Intelligent control method and system for AI refrigeration house based on coordination of state and fresh-keeping temperature

Abstract

The invention relates to the technical field of fruit preservation, and particularly discloses an AI refrigeration house intelligent control method and system based on coordination of state and preservation temperature. A data processing technology based on machine learning is adopted to carry out data analysis on surface state images and preservation temperature of cherries in a house; according to the method, the biological state characteristics of the cherries and the time sequence dynamic change mode of the preservation temperature in the library are mined, and then cross-modal deep fusion coding is carried out on the biological state characteristics and the time sequence dynamic change mode, so that the internal correlation between the biological state characteristics and the time sequence dynamic change mode is fully mined, deep understanding of cherry preservation requirements is realized, and whether the current preservation temperature value is appropriate or not is intelligently identified. In this way, the fresh-keeping requirements of different cherry varieties under different environmental conditions can be fully met, and real-time monitoring and intelligent regulation and control of the fresh-keeping state of cherries are achieved, so that the fresh-keeping efficiency is improved, and the shelf life of the cherries is prolonged.

Description

Technical Field

[0001] The present application relates to the technical field of fruit preservation, and more specifically, to an AI cold storage intelligent control method and system based on the coordination of state and preservation temperature. Background Art

[0002] With the continuous expansion of the global food supply chain and the increasing demand for freshness from consumers, fruit preservation technology has become an important research direction in the field of modern agriculture and logistics. Cherries, as a fruit with high nutritional value, are widely loved by consumers for their delicious taste and unique flavor. However, the cherry picking, transportation, storage and sales processes all require strict quality control to ensure their quality and shelf life.

[0003] Some existing studies have adopted the whole cold chain preservation technology to reduce the spoilage rate of cherries and maintain their quality by controlling the temperature during the harvesting, storage, transportation and sales. However, this method of cherry preservation mainly relies on a fixed temperature set by artificial experience. Although it can extend the shelf life of cherries to a certain extent, it is often unable to carry out personalized preservation regulation based on the actual biological state of cherries, resulting in unstable preservation effect and prone to premature decay or quality degradation.

[0004] Therefore, we look forward to an AI cold storage intelligent control method based on the coordination of status and preservation temperature. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides an AI cold storage intelligent control method and system based on the coordination of state and preservation temperature, which uses machine learning-based data processing technology to perform data analysis on the surface state image and preservation temperature of cherries in the warehouse, and mines out the biological state characteristics of cherries and the temporal dynamic change pattern of the preservation temperature in the warehouse, and then through cross-modal deep fusion encoding of the two, to fully mine the inherent relationship between the two, to achieve a deep understanding of the preservation needs of cherries, so as to intelligently identify whether the current preservation temperature value is appropriate. In this way, it can fully adapt to the preservation needs of different cherry varieties and different environmental conditions, realize real-time monitoring and intelligent regulation of the preservation state of cherries, thereby improving the preservation efficiency and extending the shelf life of cherries. The warehouse refers to the inside of the AI ​​preservation cold storage.

[0006] Accordingly, according to one aspect of the present application, an AI cold storage intelligent control method based on state and fresh-keeping temperature coordination is provided, which includes: Obtain the surface state image of cherries in the warehouse collected by the camera and the time queue of the real-time preservation temperature value collected by the temperature sensor; Extracting biological state features of the surface state image of the cherries in the warehouse to obtain a biological state feature map of the cherries in the warehouse; Extracting the time series correlation features of the time queue of the real-time fresh-keeping temperature value to obtain the fresh-keeping temperature time series correlation implicit feature vector; Cross-modal fusion is performed on the biological state feature graph of cherries in the warehouse and the time series associated implicit feature vector of the fresh-keeping temperature to obtain a deep cross-modal fusion coding feature graph of cherry biological feature-fresh-keeping temperature; Based on the cherry biological feature-fresh-keeping temperature deep cross-modal fusion coding feature map, determine whether the fresh-keeping temperature value is within the recommended fresh-keeping temperature range.

[0007] According to another aspect of the present application, an AI cold storage intelligent control system based on state and fresh-keeping temperature coordination is provided, which includes: The cherry preservation information acquisition module is used to obtain the surface state image of the cherries in the warehouse collected by the camera and the time queue of the real-time preservation temperature value collected by the temperature sensor; A biological state feature extraction module is used to extract biological state features of the surface state image of the cherries in the warehouse to obtain a biological state feature map of the cherries in the warehouse; A fresh-keeping temperature time series encoding module is used to extract the time series correlation features of the time queue of the real-time fresh-keeping temperature value to obtain a fresh-keeping temperature time series correlation implicit feature vector; A cross-modal fusion module is used to perform cross-modal fusion on the biological state feature map of cherries in the warehouse and the time series associated implicit feature vector of the fresh-keeping temperature to obtain a deep cross-modal fusion coding feature map of cherry biological feature-fresh-keeping temperature; The fresh-keeping temperature suitability assessment module is used to determine whether the fresh-keeping temperature value is within the recommended fresh-keeping temperature range based on the cherry biological feature-fresh-keeping temperature deep cross-modal fusion encoding feature map.

[0008] Compared with the prior art, the AI ​​cold storage intelligent control method and system based on state and fresh-keeping temperature coordination provided by this application uses machine learning-based data processing technology to analyze the surface state image and fresh-keeping temperature of cherries in the warehouse, and mines out the biological state characteristics of cherries and the temporal dynamic change pattern of the fresh-keeping temperature in the warehouse, and then through cross-modal deep fusion encoding of the two, to fully mine the inherent relationship between the two, to achieve a deep understanding of the cherry preservation needs, so as to intelligently identify whether the current fresh-keeping temperature value is appropriate. In this way, it can fully adapt to the preservation needs of different cherry varieties and different environmental conditions, realize real-time monitoring and intelligent regulation of the cherry preservation state, thereby improving the preservation efficiency and extending the shelf life of cherries. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0010] Figure 1 This is a flow chart of an AI cold storage intelligent control method based on coordination of state and preservation temperature according to an embodiment of the present application.

