Method and system for obtaining image of ice crystal structure of litchi pulp under low-frequency electric field

Abstract

The application provides a method and system for obtaining an image of an ice crystal structure of litchi pulp under a low-frequency electric field, the method comprising: performing magnetic resonance scanning on litchi treated by a low-frequency electric field and stored under an ice temperature condition to obtain k-space data of the litchi pulp; performing Fourier transform on the k-space data to obtain image space data; inputting amplitude information and phase information of the image space data into a pre-trained convolutional neural network model as a double-channel to obtain an image of the ice crystal structure of the litchi pulp with enhanced ice crystal structure clarity; wherein the model comprises a convolutional attention module for enhancing the perception ability of the litchi ice crystal region. In the model training process, a composite loss function is used, and a particle swarm optimization algorithm is used to dynamically optimize and adjust the weights of each item in the composite loss function, the k-space data inside the litchi is obtained by combining a magnetic resonance imaging technology, and the convolutional neural network model processes and reconstructs the image space data to generate a high-definition ice crystal structure image.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of food preservation technology, and particularly relates to a method and system for obtaining an image of an ice crystal structure of litchi pulp under a low-frequency electric field. BACKGROUND

[0002] Litchi is known for its unique flavor and rich nutritional value. Litchi fruits and their tissues are rich in various physiologically active substances such as polyphenols, flavonoids, and polysaccharides. Litchi is a typical respiratory climacteric fruit, with a fast postharvest ripening rate, vigorous physiological metabolism, and a special fruit skin structure that is extremely prone to water loss and browning. These characteristics result in a very short shelf life for litchi, making it prone to rotting, deterioration, flavor deterioration, and loss of nutritional components during picking, transportation, storage, and sales, which severely affects its commercial value and limits the further development of the litchi industry.

[0003] Low-temperature storage is a commonly used and effective method to inhibit postharvest physiological metabolism of fruits and vegetables, delay aging, inhibit microbial activity, and thus prolong the shelf life. According to the different temperature intervals, low-temperature preservation can be divided into cold storage (usually above 0℃), ice temperature (between 0℃ and the tissue freezing point), slight freezing, and freezing. Among them, the ice temperature preservation technology has unique advantages because it can maintain food at a state close to the freezing point but not frozen, which can significantly inhibit enzyme activity and microbial growth at a lower temperature compared to traditional cold storage, while avoiding ice crystal damage and quality deterioration caused by freezing.

[0004] Although ice temperature preservation shows potential in other foods, current research on ice temperature preservation for litchi is still insufficient. Especially crucial is that under ice temperature conditions, the tissue inevitably faces the problem of ice crystal formation. The size, shape, distribution of ice crystals and their interaction with cell structures are directly related to the integrity of fruit tissue structure, juice retention ability, and final eating quality. However, a deep understanding of the micro-mechanism of ice crystal formation in litchi pulp under ice temperature conditions is crucial for optimizing preservation technology and maximizing litchi quality. Existing conventional detection methods often have difficulty in non-destructively, intuitively, and finely revealing the microstructure evolution rules of litchi during preservation, especially the ice crystals.

[0005] Therefore, there is an urgent need for a technical means that can effectively observe the formation of ice crystals in litchi during ice temperature preservation. SUMMARY

[0006] To overcome the problems in the related art, one of the purposes of the present application is to provide a method for obtaining an image of the ice crystal structure of litchi flesh under a low-frequency electric field, which combines magnetic resonance imaging (MRI) technology to obtain k-space data of the inside of litchi, and uses a specific optimized convolutional neural network (CNN) model to process and reconstruct the image space data after Fourier transform of the data, to generate a high-definition ice crystal structure image, so as to overcome the problem that there is a lack of effective, non-destructive, intuitive and fine technical means to observe and analyze the formation, distribution and evolution of the internal micro-ice crystals of litchi under ice-temperature preservation conditions in the prior art.

