Cold chain transportation commodity quality monitoring system and method based on image analysis technology

Abstract

The invention relates to the technical field of cold-chain transportation monitoring, and particularly discloses a cold-chain transportation commodity quality monitoring system and method based on an image analysis technology, which can perform defogging processing on a fruit state image after the fruit state image of a cold-chain fruit is collected through a camera, so that the quality of the cold-chain transportation commodity is monitored. And inputting the defogged fruit state image into an image processing and analysis algorithm based on artificial intelligence and deep learning for analysis so as to automatically learn and capture surface state shallow layer-semantic fine granularity interaction significant fusion features in the fruit state image, and decoding to determine a decoded value of a fruit deterioration evaluation coefficient. Therefore, monitoring and management of the quality of the fruits transported by the cold-chain logistics can be realized in a more intelligent manner, so that the deteriorated cold-chain fruits can be found and treated in time, and the freshness keeping level of the fruits in the cold-chain logistics transportation can be improved.

Description

Technical Field

[0001] The present application relates to the field of cold chain transportation monitoring technology, and more specifically, to a cold chain transportation commodity quality monitoring system and method based on image analysis technology. Background Art

[0002] With the improvement of living standards, consumers' demand for fresh fruits is increasing, and they are also paying more and more attention to the safety and quality of food. In the existing cold chain logistics technology, the quality monitoring of perishable foods such as fruits mainly relies on regular manual inspections and simple environmental parameter monitoring, such as temperature and humidity. This method is not only inefficient, but also difficult to ensure the accuracy and timeliness of monitoring, which increases the risk of fruit quality degradation during the cold chain logistics process.

[0003] Chinese patent CN112978128A proposes a cold chain logistics and transportation commodity quality monitoring and management system based on big data and image analysis technology. After collecting fruit images, it can segment and splice feature areas and remove background images. Then, the processed fruit images are sent to the modeling and analysis server for further analysis to obtain information such as the area of ​​the fruit, the browning area, and calculate the proportion of the browning area. By analyzing the water loss rate, hardness changes, and environmental parameters (temperature, humidity, ethylene concentration, etc.) of the fruit, the fruit deterioration assessment coefficient is calculated to obtain the quality status of the fruit in the cold chain logistics process.

[0004] In the above-mentioned cold chain logistics and transportation commodity quality monitoring and management system, although it is possible to evaluate the quality changes of fruits through image analysis and multi-parameter monitoring, it increases the cost of cold chain logistics and cannot effectively capture the subtle features of the fruit surface, such as color changes, spots and semantics, which are crucial for evaluating fruit quality and can be perceived and evaluated in the early stage of fruit deterioration. In addition, the above-mentioned monitoring and management system lacks sufficient adaptability when monitoring the quality of different types of fruits under different lighting conditions and environmental conditions, that is, when facing different environments and different types of fruits, it is necessary to manually adjust parameters or models, which makes the process of monitoring the quality of cold chain logistics and transportation fruits more cumbersome and affects the accuracy of fruit deterioration detection.

[0005] Therefore, an optimized cold chain transportation commodity quality monitoring system is desired. Summary of the invention

[0006] The present application provides a cold chain transportation commodity quality monitoring system and method based on image analysis technology, which can monitor and manage the quality of fruits in cold chain logistics transportation in a more intelligent way, so as to timely detect and deal with spoiled cold chain fruits, and help improve the freshness of fruits in cold chain logistics transportation.

[0007] In a first aspect, a cold chain transportation commodity quality monitoring system based on image analysis technology is provided, including: a fruit state image acquisition module, used to obtain a fruit state image of a monitored cold chain fruit acquired by a camera; a fruit state image defogging processing module, used to perform image defogging on the fruit state image to obtain a defogged fruit state image; a fruit surface state multidimensional feature extraction module, used to extract texture features and semantic features based on the fruit surface of the defogged fruit state image to obtain a fruit surface texture feature map and a fruit surface state semantic feature map; a fruit surface state multidimensional feature fine-grained interactive fusion module, used to input the fruit surface texture feature map and the fruit surface state semantic feature map into a bidirectional global attention joint perception module based on fine-grained deconstruction to obtain a fruit surface state shallow-semantic fine-grained interactive significant fusion feature map; a fruit deterioration assessment coefficient generation module, used to determine a decoding value of a fruit deterioration assessment coefficient based on the fruit surface state shallow-semantic fine-grained interactive significant fusion feature map.

[0008] In a possible implementation, the fruit surface state multidimensional feature extraction module includes: a fruit surface texture feature capture unit, used to input the dehazed fruit state image into a fruit surface texture feature extractor based on a first atrous convolutional neural network model to obtain the fruit surface texture feature map; a fruit surface state semantic feature capture unit, used to input the dehazed fruit state image into a fruit surface state semantic feature extractor based on a second atrous convolutional neural network model to obtain the fruit surface state semantic feature map.

[0009] In a possible implementation, the fruit surface state multidimensional feature fine-grained interactive fusion module includes: a fruit surface state multidimensional feature fine-grained deconstruction unit, which is used to perform feature fine-grained deconstruction on the fruit surface texture feature map and the fruit surface state semantic feature map to obtain a set of local feature matrices of fruit surface texture channel dimension and a set of local feature matrices of fruit surface state semantic channel dimension; a first unidirectional global attention interactive processing unit, which is used to use each local feature matrix of fruit surface texture channel dimension in the set of local feature matrices of fruit surface texture channel dimension as a query feature matrix, and use the set of local feature matrices of fruit surface state semantic channel dimension as a set of key feature matrices, and input the query feature matrix and the set of key feature matrices into the unidirectional global attention interactive module based on the first converter structure to obtain a set of local feature matrices of fruit surface texture channel dimension optimized by unidirectional global attention; a second unidirectional global attention interactive processing unit, which is used to use each local feature matrix of fruit surface texture channel dimension in the set of local feature matrices of fruit surface state semantic channel dimension as a query feature matrix, and use the set of local feature matrices of fruit surface state semantic channel dimension as a key feature matrix, and input the query feature matrix and the set of key feature matrices into the unidirectional global attention interactive module based on the first converter structure to obtain a set of local feature matrices of fruit surface texture channel dimension optimized by unidirectional global attention; The local feature matrix of the semantic channel dimension of the fruit surface state is used as the query feature matrix, and the set of the local feature matrix of the fruit surface texture channel dimension is used as the set of key feature matrices. The query feature matrix and the set of the key feature matrix are input into the one-way global attention interaction module based on the second converter structure to obtain the set of local feature matrices of the semantic channel dimension of the fruit surface state optimized by one-way global attention; the fruit surface state coupling unit is used to couple the set of local feature matrices of the one-way global attention optimized fruit surface texture channel dimension and the set of local feature matrices of the one-way global attention optimized fruit surface state semantic channel dimension along the channel dimension to obtain a one-way global interaction optimized fruit surface texture feature map and a one-way global interaction optimized fruit surface state semantic feature map; the fruit surface state multidimensional feature weighted sum unit is used to calculate the position-weighted sum between the one-way global interaction optimized fruit surface texture feature map and the one-way global interaction optimized fruit surface state semantic feature map to obtain the shallow-semantic fine-grained interactive significant fusion feature map of the fruit surface state.