[0011] Figure 2 This is a data flow diagram of the AI ​​cold storage intelligent control method based on the coordination of state and preservation temperature according to an embodiment of the present application.

[0012] Figure 3 This is a flowchart of step S140 in the AI ​​cold storage intelligent control method based on coordination of state and preservation temperature according to an embodiment of the present application.

[0013] Figure 4 This is a flowchart of step S142 in the AI ​​cold storage intelligent control method based on coordination of state and preservation temperature according to an embodiment of the present application.

[0014] Figure 5 This is a block diagram of an AI cold storage intelligent control system based on coordination of state and preservation temperature according to an embodiment of the present application.

[0015] Figure 6 This is a schematic diagram comparing the TSS content of cherries in AI fresh-keeping storage and ordinary cold storage.

[0016] Figure 7 This is a schematic diagram comparing the color of cherries in the AI ​​fresh-keeping storage and ordinary cold storage.

[0017] Figure 8 Schematic diagram of the brightness comparison of cherries in the AI ​​fresh-keeping storage and ordinary cold storage.

[0018] Fig. 9 This is a schematic diagram comparing the saturation of cherries in the AI ​​fresh-keeping storage and ordinary cold storage.

[0019] Fig.10 This is a schematic diagram comparing the hardness of cherries in the AI ​​fresh-keeping storage and ordinary cold storage.

[0020] Fig.11 This is a schematic diagram comparing the respiration intensity of cherries in the AI ​​fresh-keeping storage and ordinary cold storage.

[0021] Fig.12 This is a schematic diagram comparing the MDA content of cherries in AI fresh-keeping storage and ordinary cold storage.

[0022] Fig.13 This is a schematic diagram comparing the quality of cherries in the AI ​​fresh-keeping storage and ordinary cold storage.

[0023] The AI ​​fresh-keeping warehouse is an AI fresh-keeping warehouse constructed based on the crisp persimmon preservation control method based on multi-parameter coupling analysis of the present application. DETAILED DESCRIPTION

[0024] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0025] Figure 1 This is a flow chart of an AI cold storage intelligent control method based on coordination of state and preservation temperature according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of the AI ​​cold storage intelligent control method based on state and fresh-keeping temperature coordination according to the embodiment of the present application. Figure 1 and Figure 2 As shown, according to the embodiment of the present application, the AI ​​cold storage intelligent control method based on the coordination of state and preservation temperature includes the following steps: S110, obtaining a surface state image of cherries in the storage collected by a camera, and a time queue of real-time preservation temperature values ​​collected by a temperature sensor; S120, extracting the biological state features of the surface state image of cherries in the storage to obtain a biological state feature map of cherries in the storage; S130, extracting the time series correlation features of the time queue of real-time preservation temperature values ​​to obtain a preservation temperature time series correlation implicit feature vector; S140, cross-modally fusing the biological state feature map of cherries in the storage and the preservation temperature time series correlation implicit feature vector to obtain a cherry biological feature-preservation temperature deep cross-modal fusion coding feature map; S150, determining whether the preservation temperature value is within the recommended preservation temperature range based on the cherry biological feature-preservation temperature deep cross-modal fusion coding feature map.

[0026] In the above-mentioned AI cold storage intelligent control method based on the coordination of state and preservation temperature, the step S110 obtains the surface state image of the cherries in the storage collected by the camera, and the time queue of the real-time preservation temperature value collected by the temperature sensor. It should be understood that the appearance of cherries can directly reflect their freshness and quality. By analyzing the color, shape, texture and other characteristics of the cherry surface, the biological state information such as the maturity of the cherry, whether there are pests, damage, rot or spots, etc. can be reflected, so as to evaluate its preservation effect. Temperature is one of the key factors affecting the preservation effect of cherries. Therefore, by jointly analyzing the biological state and ambient temperature of cherries, we can fully understand the impact of ambient temperature on the biological state of cherries, as well as the demand for ambient temperature of the biological state of cherries under the current situation, which helps to more accurately analyze whether the current ambient temperature is suitable, thereby providing an important basis for the regulation of cherry preservation conditions.

[0027] The collection of cherry surface state images is completed by multiple cameras installed inside the cold storage. In order to accurately capture the surface state of cherries, it is crucial to select cameras with high resolution and good color reproduction capabilities. The selected camera can clearly record the subtle features of cherries such as color, shape, size, etc., which is helpful for subsequent image analysis. Considering the special environment in the cold storage, such as low temperature and high humidity, the selected camera should have high weather resistance to ensure that it can still work normally under harsh conditions. Cameras are usually installed in multiple locations inside the cold storage, including near the door, walls, ground and central areas, etc., to ensure that all areas where cherries are stored are covered, avoid blind spots, and achieve all-round monitoring. The location layout of the cameras needs to be carefully designed to ensure the comprehensiveness of image acquisition and avoid mutual interference.

