Humidity monitoring method and system in agricultural product cold-chain logistics transportation process

Abstract

The invention relates to the technical field of agricultural product transportation, in particular to an agricultural product cold-chain logistics transportation process humidity monitoring method and system, and the method comprises the steps: collecting microenvironment humidity data of the surface of an agricultural product and visual form data representing the moisture state of the agricultural product; calculating to obtain a real-time moisture state evaluation result of the agricultural product based on the microenvironment humidity data and the visual form data; the moisture state evaluation result is mapped into the transportation humidity grade of the agricultural product, so that the humidity monitoring result can truly reflect the actual moisture condition of the agricultural product, and the problem of insufficient precision caused by dislocation of monitoring objects in a traditional method is solved.

Description

Technical Field

[0001] This invention relates to the field of agricultural product transportation technology, specifically to a method and system for monitoring humidity during the cold chain logistics transportation of agricultural products. Background Technology

[0002] Maintaining a suitable humidity environment during the cold chain logistics transportation of agricultural products is crucial for ensuring the quality and freshness of fruits, vegetables, flowers, and other agricultural products, and can effectively reduce economic losses caused by wilting due to dehydration or spoilage due to excessive moisture.

[0003] Currently, humidity monitoring methods mainly rely on deploying environmental humidity sensors inside refrigerated truck compartments to measure the ambient humidity parameters of the air inside the compartment and then adjust the environment accordingly. However, because the monitored object is the ambient humidity of the space where agricultural products are located, this is an indirect indicator and cannot directly and accurately reflect the actual moisture state changes of the agricultural products themselves. Furthermore, due to factors such as airflow distribution within the compartment and the way goods are stacked, the ambient humidity often deviates from the true microenvironment humidity and physiological state of the agricultural products' surface, leading to inaccurate humidity monitoring results and making it difficult to achieve real-time, effective judgment and early warning of the preservation status of agricultural products. Summary of the Invention

[0004] To address the technical problem that existing methods only monitor the macroscopic humidity inside the vehicle compartment, which cannot directly and accurately reflect the actual humidity of the agricultural products themselves, resulting in inaccurate humidity monitoring results, this application provides a method and system for monitoring humidity during the cold chain logistics transportation of agricultural products.

[0005] The humidity monitoring method and system for agricultural product cold chain logistics transportation provided in this application adopts the following technical solution: A method for monitoring humidity during the cold chain logistics transportation of agricultural products includes: Collect microenvironmental humidity data on the surface of agricultural products and visual morphological data characterizing the moisture state of agricultural products; Based on microenvironment humidity data and visual morphology data, the real-time moisture status assessment results of agricultural products are calculated. The moisture status assessment results are mapped to the transport humidity level of agricultural products.

[0006] Further steps include collecting microenvironmental humidity data on the surface of agricultural products and visual morphological data characterizing the moisture state of agricultural products: The microenvironment humidity data of agricultural product surfaces is collected using a flexible humidity sensor array; Initial visual data characterizing the moisture state of agricultural products is acquired using a multispectral camera. Based on a preset image segmentation algorithm, regions of interest (ROIs) on the surface of agricultural products are extracted from the initial visual data. Texture feature analysis and morphological quantization are then performed on the ROIs to obtain visual morphological data.

[0007] Further steps, including texture feature analysis and morphological quantization of the region of interest to obtain visual morphological data, include: Texture feature analysis is performed on the region of interest to obtain multi-scale texture feature vectors; Frequency domain texture features are extracted from multi-scale texture feature vectors, and principal component analysis is performed on the frequency domain texture features to reduce their dimensionality, resulting in a subset of texture features. Morphological quantization is also performed on the region of interest to obtain a set of morphological feature parameters. The texture feature subset and the morphological feature parameter set are fused to obtain visual morphological data.

[0008] Furthermore, based on microenvironmental humidity data and visual morphology data, the steps for calculating the real-time moisture status assessment results of agricultural products include: Based on a pre-set deep neural network, humidity depth features are extracted from microenvironment humidity data, and visual depth features are extracted from visual morphology data through a pre-set convolutional neural network. Humidity depth features and visual depth features are fused to obtain multimodal fusion features; The multimodal fusion features are input into a preset moisture state assessment model to calculate the moisture state assessment results.