[0010] In a possible implementation, the first unidirectional global attention interaction processing unit is used to: select a predetermined fruit surface texture channel dimension local feature matrix from the set of fruit surface texture channel dimension local feature matrices as a query feature matrix; calculate the product of the predetermined fruit surface texture channel dimension local feature matrix and the transposed matrices of each of the fruit surface state semantic channel dimension local feature matrices in the set of fruit surface state semantic channel dimension local feature matrices to obtain a set of fruit surface state texture-semantic channel dimension local semantic interaction feature matrices; divide each of the fruit surface state texture-semantic channel dimension local semantic interaction feature matrices in the set of the fruit surface state texture-semantic channel dimension local semantic interaction feature matrices by the scale square root of the predetermined fruit surface texture channel dimension local feature matrix according to the position point, and then use The function performs soft maximum normalization processing on each feature matrix in the obtained set of feature matrices to obtain a set of local weight matrices of the fruit surface texture channel dimension; using the set of local weight matrices of the fruit surface texture channel dimension as weighted weights, calculates the position-weighted sum of the local feature matrices of the fruit surface texture channel dimension in the set of local feature matrices of the fruit surface texture channel dimension to obtain the unidirectional global attention optimized local feature matrix of the fruit surface texture channel dimension.

[0011] In a possible implementation, the second unidirectional global attention interaction processing unit is used to: select a predetermined fruit surface state semantic channel dimension local feature matrix from the set of fruit surface state semantic channel dimension local feature matrices as a query feature matrix; calculate the product of the predetermined fruit surface state semantic channel dimension local feature matrix and the transposed matrices of each of the fruit surface texture channel dimension local feature matrices in the set of fruit surface texture channel dimension local feature matrices to obtain a set of fruit surface state semantic-texture channel dimension local semantic interaction feature matrices; and transform each fruit surface state in the set of fruit surface state semantic-texture channel dimension local semantic interaction feature matrices into a query feature matrix. The local semantic interaction feature matrix of the semantic-texture channel dimension is divided by the scale square root of the predetermined local feature matrix of the fruit surface state semantic channel dimension at each position point, and then the softmax function is used to perform soft maximum normalization processing on each feature matrix in the set of obtained feature matrices to obtain a set of local weight matrices of the fruit surface state semantic channel dimension; using the set of local weight matrices of the fruit surface state semantic channel dimension as weighted weights, the position-weighted sum of each local feature matrix of the fruit surface state semantic channel dimension in the set of local feature matrices of the fruit surface state semantic channel dimension is calculated to obtain the unidirectional global attention optimized local feature matrix of the fruit surface state semantic channel dimension.

[0012] In a possible implementation, the fruit deterioration assessment coefficient generation module is used to: input the shallow-semantic fine-grained interactive significant fusion feature map of the fruit surface state into a decoder-based fruit deterioration assessment module to obtain a decoded value of the fruit deterioration assessment coefficient.

[0013] In a possible implementation, the fruit deterioration assessment coefficient generation module is used to: flatten the feature of the shallow-semantic fine-grained interactive significant fusion feature map of the fruit surface state to obtain a shallow-semantic fine-grained interactive significant fusion feature vector of the fruit surface state; multiply the decoding weight matrix of the decoder by the shallow-semantic fine-grained interactive significant fusion feature vector of the fruit surface state to obtain a decoded shallow-semantic fine-grained interactive significant fusion feature vector of the fruit surface state; and accumulate and sum all eigenvalues ​​of the decoded shallow-semantic fine-grained interactive significant fusion feature vector of the fruit surface state to obtain a decoded value of the fruit deterioration assessment coefficient.

[0014] In a second aspect, a method for monitoring the quality of cold chain transported goods based on image analysis technology is provided, comprising: obtaining a fruit state image of the monitored cold chain fruit collected by a camera; performing image defogging on the fruit state image to obtain a defogged fruit state image; performing texture feature and semantic feature extraction based on the fruit surface on the defogged fruit state image to obtain a fruit surface texture feature map and a fruit surface state semantic feature map; inputting the fruit surface texture feature map and the fruit surface state semantic feature map into a bidirectional global attention joint perception module based on fine-grained deconstruction to obtain a fruit surface state shallow-semantic fine-grained interactive significant fusion feature map; and determining a decoding value of a fruit deterioration assessment coefficient based on the fruit surface state shallow-semantic fine-grained interactive significant fusion feature map.

[0015] In a possible implementation, texture features and semantic features based on the fruit surface are extracted from the dehazed fruit state image to obtain a fruit surface texture feature map and a fruit surface state semantic feature map, including: inputting the dehazed fruit state image into a fruit surface texture feature extractor based on a first atrous convolutional neural network model to obtain the fruit surface texture feature map; inputting the dehazed fruit state image into a fruit surface state semantic feature extractor based on a second atrous convolutional neural network model to obtain the fruit surface state semantic feature map.

[0016] In a possible implementation, based on the shallow-semantic fine-grained interactive significant fusion feature map of the fruit surface state, a decoded value of the fruit deterioration assessment coefficient is determined, including: inputting the shallow-semantic fine-grained interactive significant fusion feature map of the fruit surface state into a decoder-based fruit deterioration assessment module to obtain a decoded value of the fruit deterioration assessment coefficient.

[0017] The present application provides a cold chain transportation commodity quality monitoring system and method based on image analysis technology, which can collect the fruit state image of cold chain fruit through a camera, defog the fruit state image, and input the defogged fruit state image into an image processing and analysis algorithm based on artificial intelligence and deep learning for analysis, so as to automatically learn and capture the surface state shallow-semantic fine-grained interactive significant fusion features in the fruit state image, and decode it to determine the decoded value of the fruit deterioration assessment coefficient. In this way, the monitoring and management of the quality of fruits in cold chain logistics transportation can be realized in a more intelligent way, so as to timely discover and deal with spoiled cold chain fruits, which helps to improve the level of freshness retention of fruits in cold chain logistics transportation. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a schematic block diagram of a cold chain transportation commodity quality monitoring system based on image analysis technology according to an embodiment of the present application.

[0019] Figure 2 This is a data flow diagram of a cold chain transportation commodity quality monitoring system based on image analysis technology in an embodiment of the present application.

[0020] Figure 3 This is a schematic block diagram of a multi-dimensional feature extraction module for fruit surface state in a cold chain transportation commodity quality monitoring system based on image analysis technology in an embodiment of the present application.

[0021] Figure 4 This is a schematic block diagram of a fine-grained interactive fusion module of multi-dimensional features of fruit surface state in a cold chain transportation commodity quality monitoring system based on image analysis technology in an embodiment of the present application.

[0022] Figure 5 This is a schematic flow chart of a method for monitoring the quality of cold chain transported goods based on image analysis technology according to an embodiment of the present application. DETAILED DESCRIPTION

[0023] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work also fall within the scope of protection of the present application.