[0007] A method for obtaining an image of the ice crystal structure of litchi flesh under a low-frequency electric field, comprising:

[0008] Performing magnetic resonance scanning on litchi that has been treated by a low-frequency electric field and stored under ice-temperature conditions to obtain k-space data of the flesh of the litchi;

[0009] Performing Fourier transform on the k-space data to obtain image space data containing amplitude information and phase information;

[0010] Taking the amplitude information and phase information of the image space data as a dual-channel input, inputting the dual-channel input into a pre-trained convolutional neural network model, and obtaining an image of the ice crystal structure of the flesh of the litchi with enhanced clarity of the ice crystal structure;

[0011] The convolutional neural network model comprises a convolutional attention module, and the convolutional attention module is used to enhance the perception ability of the convolutional neural network model to the feature channels and spatial positions of the ice crystal region of litchi.

[0012] By combining the k-space data obtained by magnetic resonance scanning with a specific convolutional neural network model containing a convolutional attention module, and processing the dual-channel data containing amplitude and phase information, a microstructure image of the internal flesh of litchi can be obtained non-destructively. Compared with traditional methods, this method uses the powerful image reconstruction capability of the deep learning model and the focusing capability of the attention mechanism on the key features (ice crystal region), significantly improves the clarity and detail of the ice crystal structure, and overcomes the difficulty of traditional imaging technology in observing the small ice crystal structure inside litchi under ice-temperature preservation, providing a high-quality image basis for in-depth study of the ice crystal formation mechanism.

[0013] Further, the pre-trained convolutional neural network model is obtained through a training process comprising the following steps:

[0014] Obtain a litchi nuclear magnetic resonance image data set for training, and label the ice crystal region of the litchi flesh in the litchi nuclear magnetic resonance image data set;

[0015] dividing the litchi magnetic resonance image dataset into a training set and a validation set;

[0016] training the convolutional neural network model using the training set and verifying the convolutional neural network model through the validation set.

[0017] That is, even if the model is trained using a litchi magnetic resonance image dataset specially obtained and artificially annotated with ice crystal regions, and the training set and the validation set are divided, it is ensured that the convolutional neural network model is optimized for the specific task of "litchi flesh ice crystal structure recognition and enhancement". Compared with using a general image processing model or a model not trained with specific data, the model trained by the method has higher pertinence and accuracy, can more effectively identify and reconstruct the complex features of litchi ice crystals, and thus improves the reliability and detail fidelity of the final output image.

[0018] Further, in the step of training the convolutional neural network model using the training set, a composite loss function is used for model optimization.

[0019] In the model training process, the composite loss function is used for optimization, so that the model can consider multiple evaluation indicators (such as pixel-level accuracy, region overlap, and edge sharpness) during the learning process. Compared with using only a single loss function (such as only focusing on pixel accuracy or region overlap), the composite loss function can guide the model to achieve a more balanced and comprehensive optimization goal, which helps to generate image results that perform better in multiple key dimensions, especially for ice crystal morphology and boundary analysis.

[0020] Further, the composite loss function is a weighted combination of a binary cross-entropy loss term, a region overlap loss term, and an edge loss term.

[0021] The composite loss function is a weighted combination of binary cross-entropy loss (focusing on pixel classification accuracy), region overlap loss (focusing on the degree of fit between predicted regions and true regions, sensitive to small targets), and edge loss (focusing on edge sharpness), which can more accurately optimize the core needs of ice crystal image reconstruction. This combination is particularly beneficial for handling the characteristics of ice crystal structure (small size, boundary needs to be clearly defined), ensuring that the model not only roughly locates the ice crystals, but also accurately segments their regions and sharpens their edges, thereby obtaining more valuable images for subsequent analysis.

[0022] Further, the edge loss term is calculated by the following method:

[0023] applying an edge detection operator to the predicted ice crystal structure graph output by the convolutional neural network model and to the corresponding real-annotated ice crystal structure graph to obtain a predicted edge graph and a real edge graph, respectively;

[0024] calculating a mean square error between the predicted edge graph and the real edge graph to obtain a value of the edge loss term.

[0025] By applying an edge detection operator to the predicted graph output by the model and the real-annotated graph and calculating the mean square error between the edge graphs to define the edge loss term, the focus of model optimization is directly directed to the "edge consistency". This loss calculation method specifically for edges can more effectively force the model to learn and reproduce the precise contours and boundary details of the ice crystal structure, thereby significantly improving the clarity and continuity of the ice crystal edges in the final image.