[0028] After the camera is installed, the frequency of image acquisition needs to be determined. The deterioration rate of cherries will affect the frequency of image acquisition. Generally speaking, if the cherries mature quickly, it may be necessary to take images every hour or even more frequently; otherwise, the acquisition frequency can be appropriately reduced. A reasonable acquisition frequency can not only detect changes on the surface of cherries in a timely manner, but also effectively balance the amount of data and storage costs. The collected original images may be affected by noise or background interference, so preprocessing is required, including denoising, cropping, adjusting contrast, etc., to improve the image quality and the accuracy of subsequent analysis. The preprocessed images are stored in chronological order and transmitted to the back-end server via a wired or wireless network for further processing. Due to the large amount of image data, compression technology can be used to save storage space and bandwidth resources to ensure the efficiency of data transmission. In addition, in order to ensure the security and integrity of image data, a series of network security measures need to be taken, such as encrypted transmission and access control, to prevent data leakage or illegal tampering.

[0029] At the same time, the collection of real-time preservation temperature values ​​is also a crucial link. Temperature is one of the key factors affecting the preservation effect of cherries. Therefore, by selecting temperature sensors with high accuracy and fast response and distributing them evenly throughout the cold storage, changes in ambient temperature can be fully monitored. The layout of the sensors should cover key locations such as near the door, wall, ground and central area, and the average temperature collected by each sensor should be calculated as the real-time preservation temperature value in the warehouse to ensure the comprehensiveness and representativeness of the temperature data. In addition, the sensor should be away from factors that may cause interference, such as human heat sources, lights, etc., to avoid data distortion. The acquisition frequency of temperature data is usually higher than the image acquisition frequency because the temperature changes are relatively fast and continuous. In general, recording the temperature value once a minute can meet the needs of most application scenarios. However, for some specific scenarios, such as environments with large temperature fluctuations, a higher acquisition frequency may be required.

[0030] To ensure the accuracy of temperature data, regular calibration of sensors is an indispensable step. During the calibration process, a standard temperature source should be used to compare with the sensor to correct possible errors. In addition, strict quality audits should be implemented for the collected temperature data to eliminate outliers, that is, data points that significantly deviate from the normal range. Such outliers may be caused by sensor failure or interference from external factors. If not handled, they will have an adverse effect on the accuracy of subsequent analysis. The calibrated temperature data needs to be recorded in chronological order to form time series data and transmitted to the back-end server through the network so that it can be comprehensively analyzed together with the image data. To ensure the security of data transmission, a series of network security protection measures should be adopted, including encrypted transmission, access control, etc., to effectively prevent the risk of data leakage or illegal tampering.

[0031] It is worth noting that during the data collection process, the overall operation and maintenance management of the system must also be fully considered. The first task is to regularly check the working status of the camera and temperature sensor. Once a faulty device is found, it should be replaced immediately to ensure the long-term stable operation of the system. Secondly, a complete data backup mechanism needs to be built to deal with data loss caused by unexpected situations. Finally, a data management system needs to be established to realize the automatic collection, processing and storage of data, thereby improving work efficiency. For example, special data management software can be developed to monitor equipment status, automatically schedule data collection tasks, and display data collection progress and results in real time, etc., to provide managers with an intuitive operation interface and simplify daily maintenance work.

[0032] In addition, the efficiency and accuracy of data collection also depend on the integration and compatibility of the system. In actual deployment, cameras and temperature sensors need to be seamlessly connected to the data processing platform to ensure smooth data transmission and processing. To this end, standardized data interface protocols such as Modbus and OPC UA can be used to achieve interconnection between different devices. At the same time, the data processing platform should have strong data processing capabilities, be able to quickly parse and store large amounts of data, support real-time query and historical backtracking, and provide a solid foundation for subsequent intelligent analysis. For example, distributed database technology can be used to store data in multiple nodes to improve data access speed and reliability; cloud computing platforms can also be used to use elastic computing resources to dynamically adjust computing power according to the increase or decrease in data volume to ensure efficient operation of the system.

[0033] In the above-mentioned AI cold storage intelligent control method based on the coordination of state and preservation temperature, the step S120 extracts the biological state characteristics of the surface state image of the cherries in the warehouse to obtain the biological state characteristic map of the cherries in the warehouse. In a specific example of the present application, the step S120 includes: inputting the surface state image of the cherries in the warehouse into a cherry biological characteristic extractor based on a deep separable convolutional neural network model to obtain the biological state characteristic map of the cherries in the warehouse. That is, in order to fully capture the biological state characteristics of cherries, the present application uses a deep separable convolutional neural network model to process the surface state image of cherries in the warehouse, so as to utilize the powerful learning ability of the deep separable convolutional neural network model to extract key biological state characteristics such as cherry color distribution, contour shape, and texture structure in the cherry surface state image, so as to obtain the biological state characteristic map of cherries in the warehouse.

[0034] The deep separable convolutional neural network can effectively reduce the number of model parameters and maintain high feature extraction by decomposing the standard convolution into depthwise convolution and pointwise convolution. Among them, the deep convolution is responsible for applying filters independently on each channel of the cherry surface state image in the library, while the pointwise convolution integrates the feature information of all channels through a 1x1 convolution kernel, which not only reduces the number of parameters, but also reduces the computational complexity, making the model more lightweight while maintaining a high accuracy rate.

[0035] The following is an example of a network structure based on DS-CNN, which is used for feature extraction of cherry surface state images. Assume that the size of the input cherry surface state image is 224×224×3224×224×3 (width×height×number of channels). First, an initial convolution layer is used to convert the input image into a feature map suitable for subsequent depthwise separable convolution operations. The parameters of the initial convolution layer are set as follows: the convolution kernel size is 3×33×3, the stride is 1, the padding is same (keep the input and output sizes consistent), the number of output channels is 32, and the activation function is ReLU. After the initial convolution layer, the size of the output feature map is 224×224×32224×224×32. Next, enter the depthwise separable convolution layer. The depthwise separable convolution layer consists of depthwise convolution and pointwise convolution. First, perform depthwise convolution with a convolution kernel size of 3×33×3, a stride of 1, padding of same, an input channel number of 32, and an output channel number of 32 (each input channel is processed independently). The feature map after depth convolution is processed by the ReLU activation function, and the output size is 224×224×32224×224×32. Then point-by-point convolution is performed, with a convolution kernel size of 1×11×1, a stride of 1, padding of the same, an input channel number of 32, and an output channel number of 64. The feature map after point-by-point convolution is also processed by the ReLU activation function, and the output size is 224×224×64224×224×64.