[0024] Although the current cold chain commodity quality monitoring system can evaluate the quality changes of fruits through images and other parameters, this not only increases the cost of cold chain logistics, but also makes it difficult to detect these changes in the early stages of fruit deterioration due to the system's inability to capture subtle features such as color changes and spots on the surface of fruits. In addition, when environmental conditions change or the types of fruits being monitored are different, the existing system often requires manual adjustment of parameters to adapt to the new situation. This lack of flexibility makes the fruit quality monitoring process complicated and may affect the accuracy of detecting fruit deterioration.

[0025] In view of the above technical problems, in the technical solution of the present application, a cold chain transport commodity quality monitoring system based on image analysis technology is proposed, which can collect the fruit state image of the cold chain fruit through the camera, defog the fruit state image, and input the defogged fruit state image into the image processing and analysis algorithm based on artificial intelligence and deep learning for analysis, so as to automatically learn and capture the complex features in the fruit state image. In particular, the processing based on the algorithm of artificial intelligence and deep learning can make the system have better generalization ability, and can perform well on different types of fruits under different lighting conditions. It is worth mentioning that once the deep learning model is trained, it can quickly and accurately identify the slight changes on the surface of the fruit, such as slight spots, browning or damage, as well as the deep semantic information of the surface state of the fruit, which helps to find the problem of fruit quality change earlier, so as to take timely measures to reduce losses. In addition, the deep learning model can also continuously improve its detection accuracy as more data accumulates, and is more suitable for large-scale and diversified cold chain logistics environments. In this way, the quality of fruits transported in cold chain logistics can be monitored and managed in a more intelligent way, so as to timely detect and deal with spoiled cold chain fruits, which helps to improve the freshness of fruits in cold chain logistics transportation.

[0026] Figure 1 This is a schematic block diagram of a cold chain transportation commodity quality monitoring system based on image analysis technology according to an embodiment of the present application. Figure 2 This is a data flow diagram of the cold chain transportation commodity quality monitoring system based on image analysis technology in an embodiment of the present application. Figure 1 and Figure 2As shown, the cold chain transportation commodity quality monitoring system based on image analysis technology includes: a fruit state image acquisition module 110, which is used to obtain the fruit state image of the monitored cold chain fruit acquired by the camera; a fruit state image defogging processing module 120, which is used to perform image defogging on the fruit state image to obtain the defogged fruit state image; a fruit surface state multidimensional feature extraction module 130, which is used to extract the texture feature and semantic feature based on the fruit surface of the defogged fruit state image to obtain a fruit surface texture feature map and a fruit surface state semantic feature map; a fruit surface state multidimensional feature fine-grained interactive fusion module 140, which is used to input the fruit surface texture feature map and the fruit surface state semantic feature map into a two-way global attention joint perception module based on fine-grained deconstruction to obtain a fruit surface state shallow-semantic fine-grained interactive significant fusion feature map; a fruit deterioration assessment coefficient generation module 150, which is used to determine the decoding value of the fruit deterioration assessment coefficient based on the fruit surface state shallow-semantic fine-grained interactive significant fusion feature map.

[0027] In the above-mentioned cold chain transportation commodity quality monitoring system based on image analysis technology, the fruit status image acquisition module 110 is used to obtain the fruit status image of the monitored cold chain fruit collected by the camera. That is, because the quality of fruits in cold chain logistics transportation is directly related to the food safety and satisfaction of the final consumers. By taking images of fruits in real time or regularly by the camera, image analysis technology can be used to monitor the status of the fruits, and then judge their freshness and quality. Compared with traditional manual inspection, this non-contact monitoring method is more efficient and accurate, and can be carried out continuously, which helps to discover any potential quality problems in time, so as to take measures quickly to prevent the expansion of losses.

[0028] Optionally, in one embodiment of the present application, the process of obtaining the fruit status image of the monitored cold chain fruit collected by the camera mainly includes the following steps: First, a professional camera suitable for environmental conditions is installed at the key node of the cold chain logistics. Then, the working parameters of the camera are set, such as the time interval of shooting, image resolution, etc. Finally, the camera is started to collect images at a fixed time or according to the set conditions. Through such a process, it is possible to ensure the acquisition of high-quality cold chain fruit status images, providing a reliable data basis for subsequent image analysis and fruit quality monitoring.

[0029] In the above-mentioned cold chain transportation commodity quality monitoring system based on image analysis technology, the fruit status image defogging processing module 120 is used to perform image defogging on the fruit status image to obtain a defogged fruit status image. It should be understood that, considering that during the cold chain logistics transportation process, especially in a refrigerated environment, the image captured by the camera may be obstructed by fog, steam or other forms. These factors will cause the image to be blurred and reduce the image quality, thereby affecting the accuracy of subsequent image analysis. Therefore, in order to improve the quality of the image and ensure that the subsequent processing steps can accurately analyze the status of the fruit, in the technical solution of the present application, the fruit status image is further defogged to obtain a defogged fruit status image.

[0030] Optionally, in one embodiment of the present application, a known image defogging algorithm can be used to process the acquired fruit status image. Common image defogging algorithms include but are not limited to dark channel priors, atmospheric scattering models, etc. These algorithms attempt to estimate the fog concentration in the image and remove the effects of fog in the image. When implementing the algorithm, the image formation model can be considered, that is, the formation of the image includes elements such as scene radiance, atmospheric light, and transmittance. The transmittance map and atmospheric light value are calculated by the algorithm, and a clear image is restored accordingly.

[0031] In the above-mentioned cold chain transportation commodity quality monitoring system based on image analysis technology, the fruit surface state multidimensional feature extraction module 130 is used to extract the texture features and semantic features based on the fruit surface of the defogging fruit state image to obtain a fruit surface texture feature map and a fruit surface state semantic feature map. It should be understood that in order to obtain feature information of different levels and categories about the fruit state, so as to help more comprehensively understand the quality of the fruit in the cold chain logistics transportation process, in the technical solution of the present application, the defogging fruit state image is further input into the fruit surface texture feature extractor based on the first hole convolutional neural network model to obtain the fruit surface texture feature map, and the defogging fruit state image is input into the fruit surface state semantic feature extractor based on the second hole convolutional neural network model to obtain the fruit surface state semantic feature map. It should be understood that the use of the fruit surface texture feature extractor based on the first hole convolutional neural network model to extract the texture features of the fruit surface can capture the texture details of the fruit surface state, such as spots, cracks, depressions and other subtle physical characteristics. These texture features are very important for evaluating the integrity, freshness and health of fruits. In addition, the fruit surface state semantic feature extractor based on the second atrous convolutional neural network model can extract the state semantic features of the fruit surface. This semantic feature extraction aims to understand the meaning information of the fruit surface state contained in the image, such as the color semantics of the fruit, whether there are signs of rot, whether there is mildew, etc. Semantic feature extraction can help identify whether the fruit has begun to deteriorate, whether there is browning, etc., which helps to detect spoiled cold chain fruits early and take corresponding measures.