[0026] Further, the edge detection operator is a Sobel operator.

[0027] The Sobel operator is sensitive to gradient changes in the image and can effectively extract the edge information of the ice crystal structure. Using this specific operator not only ensures the effectiveness of the edge loss calculation, but also improves the implementability of the method and the reproducibility of the results.

[0028] Further, the weight coefficient of the composite loss function is dynamically adjusted by a particle swarm optimization algorithm.

[0029] The particle swarm optimization (PSO) algorithm is used to dynamically adjust the weight coefficients of each loss term in the composite loss function, overcoming the difficulty of manually setting weights to achieve optimality and the difficulty of different training stages possibly requiring different weights. PSO can adaptively search for the optimal weight combination, so that the model training process can better balance the contributions of different loss terms, thereby possibly achieving better overall optimization effect than fixed weights, further improving the performance of the model and the quality of the final image.

[0030] Further, the particle swarm optimization algorithm is based on a preset velocity update formula and a position update formula, with the weighted combination value of the composite loss function as the optimization target to iteratively search for the optimal weight combination.

[0031] The global optimization strategy based on swarm intelligence can more efficiently and intelligently find a weight combination close to the global optimum than simple grid search or random search, ensuring the effectiveness and scientificity of weight optimization and helping the model to fully converge to a better state.

[0032] Further, the convolutional attention module is a CBAM module, which sequentially includes a channel attention submodule and a spatial attention submodule.

[0033] The channel attention sub-module is used for learning the importance weight of different feature channels in the litchi ice crystal image.

[0034] The spatial attention sub-module is used for learning the importance weight of different spatial positions in the litchi ice crystal image.

[0035] The second object of the present application is to provide a system for obtaining a litchi flesh micro-ice crystal structure image, which is used for executing the method for obtaining a litchi flesh ice crystal structure image under a low-frequency electric field as described above, and the system comprises:

[0036] A magnetic resonance scanning unit is configured to perform magnetic resonance scanning on litchi that has been treated by a low-frequency electric field and stored under ice temperature conditions, so as to obtain k-space data of the litchi flesh.

[0037] A data processing unit is configured to perform Fourier transform on the k-space data, so as to obtain image space data containing amplitude information and phase information.

[0038] An image reconstruction unit contains a pre-trained convolutional neural network model, the convolutional neural network model comprises a convolutional attention module, the convolutional neural network model is configured to receive the amplitude information and the phase information from the data processing unit as a double-channel input, and the pre-trained convolutional neural network model is configured to process the double-channel input, so as to obtain a litchi flesh ice crystal structure image with enhanced ice crystal structure clarity.

[0039] The present application has the following beneficial effects:

[0040] The method for obtaining a litchi flesh ice crystal structure image under a low-frequency electric field provided by the present application can obtain k-space data of litchi that has been treated by a low-frequency electric field and stored under ice temperature conditions by applying magnetic resonance imaging technology, then perform Fourier transform to obtain image space data containing amplitude and phase information, input the double-channel information into a convolutional neural network (CNN) model containing a convolutional attention module for reconstruction, and realize high-quality and non-destructive imaging of the micro-ice crystal structure inside the litchi flesh. Compared with the limitation that traditional observation methods are difficult to reveal the ice crystal morphology inside the tissue in depth and in detail, the present application uses the penetration of MRI (magnetic resonance imaging technology) to obtain k-space data of litchi flesh, and enhances the perception ability of key features by using double-channel information and an attention mechanism, which significantly improves the clarity, edge sharpness and detail performance of the ice crystal structure, and provides unprecedented intuitive and reliable image basis for accurately studying the micro-mechanism of the formation, distribution and interaction of ice crystals in litchi during ice temperature preservation. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1is a schematic diagram of the method for obtaining the image of the ice crystal structure of the litchi pulp under a low-frequency electric field provided in the present application. DETAILED DESCRIPTION

[0042] Preferred embodiments of the present application will be described in more detail with reference to the drawings. Although the preferred embodiments of the present application are shown in the drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present application is more thorough and complete, and the scope of the present application is fully conveyed to those skilled in the art.