[0009] Furthermore, before inputting the multimodal fusion features into the preset moisture state assessment model, the following steps are included: Acquire historical training sample sets under different moisture conditions. The historical training sample sets consist of historical microenvironmental humidity data and historical visual morphology data of agricultural products. The target training sample set is obtained by expanding the historical training sample set through adversarial operations; Deep learning is performed on the target training sample set to obtain an initial training model. Then, the weights of the multimodal features in the initial training model are optimized to obtain a preset moisture state assessment model.

[0010] Furthermore, prior to the step of mapping the moisture status assessment results to the transport moisture level of agricultural products, the following steps are included: Obtain historical moisture status assessment results and establish an initial mapping relationship library between historical moisture status assessment results and transportation humidity levels; Based on real-time parameters of the current cold chain logistics transportation environment, the initial mapping relation library is weighted to obtain the target mapping relation library.

[0011] Furthermore, the steps for mapping moisture status assessment results to transport humidity levels for agricultural products include: In the target mapping database, query the initial transport humidity level corresponding to the moisture status assessment result; Based on historical transportation quality data of agricultural products, the initial transportation humidity level is calibrated with confidence to obtain the transportation humidity level.

[0012] This application also provides a humidity monitoring system for the cold chain logistics transportation process of agricultural products, including: The data acquisition module is used to collect microenvironmental humidity data on the surface of agricultural products and visual morphological data characterizing the moisture state of agricultural products. The data calculation module is used to calculate the real-time moisture status assessment results of agricultural products based on microenvironmental humidity data and visual morphology data. The data mapping module is used to map the moisture status assessment results to the transportation humidity level of agricultural products.

[0013] Beneficial effects achieved: This application provides a method for monitoring humidity during the cold chain logistics transportation of agricultural products, including: collecting microenvironmental humidity data on the surface of agricultural products and visual morphological data characterizing the moisture state of agricultural products; calculating the real-time moisture state assessment result of agricultural products based on the microenvironmental humidity data and visual morphological data; and mapping the moisture state assessment result to the transportation humidity level of agricultural products.

[0014] In this application, by collecting micro-environmental humidity data on the surface of agricultural products, humidity parameters that are closest to the agricultural products themselves are obtained directly, avoiding the deviation between the macro-environmental humidity inside the vehicle and the actual humidity on the surface of the agricultural products. At the same time, visual morphological data that characterizes the moisture state of agricultural products are combined for multi-dimensional verification. Based on the fusion analysis of the two types of data that directly reflect the state of agricultural products, the moisture state assessment result is obtained. This moisture state assessment result comes from the agricultural products themselves rather than the surrounding environment, thus realizing the transformation from indirect monitoring to direct perception. Finally, the moisture state assessment result is mapped to the transportation humidity level, establishing a direct correspondence between the state of agricultural products and environmental assessment. This allows the humidity monitoring results to truly reflect the actual moisture status of agricultural products, solving the problem of insufficient accuracy caused by misalignment of the monitoring object in traditional methods. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating the steps of a method for monitoring humidity during the cold chain logistics transportation of agricultural products, as described in this application. Figure 2 This is a schematic diagram of a humidity monitoring system for the cold chain logistics transportation of agricultural products according to this application.

[0016] Explanation of icon numbers: 10. Data acquisition module; 20. Data calculation module; 30. Data mapping module; 40. Model building module. Detailed Implementation

[0017] The following is in conjunction with the appendix Figure 1 and 2 This application will be described in further detail.

[0018] In the description of this application, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0019] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0020] This application discloses a method and system for monitoring humidity during the cold chain logistics transportation of agricultural products.

[0021] Please refer to Figure 1 The humidity monitoring method for agricultural product cold chain logistics transportation proposed in this embodiment includes steps S10~S30: Step S10: Collect microenvironmental humidity data of the surface of agricultural products and visual morphological data characterizing the moisture state of agricultural products.

[0022] In this step, by collecting microenvironmental humidity data and visual morphological data of the surface of agricultural products, the limitations of indirect monitoring in existing humidity monitoring methods can be overcome. The monitoring object is changed from the cold chain transportation environment of agricultural products to the agricultural products themselves, thereby establishing a perception system of the agricultural products' own state. By acquiring humidity data that is closest to the surface of agricultural products, environmental bias is eliminated. At the same time, visual morphological data is used to capture the physical characteristic changes caused by moisture state, so as to achieve direct perception of the true humidity state of agricultural products. This fundamentally solves the problem of insufficient humidity monitoring accuracy caused by the deviation of the monitoring object in existing methods.