[0032] Figure 3 This is a schematic block diagram of a multi-dimensional feature extraction module for fruit surface state in a cold chain transportation commodity quality monitoring system based on image analysis technology in an embodiment of the present application. Figure 3As shown, optionally, in one embodiment of the present application, the fruit surface state multidimensional feature extraction module 130 includes: a fruit surface texture feature capture unit 131, which is used to input the defogging fruit state image into a fruit surface texture feature extractor based on a first hole convolutional neural network model to obtain the fruit surface texture feature map; a fruit surface state semantic feature capture unit 132, which is used to input the defogging fruit state image into a fruit surface state semantic feature extractor based on a second hole convolutional neural network model to obtain the fruit surface state semantic feature map. Here, hole convolution (also called dilated convolution or atrous convolution) is a special convolution operation that expands the receptive field by inserting holes (i.e., jumping a certain number of pixels) between the elements of the convolution kernel without increasing the number of parameters or reducing the downsampling rate. This method is particularly suitable for image segmentation tasks or application scenarios that require retaining spatial resolution because it can capture greater contextual information without changing the input size.

[0033] Optionally, in a specific embodiment of the present application, the defogging fruit state image is input into a fruit surface texture feature extractor based on a first dilated convolutional neural network model to obtain the fruit surface texture feature map, including: taking the defogging fruit state image as input. Constructing a network containing multiple dilated convolutional layers. These layers use different dilation rates to capture texture information at different scales. Each layer of dilated convolution generates a new set of feature maps, which capture texture information at different levels, such as spots, cracks, depressions, etc., and finally obtain the fruit surface texture feature map. The defogging fruit state image is input into a fruit surface state semantic feature extractor based on a second dilated convolutional neural network model to obtain the fruit surface state semantic feature map, including: similarly, taking the defogging fruit state image as input. Constructing another network containing multiple dilated convolutional layers, which also use different dilation rates. Each layer of dilated convolution generates a new set of feature maps, which capture semantic information at different levels, such as the color semantics, signs of decay, mildew, etc. of the fruit, and finally obtain the semantic feature map of the surface state of the fruit.

[0034] In the above-mentioned cold chain transportation commodity quality monitoring system based on image analysis technology, the fruit surface state multidimensional feature fine-grained interactive fusion module 140 is used to input the fruit surface texture feature map and the fruit surface state semantic feature map into the two-way global attention joint perception module based on fine-grained deconstruction to obtain the shallow-semantic fine-grained interactive significant fusion feature map of the fruit surface state. It should be understood that in the cold chain logistics process, various subtle changes may occur on the fruit surface, such as small-scale damage caused by collision, slight color changes, or spots and cracks on the fruit surface, etc. These subtle information on the fruit surface are presented in its surface texture features. Therefore, in the process of detecting fruit deterioration, these detailed texture information needs to be combined with semantic features to carry out subsequent fruit deterioration assessment to further improve the accuracy of the calculation of the fruit deterioration assessment coefficient. Based on this, in order to achieve a deep understanding of the surface state of the fruit and accurately calculate the fruit deterioration assessment coefficient, in the technical solution of the present application, the fruit surface texture feature map and the fruit surface state semantic feature map are further input into the two-way global attention joint perception module based on fine-grained deconstruction to obtain the shallow-semantic fine-grained interaction significant fusion feature map of the fruit surface state. In particular, the two-way global attention joint perception module based on fine-grained deconstruction can improve the model's semantic joint perception ability of the fruit surface state through feature fine-grained deconstruction, global fine-grained interaction based on attention mechanism, feature coupling and feature fusion, thereby facilitating a more accurate calculation of the fruit deterioration assessment coefficient, so as to support better logistics management and cold chain logistics transportation fruit quality control decisions.

[0035] Figure 4 This is a schematic block diagram of a fine-grained interactive fusion module of multi-dimensional features of fruit surface state in a cold chain transportation commodity quality monitoring system based on image analysis technology in an embodiment of the present application. Figure 4As shown, optionally, in one embodiment of the present application, the fruit surface state multidimensional feature fine-grained interactive fusion module 140 includes: a fruit surface state multidimensional feature fine-grained deconstruction unit 141, which is used to perform feature fine-grained deconstruction on the fruit surface texture feature map and the fruit surface state semantic feature map to obtain a set of fruit surface texture channel dimension local feature matrices and a set of fruit surface state semantic channel dimension local feature matrices; a first unidirectional global attention interactive processing unit 142, which is used to respectively use each fruit surface texture channel dimension local feature matrix in the set of fruit surface texture channel dimension local feature matrices as a query feature matrix, and use the set of fruit surface state semantic channel dimension local feature matrices as a set of key feature matrices, and input the query feature matrix and the set of key feature matrices into the unidirectional global attention interactive module based on the first converter structure to obtain a set of unidirectional global attention optimized fruit surface texture channel dimension local feature matrices; a second unidirectional global attention interactive processing unit 143, which is used to respectively use the fruit surface state semantic channel dimension local feature matrix of each fruit surface texture channel dimension local feature matrix as a query feature matrix, and use the set of fruit surface state semantic channel dimension local feature matrix as a key feature matrix. The local feature matrices of the semantic channel dimensions of each fruit surface state in the set are used as query feature matrices, and the set of local feature matrices of the fruit surface texture channel dimension is used as the set of key feature matrices. The query feature matrix and the set of key feature matrices are input into the one-way global attention interaction module based on the second converter structure to obtain a set of local feature matrices of the semantic channel dimension of the fruit surface state optimized by one-way global attention; the fruit surface state coupling unit 144 is used to couple the set of local feature matrices of the one-way global attention optimized fruit surface texture channel dimension and the set of local feature matrices of the one-way global attention optimized fruit surface state semantic channel dimension along the channel dimension to obtain a one-way global interaction optimized fruit surface texture feature map and a one-way global interaction optimized fruit surface state semantic feature map; the fruit surface state multidimensional feature weighted sum unit 145 is used to calculate the position-weighted sum between the one-way global interaction optimized fruit surface texture feature map and the one-way global interaction optimized fruit surface state semantic feature map to obtain the shallow-semantic fine-grained interactive significant fusion feature map of the fruit surface state.

[0036] Optionally, in one embodiment of the present application, the first unidirectional global attention interaction processing unit is used to: select a predetermined fruit surface texture channel dimension local feature matrix from the set of fruit surface texture channel dimension local feature matrices as a query feature matrix; calculate the product of the predetermined fruit surface texture channel dimension local feature matrix and the transposed matrices of each of the fruit surface state semantic channel dimension local feature matrices in the set of fruit surface state semantic channel dimension local feature matrices to obtain a set of fruit surface state texture-semantic channel dimension local semantic interaction feature matrices; and convert each of the fruit surface state texture-semantic channel dimension local semantic interaction feature matrices into a query feature matrix. After dividing the scale square root of the predetermined local feature matrix of the fruit surface texture channel dimension by the position point of each local semantic interaction feature matrix of the fruit surface state texture-semantic channel dimension, the softmax function is used to perform soft maximum normalization processing on each feature matrix in the set of obtained feature matrices to obtain a set of local weight matrices of the fruit surface texture channel dimension; using the set of local weight matrices of the fruit surface texture channel dimension as the weighted weight, the position-weighted sum of the local feature matrices of the fruit surface texture channel dimension in the set of local feature matrices of the fruit surface texture channel dimension is calculated to obtain the unidirectional global attention optimized local feature matrix of the fruit surface texture channel dimension.