[0036] In order to reduce the spatial dimension of the feature map and reduce the amount of calculation, the maximum pooling layer is used next. The parameters of the maximum pooling layer are set as follows: the pooling kernel size is 2×22×2, the stride is 2, and the padding is valid. After the maximum pooling layer is processed, the size of the output feature map is 112×112×64112×112×64.

[0037] In order to further extract more advanced features, the combination of depthwise separable convolutional layers and maximum pooling layers can be repeated multiple times. Taking the second depthwise separable convolutional layer as an example, first perform depthwise convolution with a kernel size of 3×33×3, a stride of 1, padding of same, 64 input channels, and 64 output channels. The feature map after depthwise convolution is processed by the ReLU activation function, and the output size is 112×112×64112×112×64. Then perform pointwise convolution with a kernel size of 1×11×1, a stride of 1, padding of same, 64 input channels, and 128 output channels. The feature map after pointwise convolution is also processed by the ReLU activation function, and the output size is 112×112×128112×112×128. Subsequently, a maximum pooling layer is used with a kernel size of 2×22×2, a stride of 2, and padding of valid. After processing by the maximum pooling layer, the size of the output feature map is 56×56×12856×56×128.

[0038] It should be understood that by decomposing the traditional convolution operation into depth convolution and point-by-point convolution, DS-CNN significantly reduces the number of model parameters and computational complexity while maintaining a high feature extraction capability. This efficient network structure is particularly suitable for resource-constrained scenarios, such as mobile devices or embedded systems, and is also suitable for processing large-scale data sets, such as image analysis tasks in cherry preservation. Through the combination of multiple layers of depth-separable convolution and maximum pooling layers, the model can gradually extract high-level features of the cherry surface, providing strong support for intelligent identification of the cherry preservation status.

[0039] In the above-mentioned AI cold storage intelligent control method based on state and fresh-keeping temperature collaboration, the step S130 extracts the time series correlation features of the time queue of the real-time fresh-keeping temperature value to obtain the fresh-keeping temperature time series correlation implicit feature vector. In a specific example of the present application, the step S130 includes: inputting the time queue of the real-time fresh-keeping temperature value into the fresh-keeping temperature time series feature extractor based on the forward LSTM model to obtain the fresh-keeping temperature time series correlation implicit feature vector. It should be understood that considering that the fresh-keeping temperature is data that changes dynamically over time, therefore, in order to better understand the time change trend of the fresh-keeping temperature, the present application adopts a forward long short-term memory network (LSTM) model to perform time series encoding on the time queue of the real-time fresh-keeping temperature value. By introducing a gating mechanism, the LSTM model can effectively control the flow of information, thereby capturing the long-term dependency of the fresh-keeping temperature value over time, extracting the correlation change characteristics of the temperature in the time dimension, and obtaining the fresh-keeping temperature time series correlation implicit feature vector, thereby providing important information for subsequent fresh-keeping temperature analysis.

[0040] The forward long short-term memory network (LSTM) is a special recurrent neural network (RNN) that is particularly good at processing sequence data and can capture long-term dependencies in time series. In the AI ​​cold storage intelligent control method based on the coordination of state and fresh-keeping temperature, the LSTM model is used to extract the time series correlation features of the time queue of real-time fresh-keeping temperature values, thereby generating a fresh-keeping temperature time series correlation implicit feature vector. The following is an example of a network structure based on a forward LSTM model, which is used to process cherry fresh-keeping temperature data.

[0041] Assume that the time queue of the input real-time fresh-keeping temperature value is ,in represents the temperature value at the i-th time point, and n is the length of the time series. First, input data It is fed into an embedding layer, which converts the one-dimensional temperature value into a fixed-length vector. The role of the embedding layer is to map the discrete temperature value into a continuous vector space so that the LSTM model can better process it. Next, the output of the embedding layer is fed into the LSTM layer. The LSTM layer consists of multiple LSTM units, each of which contains three gating mechanisms: input gate, forget gate, and output gate. These gating mechanisms help the model remember long-term dependencies and forget irrelevant information by controlling the flow of information.

[0042] Specifically, for each time point, the calculation process of the LSTM unit is as follows: Input Gate: The input gate determines how much of the input information at the current time point will be written into the cell state. The input gate is calculated using a sigmoid activation function, and the input includes the embedding vector at the current time point and the hidden state at the previous time point.

[0043] Forget Gate: The forget gate determines how much information in the cell state will be retained. The forget gate calculation also uses a sigmoid activation function, and the input includes the embedding vector of the current time point and the hidden state of the previous time point.

[0044] Cell state update: The cell state update combines the results of the forget gate and the input gate, as well as a new candidate cell state value. The new candidate cell state value is calculated through a tanh activation function, and the input includes the embedding vector at the current time point and the hidden state at the previous time point.

[0045] Output Gate: The output gate determines how much of the cell state at the current time point will be output. The output gate is calculated using a sigmoid activation function, and the input includes the embedding vector of the current time point and the hidden state of the previous time point.

[0046] Hidden state update: The hidden state update combines the result of the output gate and the cell state at the current time point, calculated through a tanh activation function.