[0043] EMBODIMENT

[0044] As shown in the drawings, the present embodiment provides a method and system for obtaining the image of the ice crystal structure of the litchi pulp under a low-frequency electric field. The method for obtaining the image of the ice crystal structure of the litchi pulp under a low-frequency electric field has the following specific steps: Figure 1 Step S1: sample preparation and pretreatment

[0045] Select 200 fresh litchis purchased in the market, and select the standard as follows: uniform fruit size (single fruit weight about 20-25g), bright red or slightly yellow-green color, intact peel without mechanical damage and pests. Number and mark the selected litchis. Place the marked litchis in an ice temperature preservation cabinet with a temperature setting of -1℃, and apply a low-frequency alternating electric field for treatment, and store for a period of time.

[0046] Step S2: magnetic resonance scanning and data acquisition

[0047] Randomly select samples from the litchis treated in step S1, and place them in the scanning area of the magnetic resonance imaging (MRI) equipment. Scan the pulp area of the litchi using an appropriate pulse sequence, and obtain its two-dimensional k-space (frequency domain) data. The k-space data is a complex matrix, as follows:

[0048]

[0049] Where u and v represent the spatial frequency domain coordinates of the nuclear magnetic resonance equipment in collecting the ice crystal of the litchi pulp during ice temperature preservation, and represent the frequency component indexes in the vertical and horizontal directions, respectively.

[0050] The value of represents the signal amplitude and phase at this frequency component.

[0051] Step S3: Fourier transform and initial image generation

[0052] The k-space data matrix obtained in step S2 is transformed into the image domain by Fourier transform, and the initial image is generated as follows: ​The initial image data of litchi is obtained by converting the data from the frequency domain to the image space domain using two-dimensional discrete Fourier transform. The two-dimensional discrete Fourier transform formula is as follows:

[0053] ;

[0054] where M and N are the number of rows and columns of the k-space matrix (corresponding to the image resolution), is the complex-valued pixel at the image space coordinate The amplitude information and the phase information are separated from the complex image data At this time, the obtained can be regarded as a preliminary gray-scale image, but the ice crystal structure may not be clear enough.

[0055] Step S4: Preparation of training data (this step is completed before the model is applied)

[0056] In order to train the subsequent convolutional neural network model, a labeled litchi magnetic resonance image dataset needs to be prepared. A batch (for example, 200) of ice-temperature preserved litchi magnetic resonance images (which can be the amplitude map generated in step S3, or a special acquisition such as a diffusion weighted imaging (DWI) image, which is more conducive to artificial identification of ice crystals) is obtained by MRI equipment. Experienced technicians manually or semi-automatically accurately label the ice crystal regions in the litchi flesh on these images to generate corresponding binary mask images (ice crystal regions are 1 and backgrounds are 0). Divide the 200 groups of images and their corresponding labeled masks into a training set (160 groups) and a validation set (40 groups) according to an 8:2 ratio at random.

[0057] Step S5: Convolutional neural network model construction and training

[0058] A convolutional neural network (CNN) model is constructed, which is designed for image reconstruction or segmentation tasks. The key is to replace the standard attention module (if any) in the model with a CBAM (Convolutional Block Attention Module) module. The CBAM module successively contains a channel attention submodule and a spatial attention submodule.

[0059] The attention mechanism of traditional CNN models usually only focuses on the spatial or channel dimension of the image, with limited perception granularity, which is prone to produce blurred effects in small ice crystal regions and lacks the ability to focus on the features of small regions. Therefore, the CBAM (Channel-Attention Module) attention module is introduced to combine channel and spatial dimension information to enhance the resolution of the boundary of the small ice crystal region in the MRI image and automatically focus on the small edges and weak texture structure of the ice crystal tissue region, thereby improving the performance and accuracy of the model.