[0037] Optionally, in one embodiment of the present application, the second unidirectional global attention interaction processing unit is used to: select a predetermined fruit surface state semantic channel dimension local feature matrix from the set of fruit surface state semantic channel dimension local feature matrices as a query feature matrix; calculate the product of the predetermined fruit surface state semantic channel dimension local feature matrix and the transposed matrices of each of the fruit surface texture channel dimension local feature matrices in the set of fruit surface texture channel dimension local feature matrices to obtain a set of fruit surface state semantic-texture channel dimension local semantic interaction feature matrices; divide each of the fruit surface state semantic-texture channel dimension local semantic interaction feature matrices in the set of the fruit surface state semantic-texture channel dimension local semantic interaction feature matrices by the scale square root of the predetermined fruit surface state semantic channel dimension local feature matrix according to the position point, and then use The function performs soft maximum normalization processing on each feature matrix in the obtained set of feature matrices to obtain a set of local weight matrices of the fruit surface state semantic channel dimension; using the set of local weight matrices of the fruit surface state semantic channel dimension as weighted weights, calculates the position-weighted sum of the local feature matrices of the fruit surface state semantic channel dimension in the set of local feature matrices of the fruit surface state semantic channel dimension to obtain the unidirectional global attention optimized local feature matrix of the fruit surface state semantic channel dimension.

[0038] In summary, in an embodiment of the present application, the fruit surface texture feature map and the fruit surface state semantic feature map are input into the fine-grained deconstruction-based bidirectional global attention joint perception module and processed with the following bidirectional global fine-grained feature joint perception formula to obtain the fruit surface state shallow-semantic fine-grained interactive significant fusion feature map; wherein the bidirectional global fine-grained feature joint perception formula is: ; in, and represent the fruit surface texture feature map and the fruit surface state semantic feature map respectively, It is a feature fine-grained deconstruction operation. They are respectively the first, second, and third dimensions along the channel dimension in the fruit surface texture feature map. and The local feature matrix of the fruit surface texture channel dimension, They are respectively the first, second, and third and The local feature matrix of the semantic channel dimension of the fruit surface state, For the The scale of the local feature matrix of the fruit surface texture channel dimension, is matrix multiplication, for function, represents the transpose of a vector, is the number of local feature matrices of the fruit surface texture channel dimension in the fruit surface texture feature map, For the The one-way global attention optimizes the local feature matrix of the fruit surface texture channel dimension corresponding to the local feature matrix of the fruit surface texture channel dimension. For the The one-way global attention optimizes the local feature matrix of the fruit surface state semantic channel dimension corresponding to the local feature matrix of the fruit surface texture channel dimension. Indicates that the features are coupled along the channel dimension, and They are one-way global interaction optimization of fruit surface texture feature map and one-way global interaction optimization of fruit surface state semantic feature map. and are weighted hyper parameters of the one-way global interactive optimization of the fruit surface texture feature map and the one-way global interactive optimization of the fruit surface state semantic feature map, respectively. It is a shallow-semantic fine-grained interactive significant fusion feature map of the fruit surface state.

[0039] Specifically, first, the fruit surface texture feature map and the fruit surface state semantic feature map are fine-grainedly deconstructed to decompose these feature maps into smaller local feature matrices (i.e., local features of the fruit surface state). Fine-grained deconstruction enables the model to analyze the different features and local information of the fruit surface state more carefully, providing rich input for the subsequent attention mechanism, so that the model can more accurately identify and process the key areas in the feature map, so as to better detect which parts of the fruit surface have problems, and provide a basis for fruit deterioration assessment. Then, each fruit surface texture channel dimension local feature matrix in the set of the fruit surface texture channel dimension local feature matrix is ​​used as a query feature matrix, and the set of the fruit surface state semantic channel dimension local feature matrix is ​​used as a set of key feature matrices. The query feature matrix and the set of the key feature matrix are input into a unidirectional global attention interaction module based on the first converter structure to obtain a set of unidirectional global attention optimized fruit surface texture channel dimension local feature matrices. At the same time, each local feature matrix of the fruit surface state semantic channel dimension in the set of local feature matrices of the fruit surface state semantic channel dimension is used as a query feature matrix, and the set of local feature matrices of the fruit surface texture channel dimension is used as a set of key feature matrices, and the query feature matrix and the set of key feature matrices are input into a unidirectional global attention interaction module based on the second converter structure to obtain a set of local feature matrices of the fruit surface texture channel dimension optimized by unidirectional global attention.

[0040] That is, after fine-grained deconstruction of the feature map, the unidirectional global attention interaction module uses the multi-head attention mechanism in the Transformer structure to measure the correlation between the local feature matrices of the texture features and semantic features of the fruit surface state, strengthens the relationship between the features, and enables the model to pay more attention to the feature areas rich in information.

[0041] In particular, in the technical solution of the present application, each local feature matrix of the fruit surface texture channel dimension of the fruit surface texture feature map can respectively perform unidirectional semantic interaction with all local feature matrices of the fruit surface state semantic channel dimension of the fruit surface state semantic feature map to more comprehensively capture the global semantic association information between each local feature matrix of the fruit surface texture channel dimension of the fruit surface texture feature map and the fruit surface state semantic feature map. Similarly, each local feature matrix of the fruit surface state semantic channel dimension of the fruit surface state semantic feature map can respectively perform unidirectional semantic interaction with all local feature matrices of the fruit surface texture channel dimension of the fruit surface texture feature map to more comprehensively capture the global semantic association information between each local feature matrix of the fruit surface state semantic channel dimension of the fruit surface texture feature map and the fruit surface texture feature map. It should be understandable that the effect of unidirectional global attention interaction is to enhance the fine-grained channel relationship within the feature map, so that the model can pay more attention to the key information in the feature map related to the fruit spoilage detection and identification tasks, improve the quality and discrimination of feature representation, and thus help improve the performance of the model in subsequent fruit spoilage assessment technical calculation tasks.

[0042] Then, through the coupling processing along the channel dimension, the local feature matrices of the fruit surface texture and semantics that have been optimized by attention can be recombined along the channel dimension to form a complete feature map, providing a complete feature representation for the final fusion. Finally, the position-weighted sum between the one-way global interactively optimized fruit surface texture feature map and the one-way global interactively optimized fruit surface state semantic feature map is calculated to obtain the fruit surface texture-semantic joint perception feature map. That is, the weighted sum operation combines the key information in the two feature maps to generate a fruit surface state joint perception feature representation containing rich interactive information. In particular, the final generated "fruit surface state shallow-semantic fine-grained interactive significant fusion feature map" is a result that integrates the texture and semantic feature information of the fruit surface state. This joint perception method allows the system to more comprehensively reflect the state of the fruit surface, which helps to more accurately evaluate the quality of the fruit, thereby supporting better cold chain fruit logistics management and decision-making.