[0023] Step S20: Based on microenvironmental humidity data and visual morphology data, calculate the real-time moisture status assessment results of agricultural products.

[0024] By fusing and analyzing microenvironmental humidity data and visual morphology data, perceived information is transformed into a quantitative evaluation of the physiological state of agricultural products. This overcomes the limitations of single monitoring indicators. While using microenvironmental humidity data to reflect surface moisture status, visual morphology data captures changes in physical characteristics caused by moisture variations, resulting in an assessment that more accurately reflects the true moisture state of agricultural products. This approach eliminates interference from environmental factors and improves the reliability of state judgment, ultimately achieving real-time and accurate perception of the moisture state of agricultural products and providing a direct basis for humidity control during transportation.

[0025] Step S30: Map the moisture status assessment results to the transport humidity level of agricultural products.

[0026] By mapping moisture status assessment results to the transportation humidity level of agricultural products, this method aims to establish a quantitative correspondence between the physiological state of agricultural products and transportation environmental parameters. It transforms the assessment results of biological indicators into environmental level parameters that can be practically guided. By using preset mapping rules, it achieves intelligent conversion from agricultural product status to environmental requirements, enabling humidity monitoring data to directly serve the control and decision-making of the transportation environment. Ultimately, it achieves accurate classification and dynamic adjustment of transportation humidity levels, providing a scientific basis for humidity management in the cold chain logistics process, thereby effectively improving the preservation effect and transportation quality of agricultural products.

[0027] In one feasible implementation, the specific steps for collecting microenvironment humidity data and visual morphology data include S11~S12: Step S11: Collect microenvironmental humidity data on the surface of agricultural products using a flexible humidity sensor array.

[0028] By directly attaching a flexible humidity sensor array to the surface of agricultural products or integrating it into the contact layer of agricultural product packaging materials, the flexible humidity sensor array forms a conformal contact with the surface of agricultural products, thereby directly measuring the microenvironmental humidity parameters of the boundary layer of the agricultural product surface.

[0029] The flexible humidity sensor array synchronously collects humidity data from different spatial locations at a preset sampling frequency, and converts the analog signal into digital micro-environment humidity data through a signal conditioning circuit. This enables the direct acquisition of humidity parameters closest to the surface of agricultural products, avoiding measurement deviations caused by factors such as air circulation and temperature gradients in traditional environmental monitoring.

[0030] It should be noted that the flexible humidity sensor array in this embodiment refers to a collection of humidity sensors that uses flexible materials such as polyimide and PE as a substrate and has the characteristics of being bendable and conformable, such as the Sensirion SHT4xFlex series sensors.

[0031] Step S12: Initial visual data characterizing the moisture state of agricultural products is acquired using a multispectral camera. Based on a preset image segmentation algorithm, the region of interest on the surface of the agricultural products is extracted from the initial visual data. Texture feature analysis and morphological quantization are then performed on the region of interest to obtain visual morphological data.

[0032] By utilizing a multispectral camera to simultaneously acquire image information of agricultural product surfaces in multiple specific bands of visible and near-infrared light, such as 970nm and 1200nm, which are sensitive to moisture absorption, initial visual data containing spatial and spectral dimensions is formed. Then, based on a preset image segmentation algorithm, such as the U-Net network, the initial visual data is processed. This preset image segmentation algorithm has been pre-trained to learn the differences in spectral characteristics between agricultural products and the background environment in different bands. It can automatically identify and accurately segment the agricultural product area in the initial visual data, separating it from the complex transportation background as the region of interest, eliminating interference from irrelevant factors such as the background of the carriage and packaging materials. This ensures that subsequent texture feature analysis and morphological quantization processing are only performed on the surface area of ​​the agricultural product itself, thereby ensuring that the extracted visual morphological data can accurately reflect the true moisture state of the agricultural product.

[0033] The subsequent textural feature analysis and morphological quantification of the region of interest aim to transform the visual information within the region of interest into quantifiable feature parameters. By analyzing the textural features of agricultural product surfaces, such as roughness, directionality, and regularity, the changes in surface microstructure caused by moisture variations can be reflected. Simultaneously, morphological quantification, such as calculating geometric parameters like area, perimeter, and contour shape, captures macroscopic morphological changes caused by moisture imbalance. This transforms subjective visual observation into objective numerical features. The resulting visual morphological data can accurately characterize the physical changes in the moisture state of agricultural products, providing a reliable quantitative basis for subsequent moisture state assessment.