[0060] The image spatial domain information obtained based on the Fourier transform formula is used to construct the input feature map of the CNN. Considering that the edges, textures, and ice crystal structure details of litchi ice crystal tissue depend more on phase information, a dual-channel input including amplitude and phase information is adopted to capture image information more comprehensively, thereby improving the model's ability to identify the edge, texture, and structure features of the ice crystal region. The formula is as follows:

[0061]

[0062] In the CBAM module, the input feature map is first filtered through the channel attention mechanism to select important feature channels in the ice crystal tissue region that are key in the field of litchi ice temperature preservation. The channel attention mechanism uses global pooling to compress spatial information to obtain statistical features of each channel and learns the attention weights of these channels through an MLP (Multi-Layer Perceptron), thereby enabling the network to focus on feature channels that are important in litchi ice temperature preservation. The formula is as follows:

[0063]

[0064] wherein, represents a sigmoid activation function that normalizes numerical values; represents the average pooling value of the input feature map, and the formula is as follows:

[0065]

[0066] wherein, represents the height of the feature map, i.e., the number of pixels in the vertical direction in the image; represents the width of the feature map, i.e., the number of pixels in the horizontal direction of the image; represents the number of channels in the feature map;

[0067] represents the maximum value of all pixel points in the input feature map channel, and the formula is as follows:

[0068]

[0069] The input feature map is processed by the CBAM channel attention module to obtain , and the formula is as follows:

[0070]

[0071] wherein, indicates element-wise multiplication; the semantic features are extracted by the CNN model, and a prediction map is output;

[0072]

[0073] wherein, indicates a complete convolutional neural network structure. indicates the extracted deep semantic feature map;

[0074] Then, the prediction map is output through convolution transformation and an activation function, as follows:

[0075]

[0076] wherein, is a 1x1 convolution kernel, which is used to map the double-channel into a single channel; indicates a convolution operation; b indicates a scalar or a bias map with the same size as the output map; indicates the intermediate prediction value after convolution; indicates a Sigmoid function; indicates the final prediction map.

[0077] The amplitude information and the phase information obtained in step S3 are provided as double-channel input feature maps to the CNN model.

[0078] The CBAM-CNN model is trained using the training set (160 groups of images and their labels) prepared in step S4. During the training process, a composite loss function is used for optimization. The composite loss function is a weighted combination of binary cross-entropy loss (BCE), Dice loss (also known as region overlap loss), and edge loss (EdgeLoss):

[0079] wherein, is the composite loss function value.

[0080] Binary cross-entropy calculates the cross-entropy between the pixel probability predicted by the model and the true label (0 or 1). The formula is as follows:

[0081]

[0082] wherein, indicates the total number of pixels in the image , represents a pixel value in a real nuclear magnetic resonance image, represents a pixel point in a predicted image.

[0083] Region overlap loss The overlap (Dice coefficient) between the predicted ice crystal region and the real labeled region is calculated, which facilitates the improvement of the sensitivity of the model to the formation of small ice crystal particles in the ice temperature preservation of litchi flesh, and the formula is as follows:

[0084]

[0085]

[0086] Edge loss The edge relationship between the predicted image and the real nuclear magnetic resonance image is calculated, and the formula is as follows:

[0087] The Sobel convolution is used to extract the edge of the ice crystal tissue and define the convolution kernel in the horizontal direction and the vertical direction and ;

[0088]

[0089] Convolution operation is performed on the predicted image and the real nuclear magnetic resonance image to obtain and ;

[0090]

[0091] The mean square error (MSE) between the edge maps is calculated

[0092]

[0093] The model is trained using the training set, and during the training process, the weight coefficients of the loss function are dynamically adjusted using the particle swarm optimization (PSO) algorithm , and . Let represent the arithmetic function of the kth generation example , and weight combination, the particle swarm optimization algorithm includes the velocity update and position update formulas, and the velocity update formula is as follows:

[0094]

[0095] wherein, and represent the acceleration constant; and a random number representing 0-1; is an inertia weight, represents the value of the comprehensive loss function corresponding to the i-th particle in the k-th generation; represents the minimum value of the comprehensive loss function experienced by the i-th particle in the k-th generation; represents the minimum value of the comprehensive loss function of the entire population in the k-th generation;

[0096] The position update formula is as follows:

[0097] represents the position of the i-th particle in the (k+1)-th generation; represents the position of the i-th particle in the k-th generation; represents the speed of the i-th particle in the (k+1)-th generation.

[0098] The three loss functions and the corresponding weights are combined into a comprehensive loss function as the objective function R of the particle swarm optimization algorithm, and the calculation formula is as follows:

[0099]

[0100] During the training process, the validation set (40 groups) prepared in step S4 is regularly used to evaluate the model performance, prevent overfitting, and adjust the hyperparameters or determine the optimal training round number. After training is completed, the model weight optimized is saved.