[0043] In the above-mentioned cold chain transportation commodity quality monitoring system based on image analysis technology, the fruit deterioration assessment coefficient generation module 150 is used to determine the decoded value of the fruit deterioration assessment coefficient based on the shallow-semantic fine-grained interactive significant fusion feature map of the fruit surface state. Optionally, in one embodiment of the present application, the fruit deterioration assessment coefficient generation module is used to: input the shallow-semantic fine-grained interactive significant fusion feature map of the fruit surface state into the fruit deterioration assessment module based on the decoder to obtain the decoded value of the fruit deterioration assessment coefficient. In other words, the fine-grained interactive fusion feature between the shallow texture and semantics of the fruit surface state is used to evaluate the fruit deterioration, which helps to discover the problem of fruit quality changes earlier, so as to take timely measures to reduce losses. In this way, the monitoring and management of the quality of fruits in cold chain logistics transportation can be realized in a more intelligent way, so as to timely discover and deal with spoiled cold chain fruits, which helps to improve the freshness retention level of fruits in cold chain logistics transportation.

[0044] Optionally, in one embodiment of the present application, the fruit deterioration assessment coefficient generation module is used to: flatten the fruit surface state shallow-semantic fine-grained interactive significant fusion feature map to obtain a fruit surface state shallow-semantic fine-grained interactive significant fusion feature vector; multiply the decoding weight matrix of the decoder by the fruit surface state shallow-semantic fine-grained interactive significant fusion feature vector to obtain a decoded fruit surface state shallow-semantic fine-grained interactive significant fusion feature vector; and accumulate and sum all eigenvalues ​​of the decoded fruit surface state shallow-semantic fine-grained interactive significant fusion feature vector to obtain a decoded value of the fruit deterioration assessment coefficient.

[0045] Specifically, in the technical solution of the present application, the decoder is used to decode and regress the shallow-semantic fine-grained interactive significant fusion feature vector of the fruit surface state using the following decoding formula to obtain the decoded value of the fruit deterioration assessment coefficient, wherein the decoding formula is: ,in, is the shallow-semantic fine-grained interactive significant fusion feature vector of the fruit surface state, is the decoded value of the fruit deterioration assessment coefficient, is the decoding weight matrix, Represents matrix multiplication.

[0046] In a preferred example, the shallow-semantic fine-grained interactive significant fusion feature map of the fruit surface state is input into a decoder-based fruit deterioration assessment module to obtain a decoded value of the fruit deterioration assessment coefficient, including: calculating the sum of the absolute values ​​of each feature value of the shallow-semantic fine-grained interactive significant fusion feature map of the fruit surface state to obtain a first fruit surface state shallow-semantic fine-grained interactive significant fusion and modulation value, and calculating the square root of the sum of the squares of each feature value of the shallow-semantic fine-grained interactive significant fusion feature map of the fruit surface state to obtain a second fruit surface state shallow-semantic fine-grained interactive significant fusion and modulation value.

[0047] After dot-wise subtraction of the fruit surface state shallow-semantic fine-grained interactive significant fusion feature map and the second fruit surface state shallow-semantic fine-grained interactive significant fusion and modulation value, dot-wise multiplication is performed with the number of eigenvalues ​​of the fruit surface state shallow-semantic fine-grained interactive significant fusion feature map and the inverse of the first fruit surface state shallow-semantic fine-grained interactive significant fusion and modulation value, and the inverse of each eigenvalue is taken to obtain the first fruit surface state shallow-semantic fine-grained interactive significant fusion phase transition feature map.

[0048] After point-wise subtraction of the fruit surface state shallow-semantic fine-grained interactive significant fusion feature map and the first fruit surface state shallow-semantic fine-grained interactive significant fusion and modulation value, point-wise multiplication is performed with the square root of the number of eigenvalues ​​of the fruit surface state shallow-semantic fine-grained interactive significant fusion feature map and the reciprocal of the second fruit surface state shallow-semantic fine-grained interactive significant fusion and modulation value, and the reciprocal of each eigenvalue is taken to obtain the second fruit surface state shallow-semantic fine-grained interactive significant fusion phase transition feature map.

[0049] The first fruit surface state shallow-semantic fine-grained interactive significant fusion phase transition feature map is point-subtracted from the second fruit surface state shallow-semantic fine-grained interactive significant fusion phase transition feature map and the dot product feature map of the weighted hyperparameter to obtain an optimized fruit surface state shallow-semantic fine-grained interactive significant fusion feature map.

[0050] The optimized shallow-semantic fine-grained interactive significant fusion feature map of the fruit surface state is passed through a decoder-based fruit deterioration assessment module to obtain a decoded value of the fruit deterioration assessment coefficient.

[0051] Here, the shallow-semantic fine-grained interactive significant fusion feature map of the fruit surface state The optimization expression is: ; in, is the shallow-semantic fine-grained interactive significant fusion feature map of the fruit surface state, represents the set of real numbers, , , Respectively represent the width, length and height of the shallow-semantic fine-grained interactive significant fusion feature map of the fruit surface state, The shallow-semantic fine-grained interactive significant fusion feature map representing the surface state of the fruit The eigenvalues ​​at the positions, The number of eigenvalues ​​representing the shallow-semantic fine-grained interactive significant fusion feature map of the surface state of the fruit, Indicates the shallow-semantic fine-grained interaction and significant fusion and modulation value of the surface state of the first fruit, Indicates the second fruit surface state shallow-semantic fine-grained interaction significant fusion and modulation value, It means point multiplication by position. is matrix multiplication, It means subtracting by position. Indicates the calculation of the inverse of each eigenvalue of the feature map, Represents the shallow-semantic fine-grained interactive fusion phase transition feature map of the surface state of the first fruit, Represents the shallow-semantic fine-grained interactive fusion phase transition feature map of the second fruit surface state, represents the weighted hyperparameter, Represents the optimized shallow-semantic fine-grained interactive significant fusion feature map of the fruit surface state.

[0052] That is, the present application takes into account that the fruit surface texture feature map and the fruit surface state semantic feature map respectively represent the texture image semantic features and state image semantic features of the fruit state image at different scales and depths based on the convolutional neural network model. When they are passed through a bidirectional global attention joint perception module based on fine-grained deconstruction, the fine-grained deconstruction differences based on the distribution of cross-scale and cross-depth image semantic features will lead to differences in bidirectional global attention weights. Therefore, it is expected to improve the detail semantic joint perception aggregation expression effect of the fruit surface state shallow-semantic fine-grained interactive significant fusion feature map based on the attention differences of cross-scale and cross-depth image semantic features.