[0047] Through the above calculations, the LSTM unit can effectively capture the long-term dependencies in the time series. In summary, the above network structure example shows how to use the forward long short-term memory network (LSTM) model to extract the time series correlation features of the time queue of real-time preservation temperature values. The one-dimensional temperature value is converted into a fixed-length vector through the embedding layer, the LSTM layer captures the long-term dependencies in the time series, and the fully connected layer extracts high-level features, and finally generates the preservation temperature time series correlation implicit feature vector. This structure is particularly suitable for processing time series data, and can effectively capture the trend of temperature changes, providing important information for the intelligent identification of the preservation status of cherries.

[0048] In the above-mentioned AI cold storage intelligent control method based on the coordination of state and preservation temperature, the step S140 cross-modally fuses the biological state feature map of cherries in the warehouse and the implicit feature vector associated with the preservation temperature time series to obtain a deep cross-modal fusion coding feature map of cherry biological features and preservation temperature. It should be understood that there is usually a complex implicit association between the biological state of cherries and the preservation temperature. For example, changes in temperature will affect the biological characteristics of cherries such as color and texture. At the same time, the biological state of cherries can in turn reflect the adaptability of cherries to current temperature conditions, thereby providing an important reference for the setting of the preservation temperature. Therefore, in order to achieve a comprehensive understanding of the preservation state of cherries, the present application further cross-modally fuses the biological state feature map of cherries in the warehouse and the implicit feature vector associated with the preservation temperature time series, and provides more powerful feature support for subsequent preservation temperature analysis by learning the intrinsic connection between the biological state of cherries and the preservation temperature.

[0049] Figure 3 FIG. 1 is a flow chart of step S140 in the AI ​​cold storage intelligent control method based on state and fresh-keeping temperature coordination according to an embodiment of the present application. Figure 3 As shown, the step S140 includes: S141, extracting the intrinsic common features between the fresh-keeping temperature time series associated implicit feature vector and the biological state characteristic map of cherries in the warehouse to obtain the intrinsic common feature vector of cherry biological feature-fresh-keeping temperature; S142, based on the intrinsic common feature vector of cherry biological feature-fresh-keeping temperature, performing dual feature modulation on the biological state characteristic map of cherries in the warehouse to obtain the deep cross-modal fusion coding feature map of cherry biological feature-fresh-keeping temperature.

[0050] Specifically, the step S141 includes: performing feature distillation on the biological state characteristic graph of cherries in the warehouse to obtain a distilled feature vector of the biological state of cherries in the warehouse; inputting the implicit feature vector of the fresh-keeping temperature time series association and the distilled feature vector of the biological state of cherries in the warehouse into a multimodal implicit association feature capture network to obtain the biological feature of cherries-fresh-keeping temperature intrinsic commonality feature vector. In a specific example of the present application, performing feature distillation on the biological state characteristic graph of cherries in the warehouse to obtain a distilled feature vector of the biological state of cherries in the warehouse includes: performing point convolution encoding on the biological state characteristic graph of cherries in the warehouse to obtain a channel-modulated biological state characteristic graph of cherries in the warehouse; performing global mean pooling processing on the biological state characteristic graph of cherries in the channel-modulated warehouse along the channel dimension and then inputting the Sigmoid function for activation processing to obtain the distilled feature vector of the biological state of cherries in the warehouse. In a specific example of the present application, the fresh-keeping temperature time series associated implicit feature vector and the biological state distilled feature vector of cherries in the warehouse are input into a multimodal implicit association feature capture network to obtain the cherry biological feature-fresh-keeping temperature intrinsic common feature vector, including: cascading and fusing the fresh-keeping temperature time series associated implicit feature vector and the biological state distilled feature vector of cherries in the warehouse and inputting them into a neural network layer based on a tanh function to obtain the cherry biological feature-fresh-keeping temperature intrinsic common feature vector.

[0051] The above step S141 can be expressed as follows:

[0052]

[0053] in, represents the implicit feature vector associated with the fresh-keeping temperature time series, A characteristic diagram of the biological state of cherries in the storage, represents point convolution, represents the average pooling operation, represents the Sigmoid activation function, represents the biological state distillation feature vector of cherries in the library, Indicates cascade, and Represent the weight parameter matrix and bias term of the neural network layer respectively, represents the hyperbolic tangent function, Represents the intrinsic common feature vector of the cherry biological feature - fresh-keeping temperature.

[0054] That is, first, the biological state feature map of cherries in the warehouse is feature distilled to remove redundant information, obtain a more compact and representative biological state distillation feature vector of cherries in the warehouse, and unify the feature dimensions so that the features in the subsequent stages can interact effectively. Then, the temporal association implicit feature vector of the fresh-keeping temperature and the biological state distillation feature vector of cherries in the warehouse are input into the multimodal implicit association feature capture network to identify and capture the potential common patterns or structures between the two, thereby constructing the intrinsic commonality feature vector of cherry biological features-fresh-keeping temperature.

[0055] Specifically, Figure 4 FIG. 1 is a flow chart of step S142 in the AI ​​cold storage intelligent control method based on state and fresh-keeping temperature coordination according to an embodiment of the present application. Figure 4 As shown, the step S142 includes: S1421, inputting the intrinsic common feature vector of the cherry biological characteristic-fresh-keeping temperature into the feature primary screening module based on the gating unit to obtain the primary storage biological state feature primary screening weight vector of the cherry; S1422, based on the primary storage biological state feature primary screening weight vector, performing preliminary feature screening on the biological state feature map of cherries in the storehouse to obtain the biological state feature map of cherries in the primary screening storehouse; S1423, inputting the intrinsic common feature vector of the cherry biological characteristic-fresh-keeping temperature into the feature secondary screening module based on the multi-level gating unit to obtain the biological state feature screening weight vector of cherries in the secondary storage; S1424, based on the cherry biological state feature screening weight vector of the secondary storage, performing secondary feature screening on the biological state feature map of cherries in the primary screening storehouse to obtain the cherry biological characteristic-fresh-keeping temperature deep cross-modal fusion coding feature map.