[0101] Step S6: Apply the trained model to image reconstruction

[0102] For a new two-channel (amplitude and phase) initial image data obtained by processing steps S1-S3, input it into the CBAM-CNN model trained in step S5. The model outputs a final litchi flesh ice crystal structure image through forward propagation, using its learned feature extraction, attention focusing and reconstruction capabilities. Compared with the initial amplitude image generated in step S3, this image has clearer ice crystal structure contours, more detailed ice crystal structure, less background noise, and significantly enhanced clarity.

[0103] The embodiment also provides a system for obtaining a litchi flesh micro-ice crystal structure image, which comprises:

[0104] A magnetic resonance scanning unit, for example, a low-field nuclear magnetic resonance instrument, is used to perform the scanning operation in step S2 to obtain k-space data.

[0105] Data processing unit: e.g. a computer configured with corresponding computing software (e.g. MATLAB, Python scientific computing library). This unit is used to perform the 2D inverse discrete Fourier transform in step S3, to calculate the initial image space data containing amplitude and phase information from the k-space data.

[0106] Image reconstruction unit: can be the same computer as the data processing unit, or a dedicated image processing workstation. This unit is loaded with the trained convolutional neural network model software containing the CBAM module in step S5. It receives the dual-channel (amplitude, phase) data from the data processing unit as input, performs the model forward propagation calculation in step S6, and finally outputs the litchi-textured ice crystal structure image with enhanced ice crystal structure clarity.

[0107] The embodiment provides a method and system for obtaining litchi flesh ice crystal structure image under low-frequency electric field, realizes non-destructive scanning of the internal structure of litchi by using MRI technology, and obtains original k-space data required for subsequent analysis. The amplitude and phase information after Fourier transform are used to form a double-channel input, which fully utilizes the advantages of phase information in reflecting edges and texture details, and provides more comprehensive image information for the subsequent CNN model than the single amplitude channel, which helps to improve the recognition ability of the model to the ice crystal structure characteristics. The CNN model containing the CBAM attention mechanism is adopted, which effectively enhances the focusing and resolution ability of the model to the small and weak texture ice crystal regions and their boundaries in the litchi MRI image through the attention ability of combining channel and spatial dimensions, and overcomes the limitations of traditional attention mechanisms. The model is trained using a litchi nuclear magnetic resonance image dataset specially labeled for litchi ice crystals, ensuring its high adaptability and accuracy for specific tasks. A compound loss function combining binary cross-entropy (pixel-level accuracy), Dice loss (region overlap, sensitive to small targets), and edge loss based on Sobel operator (boundary sharpness) is adopted, which comprehensively guides the model optimization from multiple dimensions, especially for the needs of small particle recognition and edge sharpening in ice crystal image reconstruction. The particle swarm optimization (PSO) algorithm is introduced to dynamically and adaptively adjust the weight coefficients of the compound loss function, overcoming the difficulty and suboptimality of manual setting or fixed weights, and finding a better weight balance point through an automated optimization mechanism, thereby further improving the comprehensive performance of the model and the final image quality. The finally output litchi flesh ice crystal structure image is significantly enhanced in terms of ice crystal contour sharpness, internal detail richness and background noise suppression compared with the original MRI image, providing high-quality and intuitive visual evidence for researchers, greatly facilitating the in-depth research, quantitative analysis and preservation process optimization of the ice crystal formation mechanism of litchi under specific ice temperature preservation conditions. The system provided integrates scanning, data processing and image reconstruction based on advanced models, providing a complete solution for the standardized and efficient application of the method.

[0108] Unless specifically stated otherwise, the relative arrangements of the components and steps illustrated in these embodiments and the numerical expressions and values set forth herein are not limiting. In all examples shown and discussed herein, any specific values should be interpreted as merely exemplary, and are not to be construed as limiting. Thus, other examples of the example embodiments can have different values. It is noted that like references and labels can be used to denote like items throughout the drawings, and that, unless specifically stated otherwise, no reference or label should be construed that the item being referred to is the only item to which the reference or label refers.