[0053] Therefore, in the preferred example, the difference of the different and modulated representations of the feature values ​​of the shallow-semantic fine-grained interactive significant fusion feature map of the fruit surface state relative to the overall feature set of the shallow-semantic fine-grained interactive significant fusion feature map of the fruit surface state is used as the semantic change intensity information, and a phase-like conversion corresponding to the position-based intensity modulation is performed through different and modulated representations, so that the aggregation enhancement of the semantic change phase perception can enhance the axial aggregation receptive field along the feature aggregation direction by performing a spatial translation operation based on alternating stacking under the set scale equalization of the shallow-semantic fine-grained interactive significant fusion feature map of the fruit surface state, thereby enhancing the perception effect of the aggregated semantics of the shallow-semantic fine-grained interactive significant fusion feature map of the fruit surface state for the detailed semantic changes, so as to improve the expression effect of the shallow-semantic fine-grained interactive significant fusion feature map of the fruit surface state, and improve the accuracy of the decoded value of the fruit deterioration assessment coefficient obtained by inputting it into the decoder-based fruit deterioration assessment module. In this way, the spoilage of fruits can be identified and evaluated more accurately, so that spoiled cold chain fruits can be discovered and handled in a timely manner, which helps to improve the freshness of fruits during cold chain logistics transportation.

[0054] In summary, the cold chain transportation commodity quality monitoring system based on image analysis technology according to the embodiment of the present application is explained, which can collect the fruit state image of the cold chain fruit through the camera, defog the fruit state image, and input the defogged fruit state image into the image processing and analysis algorithm based on artificial intelligence and deep learning for analysis, so as to automatically learn and capture the surface state shallow-semantic fine-grained interactive significant fusion features in the fruit state image, and decode it to determine the decoded value of the fruit deterioration assessment coefficient. In this way, the monitoring and management of the quality of fruits in cold chain logistics transportation can be realized in a more intelligent way, so as to timely discover and deal with spoiled cold chain fruits, which helps to improve the level of freshness retention of fruits in cold chain logistics transportation.

[0055] Figure 5 Schematic flow chart of the method for monitoring the quality of cold chain transported goods based on image analysis technology according to an embodiment of the present application. Figure 5As shown, a method for monitoring the quality of cold chain transportation commodities based on image analysis technology includes: S1, obtaining a fruit state image of the monitored cold chain fruit collected by a camera; S2, performing image defogging on the fruit state image to obtain a defogged fruit state image; S3, extracting texture features and semantic features based on the fruit surface on the defogged fruit state image to obtain a fruit surface texture feature map and a fruit surface state semantic feature map; S4, inputting the fruit surface texture feature map and the fruit surface state semantic feature map into a bidirectional global attention joint perception module based on fine-grained deconstruction to obtain a fruit surface state shallow-semantic fine-grained interactive significant fusion feature map; S5, determining a decoding value of a fruit deterioration assessment coefficient based on the fruit surface state shallow-semantic fine-grained interactive significant fusion feature map.

[0056] Optionally, in one embodiment of the present application, texture features and semantic features based on the fruit surface are extracted for the dehazed fruit state image to obtain a fruit surface texture feature map and a fruit surface state semantic feature map, including: inputting the dehazed fruit state image into a fruit surface texture feature extractor based on a first atrous convolutional neural network model to obtain the fruit surface texture feature map; inputting the dehazed fruit state image into a fruit surface state semantic feature extractor based on a second atrous convolutional neural network model to obtain the fruit surface state semantic feature map.

[0057] Optionally, in one embodiment of the present application, based on the shallow-semantic fine-grained interactive significant fusion feature map of the fruit surface state, a decoded value of the fruit deterioration assessment coefficient is determined, including: inputting the shallow-semantic fine-grained interactive significant fusion feature map of the fruit surface state into a decoder-based fruit deterioration assessment module to obtain a decoded value of the fruit deterioration assessment coefficient.

[0058] Here, those skilled in the art can understand that the specific operations of each step in the above-mentioned cold chain transportation commodity quality monitoring method based on image analysis technology have been referred to above. Figures 1 to 4 The description of the cold chain transportation commodity quality monitoring system based on image analysis technology has been introduced in detail, and therefore, its repeated description will be omitted.

[0059] The basic principle of the present invention is described above in conjunction with specific embodiments. However, it should be pointed out that the advantages, strengths, effects, etc. mentioned in the present invention are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. must be possessed by each embodiment of the present invention. In addition, the specific details of the above embodiments are only for the purpose of illustration and facilitation of understanding, rather than limitation, and the above details do not limit the present invention to being implemented by adopting the above specific details.

[0060] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, please refer to the relevant description of other embodiments. In the several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the module division is only a logical function division, and there may be other division methods in actual implementation. The modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0061] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0062] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference to a figure in a claim should not be considered as limiting the claim to which it relates.

[0063] In addition, it is obvious that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units stated in the device claim can also be implemented by one unit through software or hardware.

[0064] Finally, it should be noted that the above description has been given for the purpose of illustration and description. In addition, the above embodiments are only used to illustrate the technical solution of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A cold chain transportation commodity quality monitoring system based on image analysis technology, characterized in that: include: A fruit status image acquisition module, used to acquire a fruit status image of the monitored cold chain fruit acquired by a camera; The fruit state image defogging processing module is used to perform image defogging on the fruit state image to obtain the defogged fruit state image; the fruit surface state multidimensional feature extraction module is used to extract the texture features and semantic features based on the fruit surface of the defogged fruit state image to obtain a fruit surface texture feature map and a fruit surface state semantic feature map; the fruit surface state multidimensional feature fine-grained interactive fusion module is used to input the fruit surface texture feature map and the fruit surface state semantic feature map into a two-way global attention joint perception module based on fine-grained deconstruction to obtain a fruit surface state shallow-semantic fine-grained interactive significant fusion feature map; the fruit deterioration assessment coefficient generation module is used to determine the decoding value of the fruit deterioration assessment coefficient based on the fruit surface state shallow-semantic fine-grained interactive significant fusion feature map.

2. The cold chain transportation commodity quality monitoring system based on image analysis technology according to claim 1 is characterized in that: The fruit surface state multi-dimensional feature extraction module includes: a fruit surface texture feature capture unit, which is used to input the defogging fruit state image into a fruit surface texture feature extractor based on a first hole convolutional neural network model to obtain the fruit surface texture feature map; a fruit surface state semantic feature capture unit, which is used to input the defogging fruit state image into a fruit surface state semantic feature extractor based on a second hole convolutional neural network model to obtain the fruit surface state semantic feature map.