[0034] Furthermore, step S12 may specifically include steps S121 to S124: Step S121: Perform texture feature analysis on the region of interest to obtain multi-scale texture feature vectors.

[0035] First, a multi-scale image space for the region of interest is constructed using the Gaussian pyramid algorithm. At each scale level, a Gabor filter bank is applied for texture feature extraction. This filter bank contains filter kernels with different directions and frequencies. Convolution operations are used to obtain the texture response value of each pixel in the region of interest. Then, statistical features of these texture response values, including mean, variance, energy, and entropy, are calculated at each scale level to form texture feature sub-vectors for the corresponding scale level. Finally, the texture feature sub-vectors from all scale levels are concatenated and integrated to construct a multi-scale texture feature vector that comprehensively characterizes the surface texture features of agricultural products. This allows for the simultaneous capture of texture changes from macroscopic to microscopic levels on the surface of agricultural products, particularly effectively detecting subtle changes in surface texture caused by variations in moisture content, providing rich visual feature data for subsequent moisture status assessment.

[0036] Step S122: Extract frequency domain texture features from the multi-scale texture feature vector, and perform principal component analysis to reduce the dimensionality of the frequency domain texture features to obtain a subset of texture features.

[0037] By performing a Fast Fourier Transform on the multi-scale texture feature vectors, the spatial domain texture features are transformed into the frequency domain to obtain frequency domain texture features. Then, principal component analysis is performed on the frequency domain texture features for dimensionality reduction. The covariance matrix of the frequency domain texture features is calculated to quantify the correlation strength between the feature dimensions. This covariance matrix can accurately reflect the distribution structure and redundancy of the frequency domain texture features. Then, eigenvalues ​​and eigenvectors are obtained by eigenvalue decomposition of the covariance matrix. By selecting the eigenvectors corresponding to the k largest eigenvalues, a projection matrix is ​​constructed. The column vectors of this projection matrix represent the principal component directions with the largest variance in the original feature space. Finally, the frequency domain texture features are projected onto a low-dimensional space spanned by the principal component directions to achieve feature dimension compression, thereby obtaining a subset of texture features that retains the main discriminative information while eliminating redundancy.

[0038] Step S123: Perform morphological quantization on the region of interest to obtain a set of morphological feature parameters.

[0039] The region of interest (ROI) is binarized to obtain a binary image, and the minimum bounding rectangle (MBCR) contour of the binary image is extracted. The compactness parameter is obtained by calculating the ratio between the contour area of ​​the MBCR and the area of ​​the MBCR. Then, the Canny edge detection algorithm is used to extract the region boundary of the binary image, and the average distance between the boundary pixels of the binary image and the centroid of the ROI is calculated to obtain the roundness parameter. Then, the binary image is transformed by distance, and local maxima are located. The concavity-convexity parameter is obtained by analyzing the distribution characteristics of the maxima. Finally, the uniformity parameter is obtained by calculating the distribution variance of the pixel intensity inside the ROI based on the grayscale image. The compactness parameter, roundness parameter, concavity-convexity parameter, and uniformity parameter are combined to form a morphological feature parameter set, which can accurately quantify the morphological feature changes on the surface of agricultural products caused by moisture changes, and provide a reliable morphological feature basis for moisture status assessment.

[0040] Step S124: The texture feature subset and the morphological feature parameter set are fused to obtain visual morphological data.

[0041] To eliminate the influence of dimensions, the texture feature subset and morphological feature parameter set are subjected to min-max normalization. Then, a correlation model between the texture feature subset and the morphological feature parameter set is established using canonical correlation analysis. By calculating the covariance matrix and eigenvectors of the texture feature subset and the morphological feature parameter set, the projection direction that maximizes the correlation between the texture feature subset and the morphological feature parameter set is found. The texture feature subset and the morphological feature parameter set are projected onto the canonical correlation space spanned by the projection direction. The texture feature subset and the morphological feature parameter set projected onto the canonical correlation space are then spliced ​​and fused to form unified visual morphological data. By utilizing the complementarity between texture features and morphological features, visual morphological data that can comprehensively characterize the surface visual characteristics of agricultural products is constructed, providing richer and more stable feature basis for moisture status assessment.

[0042] In one feasible implementation, the specific steps for calculating the moisture state assessment results include S21~S24: Step S21: Extract humidity depth features from microenvironment humidity data based on a preset deep neural network.

