiment provides a method for processing durian CT images, including the following steps: S1. Acquire low-resolution and high-resolution CT images of durian; the specific steps are as follows: Use a camera to take a cross-sectional image of the durian; The cross-sectional image is processed to obtain the high-resolution CT image; The high-resolution CT image is downsampled to obtain the corresponding low-resolution CT image. The high-resolution image is downsampled to simulate a low-resolution input. The most common method is bicubic interpolation.
[0019] S2. Construct a training dataset using the low-resolution CT images and the high-resolution CT images; the specific steps are as follows: The high-resolution CT image is paired with the corresponding low-resolution CT image to obtain an image pair; Image enhancement processing is performed on two pairs of images in the image pair to obtain enhanced image pairs; wherein, the training dataset includes the enhanced image pairs.
[0020] Image pairing involves pairing the high-resolution CT image with the corresponding low-resolution CT image to form an "LR-HR" image pair. This image pair is the basic unit for training the neural network, and the network's goal is to learn how to reconstruct an image from the LR image that is as close as possible to the original HR image.
[0021] Data augmentation, or data enhancement, is a technique used to increase the diversity of data and improve the generalization ability of models. This typically involves performing data augmentation operations on images, such as random cropping, horizontal and / or vertical flipping, and rotation. SCNet used data augmentation techniques during training, and its training dataset contained over 3,000 high-resolution images.
[0022] S3. Constructing a super-resolution convolutional neural network; The training data construction method of super-resolution convolutional neural networks (such as SRCNN, SCNet, etc.) usually follows a standardized process. The core is to generate corresponding low-resolution (LR) image pairs by downsampling high-resolution (HR) images, thereby forming the input-output data pairs required for supervised learning.
[0023] The super-resolution convolutional neural network includes three convolutional layers: a feature extraction layer, a nonlinear mapping layer, and a reconstruction layer.
[0024] The feature extraction layer uses a set of convolutional kernels (e.g., 9×9 size, 64 filters) to extract features from low-resolution image patches, outputting a feature map. This process reduces the size of the original HR image by a specific ratio (e.g., 1 / 2, 1 / 3, 1 / 4) to generate the corresponding LR image. For example, in the SCNet network, the input low-resolution image is cropped into 64×64 patches.
[0025] The nonlinear mapping layer uses 1×1 convolutional kernels (e.g., 32 filters) to perform nonlinear transformations on the extracted features, mapping high-dimensional features to another high-dimensional space to learn more complex reconstruction patterns.
[0026] The reconstruction layer uses another set of convolutional kernels (e.g., 5×5 in size, with the same number of channels as the input image) to recombine the non-linearly mapped features, generating the final high-resolution image patch.
[0027] S4. Train the super-resolution convolutional neural network using the training dataset to obtain the image reconstruction network; Input network training: Input the generated LR-HR image pairs into the super-resolution convolutional neural network. The network uses the LR image as input and the corresponding HR image as the target (Ground Truth), and trains the model parameters by minimizing the error (such as mean square error) between the output image and the target HR image.
[0028] The loss function and optimization employ mean squared error (MSE) as the loss function to measure the pixel-level difference between the high-resolution image output by the network and the real high-resolution image. Stochastic gradient descent (SGD) combined with backpropagation (BP) is used to optimize all convolutional kernel weights and bias parameters in the network, with the goal of minimizing the loss function to maximize evaluation metrics such as PSNR (Peak Signal-to-Noise Ratio) of the reconstructed image.
[0029] The network is trained iteratively on the training set, and its performance (e.g., validation loss val_loss and validation PSNR val_psnr) is evaluated on independent validation sets. Training typically continues for multiple epochs, and stops when performance metrics (e.g., val_loss) on the validation set no longer improve to prevent overfitting. The model is saved periodically during training, with the weights of the best-performing model on the validation set saved periodically for later use in image super-resolution reconstruction.
[0030] S5. Acquire CT images of durian to be processed; S6. Input the CT image to be processed into the image reconstruction network to obtain a high-definition CT image.
[0031] In some other embodiments, a durian CT image processing system is also provided, including an image acquisition module, a data construction module, a model training module, and an image reconstruction module; The image acquisition module is used to acquire low-resolution and high-resolution CT images of durian; The data construction module is used to construct a training dataset using the low-resolution CT images and the high-resolution CT images; The model training module is used to construct a super-resolution convolutional neural network; the super-resolution convolutional neural network is trained using the training dataset to obtain an image reconstruction network; The image acquisition module is also used to acquire CT images of durians to be processed; The image reconstruction module is used to input the CT image to be processed into the image reconstruction network to obtain a high-resolution CT image.