[0056] More specifically, the step S1421 includes: multiplying the cherry biometric-fresh-keeping temperature intrinsic common feature vector by the first weight parameter matrix, and then performing dot addition with the first bias item to obtain a first-level cherry biometric-fresh-keeping temperature intrinsic common modulation vector; inputting the first-level cherry biometric-fresh-keeping temperature intrinsic common modulation vector into a Sigmoid function for nonlinear activation processing to obtain a first-level cherry biometric-fresh-keeping temperature intrinsic common modulation activation vector; inputting the first-level cherry biometric-fresh-keeping temperature intrinsic common modulation activation vector into a mask unit to obtain a preliminary screening weight vector of cherry biological state characteristics in the first-level library. Among them, the first-level cherry biometric characteristic-fresh-keeping temperature intrinsic common modulation activation vector is input into the mask unit to obtain the preliminary screening weight vector of the biological status characteristics of cherries in the first-level warehouse, including: setting a first mask threshold and a second mask threshold, wherein the second mask threshold is twice the first mask threshold; if the eigenvalue in the first-level cherry biometric characteristic-fresh-keeping temperature intrinsic common modulation activation vector is greater than the second mask threshold, it remains unchanged; if the eigenvalue in the first-level cherry biometric characteristic-fresh-keeping temperature intrinsic common modulation activation vector is greater than the first mask threshold, and is less than or equal to the second mask threshold, it is doubled and reduced; if the eigenvalue in the first-level cherry biometric characteristic-fresh-keeping temperature intrinsic common modulation activation vector is less than or equal to the first mask threshold, it is set to 0, thereby obtaining the preliminary screening weight vector of the biological status characteristics of cherries in the first-level warehouse.

[0057] More specifically, the step S1422 includes: taking each eigenvalue of the preliminary screening weight vector of the biological state characteristics of cherries in the first-level warehouse as a weight, weighting each feature matrix of the biological state characteristic diagram of cherries in the warehouse along the channel dimension by position, and then inputting the tanh function for activation processing to obtain the biological state characteristic diagram of cherries in the preliminary screening warehouse.

[0058] The above steps S1421 and S1422 can be expressed as follows:

[0059]

[0060]

[0061]

[0062] in, and Represent the first weight parameter matrix and the first bias term respectively, represents the intrinsic common modulation activation vector of the first-level cherry biological feature - fresh-keeping temperature, represents the mask function, represents the first mask threshold, represents the matrix multiplication operation, represents the weight vector for screening the biological status characteristics of cherries in the primary storage, represents dot product, This is a characteristic diagram of the biological status of cherries in the initial screening warehouse. represents the hyperbolic tangent function, represents the Sigmoid activation function, represents the intrinsic commonality feature vector of the cherry biological feature-fresh-keeping temperature, represents the eigenvalue in the intrinsic common modulation activation vector of the first-level cherry biological feature - fresh-keeping temperature, A characteristic diagram of the biological status of cherries in the warehouse is shown.

[0063] That is, based on the gated mask mechanism, the intrinsic common feature vector of the cherry biological characteristics-fresh-keeping temperature is gated and masked to achieve feature importance evaluation and selection, identify the significant correlation characteristics between the cherry biological characteristics and the fresh-keeping temperature, and generate a corresponding weight vector to perform a preliminary screening of the cherry biological state characteristic diagram to suppress irrelevant or minor parts therein, thereby obtaining a cherry biological state characteristic diagram in the preliminary screening library that focuses on the significant correlation with the fresh-keeping temperature.

[0064] More specifically, the step S1423 includes: multiplying the first-level cherry biometric characteristic-fresh-keeping temperature intrinsic common modulation vector by the second weight parameter matrix, and then dot-adding it with the second bias item to obtain the second-level cherry biometric characteristic-fresh-keeping temperature intrinsic common modulation vector; inputting the second-level cherry biometric characteristic-fresh-keeping temperature intrinsic common modulation vector into the Sigmoid function for nonlinear activation processing to obtain the second-level cherry biometric characteristic-fresh-keeping temperature intrinsic common modulation activation vector; inputting the second-level cherry biometric characteristic-fresh-keeping temperature intrinsic common modulation activation vector into the mask unit to obtain the cherry biological state feature screening weight vector in the second-level library.

[0065] More specifically, the step S1424 includes: taking each eigenvalue of the cherry biological state feature screening weight vector in the secondary warehouse as a weight, weighting each feature matrix of the cherry biological state feature map in the primary screening warehouse along the channel dimension by position, and then inputting the Sigmoid function for activation processing to obtain the cherry biological feature-shelf temperature deep cross-modal fusion coding feature map.

[0066] The above steps S1423 and S1424 can be expressed as follows:

[0067]

[0068]

[0069] in, and Represent the second weight parameter matrix and the second bias term respectively, and Represent the first weight parameter matrix and the first bias term respectively, represents the intrinsic commonality feature vector of the cherry biological feature-fresh-keeping temperature, represents the Sigmoid activation function, represents the above mask function, represents the intrinsic common modulation activation vector of the secondary cherry biological feature - fresh-keeping temperature, represents the weight vector for screening the biological status characteristics of cherries in the secondary library, represents the cherry biometric-keeping temperature deep cross-modal fusion encoding feature map, This is a characteristic diagram of the biological status of cherries in the initial screening warehouse. represents the matrix multiplication operation, Represents dot product.