[0043] Microenvironmental humidity data is input into a pre-defined deep neural network containing multiple fully connected layers. The encoder part of this pre-defined deep neural network performs nonlinear transformation and feature compression on the input data through multiple hidden layers. The ReLU activation function is used to process the output in each hidden layer. Finally, the bottleneck layer transforms the microenvironmental humidity data into a low-dimensional dense vector. This vector is the humidity depth feature that can characterize the essential features of the microenvironmental humidity data. By automatically learning the complex nonlinear patterns and temporal dependencies in the microenvironmental humidity data, deep feature representations that are difficult to design manually are captured, providing more effective feature input for subsequent multimodal feature fusion.

[0044] Step S22: Extract visual depth features from visual morphology data using a pre-defined convolutional neural network.

[0045] Visual morphological data is reconstructed into a two-dimensional feature map and input into a pre-defined convolutional neural network. This network extracts features step by step through multiple convolutional layers. The bottom convolutional kernels are responsible for capturing basic features such as local texture in the visual morphological data. The higher convolutional kernels gradually abstract more complex morphological feature representations from the visual morphological data by stacking and combining receptive fields. Pooling operations are used after each convolutional layer to reduce feature size and enhance feature invariance. Finally, a fully connected layer maps the abstracted morphological feature representations into fixed-dimensional visual depth feature vectors. By automatically learning high-level feature representations related to moisture status in the visual morphological data, the network effectively captures the visual feature change patterns on the surface of agricultural products caused by moisture changes, providing a feature representation with strong discriminative power for moisture status assessment.

[0046] Step S23: The humidity depth features and visual depth features are fused to obtain multimodal fusion features.

[0047] Humidity depth features and visual depth features are standardized to have the same scale. Then, an attention-based feature fusion method is used to calculate the correlation weight between humidity depth features and visual depth features. This correlation weight is then used to dynamically weight and combine the humidity depth features and visual depth features. The correlation weight is obtained by calculating the dot product of the humidity depth features and visual depth features and normalizing it using the softmax function. Finally, the weighted humidity depth features and visual depth features are concatenated into a unified multimodal fusion feature vector. By automatically adjusting their contribution to the fusion feature based on the intrinsic correlation between the humidity depth features and visual depth features, the advantages of the two depth feature information are complemented, resulting in a feature representation that is more discriminative than single feature information.

[0048] Step S24: Input the multimodal fusion features into the preset moisture state assessment model and calculate the moisture state assessment result.

[0049] Multimodal fusion features are input as input layer data into a pre-defined moisture state assessment model. This model typically employs a multi-layer fully connected neural network structure, performing linear transformations and nonlinear activation functions layer by layer through a forward propagation algorithm. The linear transformation calculates the input value of each multimodal fusion feature using a weight matrix and a bias vector, while nonlinear activation functions such as ReLU introduce the model's nonlinear expressive power. After feature transformations through multiple hidden layers, the original output value of the neural network is converted into a probability distribution corresponding to different moisture state levels using a Softmax activation function at the output layer. The category with the highest probability value is selected as the moisture state assessment result for output. By constructing an end-to-end pre-defined moisture state assessment model from multimodal fusion features to moisture state levels, this model can automatically learn the complex nonlinear relationship between multimodal fusion features and moisture state, achieving intelligent and accurate assessment of the moisture state of agricultural products.

[0050] Steps S241 to S243 may be included before step S24: Step S241: Obtain historical training sample sets under different moisture conditions. The historical training sample sets are historical microenvironmental humidity data and historical visual morphology data of agricultural products.

[0051] In a controlled experimental environment, agricultural product samples were subjected to different gradients of moisture treatment to reach various preset states, including moisture saturation, normal moisture content, slight water loss, and severe water loss. Simultaneously, a flexible humidity sensor array was used to collect historical microenvironmental humidity data of the agricultural product surface, and a multispectral imaging system was used to collect historical visual morphological data for each state. Data collection for each moisture state was repeated a sufficient number of times to ensure statistical significance. Finally, the collected historical microenvironmental humidity data, historical visual morphological data, and the true values ​​of moisture states obtained through standard measurement methods were strictly correlated and labeled to construct a historical training sample set containing input features and target outputs. This provides sufficient and high-quality supervised learning data for subsequent model training operations, ensuring that the preset moisture state assessment model can effectively learn the multimodal feature patterns corresponding to different moisture states from the data.