3. The cold chain transportation commodity quality monitoring system based on image analysis technology according to claim 2 is characterized in that: The fruit surface state multi-dimensional feature fine-grained interactive fusion module includes: a fruit surface state multi-dimensional feature fine-grained deconstruction unit, which is used to perform feature fine-grained deconstruction on the fruit surface texture feature map and the fruit surface state semantic feature map to obtain a set of local feature matrices of fruit surface texture channel dimension and a set of local feature matrices of fruit surface state semantic channel dimension; a first unidirectional global attention interactive processing unit, which is used to use each fruit surface texture channel dimension local feature matrix in the set of fruit surface texture channel dimension local feature matrices as a query feature matrix, and use the set of local feature matrices of fruit surface state semantic channel dimension as a set of key feature matrices, and input the query feature matrix and the set of key feature matrices into the unidirectional global attention interactive module based on the first converter structure to obtain a set of local feature matrices of fruit surface texture channel dimension optimized by unidirectional global attention; a second unidirectional global attention interactive processing unit, which is used to use each fruit surface state in the set of local feature matrices of fruit surface state semantic channel dimension as a query feature matrix, and use the set of local feature matrices of fruit surface state semantic channel dimension as a key feature matrix. The semantic channel dimension local feature matrix is ​​used as the query feature matrix, and the set of the fruit surface texture channel dimension local feature matrix is ​​used as the set of key feature matrices. The query feature matrix and the set of the key feature matrix are input into the one-way global attention interaction module based on the second converter structure to obtain the set of one-way global attention optimized fruit surface state semantic channel dimension local feature matrices; the fruit surface state coupling unit is used to couple the set of the one-way global attention optimized fruit surface texture channel dimension local feature matrix and the one-way global attention optimized fruit surface state semantic channel dimension local feature matrix along the channel dimension to obtain a one-way global interaction optimized fruit surface texture feature map and a one-way global interaction optimized fruit surface state semantic feature map; the fruit surface state multidimensional feature weighted sum unit is used to calculate the position-weighted sum between the one-way global interaction optimized fruit surface texture feature map and the one-way global interaction optimized fruit surface state semantic feature map to obtain the fruit surface state shallow-semantic fine-grained interactive significant fusion feature map.

4. The cold chain transportation commodity quality monitoring system based on image analysis technology according to claim 3 is characterized in that: The first unidirectional global attention interaction processing unit is used to: select a predetermined fruit surface texture channel dimension local feature matrix from the set of fruit surface texture channel dimension local feature matrices as a query feature matrix; calculate the product of the predetermined fruit surface texture channel dimension local feature matrix and the transposed matrix of each fruit surface state semantic channel dimension local feature matrix in the set of fruit surface state semantic channel dimension local feature matrix to obtain a set of fruit surface state texture-semantic channel dimension local semantic interaction feature matrices; divide each fruit surface state texture-semantic channel dimension local semantic interaction feature matrix in the set of fruit surface state texture-semantic channel dimension local semantic interaction feature matrix by the scale square root of the predetermined fruit surface texture channel dimension local feature matrix according to the position point, and then use The function performs soft maximum normalization processing on each feature matrix in the obtained set of feature matrices to obtain a set of local weight matrices of the fruit surface texture channel dimension; using the set of local weight matrices of the fruit surface texture channel dimension as weighted weights, calculates the position-weighted sum of the local feature matrices of the fruit surface texture channel dimension in the set of local feature matrices of the fruit surface texture channel dimension to obtain the unidirectional global attention optimized local feature matrix of the fruit surface texture channel dimension.

5. The cold chain transportation commodity quality monitoring system based on image analysis technology according to claim 4 is characterized in that: The second unidirectional global attention interaction processing unit is used to: select a predetermined fruit surface state semantic channel dimension local feature matrix from the set of fruit surface state semantic channel dimension local feature matrices as a query feature matrix; calculate the product of the predetermined fruit surface state semantic channel dimension local feature matrix and the transposed matrix of each fruit surface texture channel dimension local feature matrix in the set of fruit surface texture channel dimension local feature matrix to obtain a set of fruit surface state semantic-texture channel dimension local semantic interaction feature matrices; and convert each fruit surface state semantic-texture channel dimension local semantic interaction feature matrix into a query feature matrix. After dividing the scale square root of the predetermined fruit surface state semantic channel dimension local feature matrix by the position point of the local semantic interaction feature matrix of the channel dimension, the softmax function is used to perform soft maximum normalization processing on each feature matrix in the set of the obtained feature matrices to obtain a set of local weight matrices of the fruit surface state semantic channel dimension; using the set of local weight matrices of the fruit surface state semantic channel dimension as the weighted weight, the position-weighted sum of each local feature matrix of the fruit surface state semantic channel dimension in the set of local feature matrices of the fruit surface state semantic channel dimension is calculated to obtain the unidirectional global attention optimized fruit surface state semantic channel dimension local feature matrix.

6. The cold chain transportation commodity quality monitoring system based on image analysis technology according to claim 5 is characterized in that: The fruit deterioration assessment coefficient generation module is used to: input the shallow-semantic fine-grained interactive significant fusion feature map of the fruit surface state into the decoder-based fruit deterioration assessment module to obtain the decoded value of the fruit deterioration assessment coefficient.

7. The cold chain transportation commodity quality monitoring system based on image analysis technology according to claim 6 is characterized in that: The fruit deterioration assessment coefficient generation module is used to: flatten the shallow-semantic fine-grained interactive significant fusion feature map of the fruit surface state to obtain a shallow-semantic fine-grained interactive significant fusion feature vector of the fruit surface state; multiply the decoding weight matrix of the decoder by the shallow-semantic fine-grained interactive significant fusion feature vector of the fruit surface state to obtain a decoded shallow-semantic fine-grained interactive significant fusion feature vector of the fruit surface state; All eigenvalues ​​of the decoded shallow-semantic fine-grained interactive significant fusion feature vector of the fruit surface state are accumulated and summed to obtain the decoded value of the fruit deterioration assessment coefficient.

8. A method for monitoring the quality of cold chain transported goods based on image analysis technology, characterized in that: include: Acquire the fruit status image of the monitored cold chain fruit collected by the camera; Defogging the fruit state image to obtain a defogged fruit state image; The dehazed fruit state image is subjected to texture feature and semantic feature extraction based on the fruit surface to obtain a fruit surface texture feature map and a fruit surface state semantic feature map; the fruit surface texture feature map and the fruit surface state semantic feature map are input into a bidirectional global attention joint perception module based on fine-grained deconstruction to obtain a shallow-semantic fine-grained interactive significant fusion feature map of the fruit surface state; based on the shallow-semantic fine-grained interactive significant fusion feature map of the fruit surface state, a decoding value of the fruit deterioration assessment coefficient is determined.

9. The method for monitoring the quality of cold chain transported goods based on image analysis technology according to claim 8, characterized in that: The dehazed fruit state image is subjected to texture feature and semantic feature extraction based on the fruit surface to obtain a fruit surface texture feature map and a fruit surface state semantic feature map, including: inputting the dehazed fruit state image into a fruit surface texture feature extractor based on a first atrous convolutional neural network model to obtain the fruit surface texture feature map; inputting the dehazed fruit state image into a fruit surface state semantic feature extractor based on a second atrous convolutional neural network model to obtain the fruit surface state semantic feature map.

10. The method for monitoring the quality of cold chain transported goods based on image analysis technology according to claim 9, characterized in that: Based on the shallow-semantic fine-grained interactive significant fusion feature map of the fruit surface state, determining the decoding value of the fruit deterioration assessment coefficient includes: inputting the shallow-semantic fine-grained interactive significant fusion feature map of the fruit surface state into a fruit deterioration assessment module based on a decoder to obtain the decoding value of the fruit deterioration assessment coefficient.

Images