[0070] Here, in order to further improve the performance of the model, the common feature vector of the cherry biological characteristics-fresh-keeping temperature is screened again by multi-level gated screening, and the process of information transmission is managed more finely through the multi-level gated architecture, and the previously formed cherry biological state feature map in the initial screening library is refined for the second time. After two targeted feature selections and reconstructions, the cherry biological characteristics-fresh-keeping temperature deep cross-modal fusion coding feature map rich in significant correlation information between the cherry biological state and the fresh-keeping temperature is finally obtained, thereby achieving a comprehensive and in-depth understanding of the cherry fresh-keeping state, and providing a more accurate and reliable basis for the current fresh-keeping temperature regulation analysis.

[0071] In the above-mentioned AI cold storage intelligent control method based on state and fresh-keeping temperature collaboration, the step S150 determines whether the fresh-keeping temperature value is within the recommended fresh-keeping temperature range based on the cherry biometric-fresh-keeping temperature deep cross-modal fusion coding feature map.

[0072] In a specific example of the present application, the step S150 includes: inputting the cherry biometric-fresh-keeping temperature deep cross-modal fusion coding feature map into a classifier-based fresh-keeping optimization module to obtain an optimization result, and the optimization result is used to indicate whether the fresh-keeping temperature value is within the recommended fresh-keeping temperature range. That is, the classifier is used to perform feature analysis on the cherry biometric-fresh-keeping temperature deep cross-modal fusion coding feature map to determine whether the current fresh-keeping temperature is suitable. The classifier can learn the requirements of cherries for fresh-keeping temperature under different biological states based on the training data set, and establish a corresponding classification decision boundary, so that it can accurately determine whether the current fresh-keeping temperature is within an appropriate range based on the feature information contained in the cherry biometric-fresh-keeping temperature deep cross-modal fusion coding feature map, thereby achieving a suitability assessment of the current fresh-keeping temperature.

[0073] In a specific embodiment of the present application, when the classifier determines that the fresh-keeping temperature deviates from the recommended range, a corresponding warning signal can be issued to prompt the operator to adjust the temperature setting of the refrigeration equipment in time to ensure that the cherries are kept fresh in the best way. In this way, intelligent management of cherry preservation can be achieved to ensure the freshness and quality of cherries during transportation and storage, thereby improving economic benefits and consumer satisfaction.

[0074] Preferably, when the biological state feature map of cherries in the warehouse and the temporal association implicit feature vector of the fresh-keeping temperature respectively represent the image semantic features of the surface state image of cherries in the warehouse and the forward short-range-long-range temporal implicit association features of the real-time fresh-keeping temperature value, when cross-modal encoding is performed on them, the difference in the distribution of cross-modal semantic implicit associations will also make the cross-modal spatial composite coding semantic representation of the cherry biological feature-fresh-keeping temperature deep cross-modal fusion coding feature map have a complex spatial structure based on different semantic association distribution structures, thereby expecting to improve the convergence and generalization effects of classification regression under complex spatial structures.

[0075] Therefore, in a preferred example of the present application, when the cherry biometric-preservation temperature deep cross-modal fusion coding feature map is input into the classifier-based preservation optimization module to obtain the optimization result, the cherry biometric-preservation temperature deep cross-modal fusion coding feature map is first optimized, specifically including: Determine the feature mean and feature variance of the cherry biometric-fresh-keeping temperature deep cross-modal fusion coding feature map, and divide the feature mean by the feature variance to obtain the cherry biometric-fresh-keeping temperature deep cross-modal fusion coding probability statistics value:

[0076] in, represents the probability statistics of cherry biometrics-keeping temperature deep cross-modal fusion encoding, represents the feature mean, represents the characteristic variance; The cherry biometric-preservation temperature deep cross-modal fusion coding feature map is point-multiplied with the inverse of the maximum eigenvalue in the cherry biometric-preservation temperature deep cross-modal fusion coding feature map to obtain the cherry biometric-preservation temperature deep cross-modal fusion coding probability constraint map:

[0077] in, represents the cherry biometric-keeping temperature deep cross-modal fusion encoding probability constraint graph, represents the cherry biometric-keeping temperature deep cross-modal fusion encoding feature map, represents the inverse of the maximum eigenvalue in the cherry biometric-keeping temperature deep cross-modal fusion encoding feature map, represents dot product; After dot-adding the cherry biometric-preservation temperature deep cross-modal fusion coding probability constraint graph and the cherry biometric-preservation temperature deep cross-modal fusion coding probability statistics value, the logarithmic value with base 2 is calculated to obtain the cherry biometric-preservation temperature deep cross-modal fusion coding information interaction graph:

[0078] in, This is a diagram showing the interaction of cherry biometrics and fresh-keeping temperature deep cross-modal fusion coding information. represents the logarithm with base 2, represents the cherry biometric-keeping temperature deep cross-modal fusion encoding probability constraint graph, Indicates point addition; After point-subtracting the cherry biometric-preservation temperature deep cross-modal fusion coding probability constraint graph with the cherry biometric-preservation temperature deep cross-modal fusion coding probability statistics value, the inverse of each eigenvalue is calculated to obtain the cherry biometric-preservation temperature deep cross-modal fusion coding sequence constraint graph:

[0079] in, Represents the cherry biological feature-keeping temperature deep cross-modal fusion coding sequence constraint graph, represents the probability statistics of cherry biometrics-keeping temperature deep cross-modal fusion encoding, Indicates point reduction, represents the reciprocal of the eigenvalue; The cherry biometric-preservation temperature deep cross-modal fusion coding information interaction graph and the cherry biometric-preservation temperature deep cross-modal fusion coding sequence constraint graph are interpolated to obtain an optimized cherry biometric-preservation temperature deep cross-modal fusion coding feature graph.