[0052] Step S242: Expand the historical training sample set by adversarial operations to obtain the target training sample set.

[0053] A conditional generative adversarial network (GAN) framework is adopted, in which the generator takes random noise and moisture state labels as input and learns to generate microenvironmental humidity data and visual morphological data consistent with the distribution of historical training sample sets. The discriminator continuously optimizes its discrimination ability by distinguishing between real samples (i.e., historical training sample sets) and generated samples through adversarial training. During the training process, the parameters of the generator and discriminator are alternately optimized through the backpropagation algorithm until Nash equilibrium is reached. At this point, the generator can produce expanded samples that are highly similar to the distribution of real data. Finally, the generated samples and the historical training sample set are combined to form the target training sample set, thereby increasing the amount and diversity of training data. This effectively solves the problem of model overfitting caused by the limited number of actual labeled samples and improves the generalization ability of the preset moisture state assessment model in complex real-world scenarios.

[0054] Step S243: Perform deep learning on the target training sample set to obtain an initial training model, and then optimize the multimodal feature weights in the initial training model to obtain a preset moisture state assessment model.

[0055] Historical microenvironmental humidity data and historical visual morphology data in the target training sample set are used as input features, and their corresponding ground truth values ​​of water state are used as supervision signals. The loss function between the prediction result and the ground truth value of water state is calculated through forward propagation of a deep neural network, and the multimodal feature weight parameters are iteratively updated using the backpropagation algorithm until the model converges, thus obtaining the initial training model.

[0056] Next, the multimodal feature weights in the initial training model are optimized based on the attention mechanism. By calculating the correlation weights between humidity depth features and visual depth features, the contribution ratio of different modal features in the final moisture state assessment is dynamically adjusted, so that the initial training model can adaptively focus on the feature modal most relevant to the current moisture state. The final preset moisture state assessment model can effectively balance the differences in importance of multimodal features and achieve a more accurate moisture state assessment effect.

[0057] In one feasible implementation, the specific steps for determining the transport humidity level include S31-S32: Step S31: Query the initial transport humidity level corresponding to the moisture status assessment result in the target mapping relation library.

[0058] A target mapping relationship database is established, which includes the correspondence between moisture status assessment results and transportation humidity levels. After calculating the moisture status assessment results using a preset moisture status assessment model, the moisture status assessment results are used as query keys to perform exact matching or nearest neighbor matching searches in the target mapping relationship database. Specifically, if the moisture status assessment results are discrete levels, the corresponding initial transportation humidity level is directly retrieved using exact matching. If the moisture status assessment results are continuous values, the initial transportation humidity level is determined by taking the mean or optimal value after determining the level range through a preset threshold range.

[0059] Once the query is successful, the corresponding initial transport humidity level will be output, thus constructing an automated conversion path from the physiological state of agricultural products to environmental control parameters, providing an immediate and accurate decision-making basis for humidity control during cold chain transportation.

[0060] Step S32: Based on historical transportation quality data of agricultural products, the confidence level of the initial transportation humidity level is calibrated to obtain the transportation humidity level.

[0061] Statistical analysis was conducted on the achievement of agricultural product quality indicators under the same initial transportation humidity level in historical transportation quality data, including the actual distribution of key quality parameters such as spoilage rate and weight loss rate. Then, the posterior probability of obtaining the expected transportation quality under the initial humidity level was calculated using a Bayesian update algorithm, and this posterior probability was used as the confidence weight of the initial humidity level.

[0062] When the confidence level is lower than a preset threshold, the initial transportation humidity level is fine-tuned based on the actual humidity control values ​​corresponding to similar moisture conditions in historical high-quality transportation cases. The final output is a transportation humidity level that has been validated and optimized through practice. In this step, by combining theoretical mapping relationships with actual transportation effects, the humidity level decision-making conforms to physiological laws and has been tested in practice, effectively improving the accuracy and reliability of transportation humidity control.

[0063] The specific creation process of the target mapping relation library includes steps S311 to S312: Step S311: Obtain historical moisture status assessment results and establish an initial mapping relationship library between historical moisture status assessment results and transportation humidity levels.

[0064] By extracting historically accurate moisture status assessment results and corresponding actual transportation humidity control records from successful historical transportation cases, cluster analysis is used to classify similar moisture status assessment results based on this data, and an optimal transportation humidity level is assigned to each category. This constructs an initial mapping relationship library with moisture status category as the index and recommended transportation humidity level as the value, thus systematically integrating discrete historical experience data into a structured knowledge base, providing a reliable, practice-verified basis for the conversion of moisture status to transportation humidity level.