[0032] Among them, the image acquisition module, data construction module, model training module, and image reconstruction module are program modules, data processing devices, computing devices, or computers.
[0033] In other embodiments, a storage medium is also provided, which stores a computer program or computer instructions that, when executed by a computer's processor, implement the steps of the durian CT image processing method described above.
[0034] The storage medium can be an internal storage unit of any data processing device described in any of the foregoing embodiments, such as a hard disk or memory. The storage medium can also be an external storage device of any data processing device, such as a plug-in hard disk, smart memory card, SD card, flash memory card, etc., mounted on the device. Furthermore, the storage medium can include both internal storage units and external storage devices of any data processing device. The computer-readable storage medium is used to store the computer program and other programs and data required by the data processing device, and can also be used to temporarily store data that has been output or will be output.
[0035] In other embodiments, a computer is also provided, including a memory and one or more processors, wherein executable code is stored in the memory, and when the one or more processors execute the executable code, the steps of the durian CT image processing method described above are implemented.
[0036] The memory can be an internal storage unit of any data processing device described in any of the foregoing embodiments, such as a hard disk or RAM. The memory can also be an external storage device of any data processing device, such as a plug-in hard disk, smart memory card, SD card, flash memory card, etc., mounted on the device. Furthermore, the memory can include both internal storage units and external storage devices of any data processing device. The memory is used to store the computer program and other programs and data required by the data processing device, and can also be used to temporarily store data that has been output or will be output.
[0037] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the concept and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for processing durian CT images, characterized in that, Includes the following steps: Acquired low-resolution and high-resolution CT images of durian; A training dataset was constructed using the low-resolution CT images and the high-resolution CT images; Constructing a super-resolution convolutional neural network; The super-resolution convolutional neural network is trained using the training dataset to obtain the image reconstruction network; Acquire CT images of durians to be processed; The CT image to be processed is input into the image reconstruction network to obtain a high-resolution CT image.
2. The durian CT image processing method according to claim 1, characterized in that, Acquiring low-resolution and high-resolution CT images of durian involves the following steps: Use a camera to take a cross-sectional image of the durian; The cross-sectional image is processed to obtain the high-resolution CT image; The high-resolution CT image is downsampled to obtain the low-resolution CT image corresponding to the high-resolution CT image.
3. The durian CT image processing method according to claim 2, characterized in that, The downsampling process is a bicubic interpolation method.
4. The durian CT image processing method according to claim 1, characterized in that, Constructing a training dataset using the low-resolution CT images and the high-resolution CT images includes the following steps: The high-resolution CT image is paired with the corresponding low-resolution CT image to obtain an image pair; Image enhancement processing is performed on two pairs of images in the image pair to obtain enhanced image pairs; wherein, the training dataset includes the enhanced image pairs.
5. The durian CT image processing method according to claim 4, characterized in that, Image enhancement processing includes random cropping, horizontal flipping, vertical flipping, and rotation.
6. The durian CT image processing method according to claim 1, characterized in that, The super-resolution convolutional neural network includes three convolutional layers: a feature extraction layer, a nonlinear mapping layer, and a reconstruction layer.
7. The durian CT image processing method according to claim 1, characterized in that, The loss function of the super-resolution convolutional neural network uses mean squared error.
8. A system employing the durian CT image processing method according to any one of claims 1 to 7, characterized by, include: The image acquisition module is used to acquire low-resolution and high-resolution CT images of durian; The data construction module is used to construct a training dataset using the low-resolution CT images and the high-resolution CT images; The model training module is used to construct a super-resolution convolutional neural network; the super-resolution convolutional neural network is trained using the training dataset to obtain an image reconstruction network; The image acquisition module is also used to acquire CT images of durians to be processed; The image reconstruction module is used to input the CT image to be processed into the image reconstruction network to obtain a high-resolution CT image.
9. A storage medium, characterized in that, The storage medium stores a computer program or computer instructions, which, when executed by a computer processor, implement the steps of the durian CT image processing method as described in any one of claims 1 to 7.
10. A computer, comprising: The device includes a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, they implement the steps of the durian CT image processing method as described in any one of claims 1 to 7.