[0109] In addition, it should be noted that the use of "first", "second", and the like, is merely to distinguish between similar objects, and does not have a special meaning, and therefore cannot be understood as a limitation on the scope of protection of the present application.

[0110] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Various modifications and changes can be made by those skilled in the art to the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A method for acquiring images of the ice crystal structure of litchi flesh under a low-frequency electric field, characterized in that, include: Magnetic resonance scanning was performed on litchi that had been treated with a low-frequency electric field and stored under ice-temperature conditions to obtain k-space data of the litchi flesh. Perform a Fourier transform on the k-space data to obtain image spatial data containing amplitude and phase information; The amplitude and phase information of the image spatial data are used as dual-channel inputs and fed into a pre-trained convolutional neural network model to obtain an image of the ice crystal structure of lychee flesh with enhanced clarity of the ice crystal structure. The convolutional neural network model includes a convolutional attention module, which is used to enhance the convolutional neural network model's ability to perceive the feature channels and spatial location of the lychee ice crystal region. The pre-trained convolutional neural network model is optimized using a composite loss function during training. The composite loss function is a weighted combination of a binary cross-entropy loss term, a region overlap loss term, and an edge loss term. The region overlap loss term is used to calculate the degree of overlap between the predicted ice crystal region and the actual labeled region. The edge loss term is calculated as follows: the edge detection operator is applied to the predicted ice crystal structure map and the corresponding real labeled ice crystal structure map output by the convolutional neural network model to obtain the predicted edge map and the real edge map, respectively; the mean square error between the predicted edge map and the real edge map is calculated to obtain the value of the edge loss term.

2. The method for acquiring images of the ice crystal structure of lychee flesh under a low-frequency electric field according to claim 1, characterized in that: The pre-trained convolutional neural network model is obtained through the following steps: Obtain a litchi nuclear magnetic resonance image dataset for training, and annotate the litchi flesh ice crystal regions in the litchi nuclear magnetic resonance image dataset; The litchi nuclear magnetic resonance image dataset was divided into a training set and a validation set; The convolutional neural network model is trained using the training set and validated using the validation set.

3. The method for acquiring images of the ice crystal structure of lychee flesh under a low-frequency electric field according to claim 1, characterized in that: The edge detection operator is the Sobel operator.

4. The method for acquiring images of the ice crystal structure of lychee flesh under a low-frequency electric field according to claim 1, characterized in that: The weight coefficients of the composite loss function are dynamically adjusted using a particle swarm optimization algorithm.

5. The method for acquiring images of the ice crystal structure of lychee flesh under a low-frequency electric field according to claim 4, characterized in that: The particle swarm optimization algorithm is based on preset velocity update formulas and position update formulas, and takes minimizing the weighted combination value of the composite loss function as the optimization objective, iteratively searching for the optimal combination of weight coefficients.

6. The method for acquiring images of the ice crystal structure of lychee flesh under a low-frequency electric field according to claim 1, characterized in that: The convolutional attention module is a CBAM module, which includes a channel attention submodule and a spatial attention submodule in sequence. The channel attention submodule is used to learn the importance weights of different feature channels in the lychee ice crystal image; The spatial attention submodule is used to learn the importance weights of different spatial locations in the lychee ice crystal image.

7. A system for acquiring images of the microscopic ice crystal structure of litchi flesh, characterized in that, A method for acquiring an image of the ice crystal structure of lychee flesh under a low-frequency electric field as described in any one of claims 1 to 6, the system comprising: The magnetic resonance scanning unit is used to perform magnetic resonance scanning on litchi that have been treated with a low-frequency electric field and stored under ice-temperature conditions to obtain k-space data of litchi flesh. The data processing unit is used to perform Fourier transform on the k-space data to obtain image spatial data containing amplitude information and phase information; The image reconstruction unit includes a pre-trained convolutional neural network model, which includes a convolutional attention module. The convolutional neural network model is used to receive the amplitude information and phase information from the data processing unit as dual-channel inputs. The pre-trained convolutional neural network model is used to process the dual-channel inputs to obtain an image of the ice crystal structure of lychee flesh with enhanced clarity of the ice crystal structure.

Images