[0065] Step S312: Based on the real-time parameters of the current cold chain logistics transportation environment, the initial mapping relation library is weighted to obtain the target mapping relation library.

[0066] Real-time parameters such as temperature fluctuation amplitude and airflow velocity distribution in the current cold chain logistics transportation environment are collected. The influence weight of each real-time parameter on the humidity control effect is calculated using the entropy weight method. Then, based on the real-time parameters and their corresponding influence weights, the mapping relationship between each historical moisture status assessment result and the transportation humidity level in the initial mapping relationship library is calculated, along with its applicability weight in the current cold chain logistics transportation environment. Finally, the mapping relationships in the initial mapping relationship library are dynamically weighted according to the applicability weights to generate a target mapping relationship library that adapts to specific transportation environment conditions. This enables the mapping relationship between moisture status assessment results and transportation humidity levels to have environmental adaptability and dynamically optimize the humidity control strategy according to changes in the actual transportation environment.

[0067] This application also provides a humidity monitoring system for the cold chain logistics transportation of agricultural products, referring to... Figure 2 As shown, it includes: The data acquisition module 10 is used to collect microenvironmental humidity data on the surface of agricultural products and visual morphological data characterizing the moisture state of agricultural products. The data calculation module 20 is used to calculate the real-time moisture status assessment results of agricultural products based on microenvironmental humidity data and visual morphology data. The data mapping module 30 is used to map the moisture status assessment results to the transportation humidity level of agricultural products.

[0068] Optionally, the data acquisition module 10 is also used for: The microenvironment humidity data of agricultural product surfaces is collected using a flexible humidity sensor array; Initial visual data characterizing the moisture state of agricultural products is acquired using a multispectral camera. Based on a preset image segmentation algorithm, regions of interest (ROIs) on the surface of agricultural products are extracted from the initial visual data. Texture feature analysis and morphological quantization are then performed on the ROIs to obtain visual morphological data.

[0069] Optionally, the data acquisition module 10 is also used for: Texture feature analysis is performed on the region of interest to obtain multi-scale texture feature vectors; Frequency domain texture features are extracted from multi-scale texture feature vectors, and principal component analysis is performed on the frequency domain texture features to reduce their dimensionality, resulting in a subset of texture features. Morphological quantization is also performed on the region of interest to obtain a set of morphological feature parameters. The texture feature subset and the morphological feature parameter set are fused to obtain visual morphological data.

[0070] Optionally, the data calculation module 20 is also used for: Based on a pre-set deep neural network, humidity depth features are extracted from microenvironment humidity data, and visual depth features are extracted from visual morphology data through a pre-set convolutional neural network. Humidity depth features and visual depth features are fused to obtain multimodal fusion features; The multimodal fusion features are input into a preset moisture state assessment model to calculate the moisture state assessment results.

[0071] Optionally, the model building module 40 is also used for: Acquire historical training sample sets under different moisture conditions. The historical training sample sets consist of historical microenvironmental humidity data and historical visual morphology data of agricultural products. The target training sample set is obtained by expanding the historical training sample set through adversarial operations; Deep learning is performed on the target training sample set to obtain an initial training model. Then, the weights of the multimodal features in the initial training model are optimized to obtain a preset moisture state assessment model.

[0072] Optionally, the data mapping module 30 is also used for: Obtain historical moisture status assessment results and establish an initial mapping relationship library between historical moisture status assessment results and transportation humidity levels; Based on real-time parameters of the current cold chain logistics transportation environment, the initial mapping relation library is weighted to obtain the target mapping relation library.

[0073] Optionally, the data mapping module 30 is also used for: In the target mapping database, query the initial transport humidity level corresponding to the moisture status assessment result; Based on historical transportation quality data of agricultural products, the initial transportation humidity level is calibrated with confidence to obtain the transportation humidity level.

[0074] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A method for monitoring humidity during the cold chain logistics transportation of agricultural products, characterized in that, include: Collect microenvironmental humidity data on the surface of agricultural products and visual morphological data characterizing the moisture state of agricultural products; Based on the microenvironment humidity data and the visual morphology data, the real-time moisture status assessment result of agricultural products is calculated. The moisture status assessment results are mapped to the transportation humidity level of the agricultural products; The steps of collecting microenvironmental humidity data on the surface of agricultural products and visual morphological data characterizing the moisture state of agricultural products include: The microenvironmental humidity data of the agricultural product surface is collected using a flexible humidity sensor array; Initial visual data characterizing the moisture state of the agricultural products is acquired using a multispectral camera. Based on a preset image segmentation algorithm, the region of interest on the surface of the agricultural products is extracted from the initial visual data, and texture feature analysis is performed on the region of interest to obtain a multi-scale texture feature vector. Frequency domain texture features are extracted from the multi-scale texture feature vector, and principal component analysis is performed on the frequency domain texture features to reduce their dimensionality, resulting in a subset of texture features. Morphological quantization is then performed on the region of interest to obtain a set of morphological feature parameters. The visual morphological data is obtained by fusing the texture feature subset and the morphological feature parameter set.

2. The method for monitoring humidity during the cold chain logistics transportation of agricultural products according to claim 1, characterized in that, The step of calculating the real-time moisture status assessment result of agricultural products based on the microenvironment humidity data and the visual morphology data includes: Based on a preset deep neural network, humidity depth features are extracted from the microenvironment humidity data, and visual depth features are extracted from the visual morphology data through a preset convolutional neural network. The humidity depth feature and the visual depth feature are fused to obtain a multimodal fusion feature; The multimodal fusion features are input into a preset moisture state assessment model to calculate the moisture state assessment result.

3. The method for monitoring humidity during the cold chain logistics transportation of agricultural products according to claim 2, characterized in that, Before the step of inputting the multimodal fusion features into the preset moisture state assessment model, the following steps are included: Obtain a historical training sample set under different moisture conditions, wherein the historical training sample set consists of historical microenvironmental humidity data and historical visual morphology data of the agricultural product. The target training sample set is obtained by expanding the historical training sample set through adversarial operations; After performing deep learning on the target training sample set to obtain an initial training model, the multimodal feature weights in the initial training model are optimized to obtain the preset moisture state assessment model.

4. The method for monitoring humidity during the cold chain logistics transportation of agricultural products according to claim 1, characterized in that, Prior to the step of mapping the moisture state assessment result to the transport humidity level of the agricultural product, the following are included: Obtain historical moisture status assessment results and establish an initial mapping relationship library between the historical moisture status assessment results and the transportation humidity level; Based on real-time parameters of the current cold chain logistics transportation environment, the initial mapping relationship library is weighted to obtain the target mapping relationship library.

5. The method for monitoring humidity during the cold chain logistics transportation of agricultural products according to claim 4, characterized in that, The step of mapping the moisture state assessment result to the transport humidity level of the agricultural product includes: Query the target mapping database to find the initial transport humidity level corresponding to the moisture state assessment result; Based on the historical transportation quality data of the agricultural products, the initial transportation humidity level is calibrated with confidence to obtain the transportation humidity level.

6. A humidity monitoring system for the cold chain logistics transportation of agricultural products, characterized in that, The humidity monitoring system for the cold chain logistics transportation process of agricultural products is applied to the humidity monitoring method for the cold chain logistics transportation process of agricultural products as described in any one of claims 1 to 5, including: The data acquisition module is used to collect microenvironmental humidity data on the surface of agricultural products and visual morphological data characterizing the moisture state of the agricultural products. The data calculation module is used to calculate the real-time moisture status assessment result of the agricultural product based on the microenvironment humidity data and the visual morphology data. The data mapping module is used to map the moisture status assessment results to the transportation humidity level of the agricultural products; The steps of collecting microenvironmental humidity data on the surface of agricultural products and visual morphological data characterizing the moisture state of the agricultural products include: The microenvironmental humidity data of the agricultural product surface is collected using a flexible humidity sensor array; Initial visual data characterizing the moisture state of the agricultural products is acquired using a multispectral camera. Based on a preset image segmentation algorithm, the region of interest on the surface of the agricultural products is extracted from the initial visual data, and texture feature analysis is performed on the region of interest to obtain a multi-scale texture feature vector. Frequency domain texture features are extracted from the multi-scale texture feature vector, and principal component analysis is performed on the frequency domain texture features to reduce their dimensionality, resulting in a subset of texture features. Morphological quantization is then performed on the region of interest to obtain a set of morphological feature parameters. The visual morphological data is obtained by fusing the texture feature subset and the morphological feature parameter set.

Images