[0080] In this preferred example, the probabilistic statistical characteristics of the cherry biometric-shelf temperature deep cross-modal fusion coding feature map are used to simulate the mesoscale interaction structure under probability constraints between the eigenvalue scale and the feature map scale of the cherry biometric-shelf temperature deep cross-modal fusion coding feature map, so as to ensure the spatial transformation invariance of the cherry biometric-shelf temperature deep cross-modal fusion coding feature map under feature space interaction, thereby improving the convergence and generalization effect of the feature set of the cherry biometric-shelf temperature deep cross-modal fusion coding feature map for classification regression under complex spatial structure representation, and improving the image quality of the optimized result.

[0081] In summary, the AI ​​cold storage intelligent control method based on the coordination of state and preservation temperature according to the embodiment of the present application is explained, which uses machine learning-based data processing technology to analyze the surface state image and preservation temperature of cherries in the warehouse, and mines the biological state characteristics of cherries and the temporal dynamic change pattern of the preservation temperature in the warehouse, and then through cross-modal deep fusion encoding of the two, to fully mine the inherent relationship between the two, to achieve a deep understanding of the preservation needs of cherries, so as to intelligently identify whether the current preservation temperature value is appropriate. In this way, it can fully adapt to the preservation needs of different cherry varieties and different environmental conditions, realize real-time monitoring and intelligent regulation of the preservation state of cherries, thereby improving the preservation efficiency and extending the shelf life of cherries.

[0082] Figure 5 FIG. 1 is a block diagram of an AI cold storage intelligent control system based on state and fresh-keeping temperature coordination according to an embodiment of the present application. Figure 5As shown, according to the embodiment of the present application, the AI ​​cold storage intelligent control system 100 based on the coordination of state and preservation temperature includes: a cherry preservation information acquisition module 110, which is used to obtain the surface state image of cherries in the warehouse collected by the camera, and the time queue of the real-time preservation temperature value collected by the temperature sensor; a biological state feature extraction module 120, which is used to extract the biological state features of the surface state image of cherries in the warehouse to obtain the biological state feature map of cherries in the warehouse; a preservation temperature time series encoding module 130, which is used to extract the time series correlation features of the time queue of the real-time preservation temperature value to obtain the preservation temperature time series correlation implicit feature vector; a cross-modal fusion module 140, which is used to cross-modally fuse the biological state feature map of cherries in the warehouse and the preservation temperature time series correlation implicit feature vector to obtain a cherry biological feature-preservation temperature deep cross-modal fusion coding feature map; a preservation temperature suitability evaluation module 150, which is used to determine whether the preservation temperature value is within the recommended preservation temperature range based on the cherry biological feature-preservation temperature deep cross-modal fusion coding feature map.

[0083] Here, those skilled in the art can understand that the specific operations of each module in the above-mentioned AI cold storage intelligent control system based on state and fresh-keeping temperature coordination have been referred to above. Figures 1 to 4 The description of the AI ​​cold storage intelligent control method based on the coordination of state and preservation temperature has been introduced in detail, and therefore, its repeated description will be omitted.

[0084] Example 1 2,000 cherry fruits of uniform size, uniform color, intact fruit stalks, no pests and mechanical damage, and maturity between 80% and 90% were selected and randomly divided into two groups, each containing 1,000 fruits. Subsequently, this application uses the above-mentioned AI cold storage intelligent control method based on the coordination of state and preservation temperature to construct a cherry AI preservation warehouse, and stores two groups of the same number of cherry fruits in the AI ​​preservation cold storage (as the experimental group) and the ordinary cold storage based on fixed temperature control (as the control group) for comparative experiments, observe the cherries at a fixed time and place, and record the fruit conditions in detail.

[0085] 1. Soluble solids (TSS) content Take 50g of fruit sample, grind it in a mortar, and filter the juice for determination. Use a digital saccharimeter to determine the soluble solid content. Gently wipe the mirror of the saccharimeter with lens paper, drop a few drops of distilled water on the mirror, and press the zero adjustment button to adjust it to zero. Use a dropper to draw the sample liquid, add it to the detection mirror, press the test button and read the value, which is the soluble solid content in the sample liquid, expressed as mass fraction. Take three samples from each group, repeat the measurement for each sample 3 times, and calculate the average value. Figure 6As shown, in the first 14 days of storage, the soluble solids content of all treated cherry fruits showed an upward trend, and the difference was not obvious. This may be because the big cherries are not yet fully mature and continue to mature during the first 14 days of storage. From the 14th day, the soluble solids content of cherry fruits began to decline, and the decline rate of ordinary refrigeration was significantly faster than that of AI fresh-keeping storage. After 35 days of storage, the soluble solids content of the fruit in the experimental group was higher than that in the control group, and the difference was significant (P<0.05). Under the storage conditions of ordinary cold storage, large areas of cherries rotted after 35 days, so the storage test was no longer continued. The results of the determination of TSS content are shown in Table 1. According to the experimental data, it can be seen that the AI ​​preservation technology described in this application can effectively maintain the soluble solids content of cherry fruits and maintain the quality of cherries.

[0086] Table 1 TSS content of cherries in AI fresh-keeping storage and ordinary cold storage

[0087] 2. Peel color After the cherries are removed from the storage environment, they should be placed at room temperature for 30 minutes to reach equilibrium. Subsequently, a measurement point is selected at the equatorial part of each fruit, and the base color is measured using a colorimeter, and the color system is represented by L*, a*, and b* values. Ten fruits are used for each measurement and the average value is calculated. The L* value represents the brightness of the material. The higher the value, the brighter the color and the milder the browning. The change in color of the sample during storage is measured by the size of the H0 value. The lower the H0 value, the more severe the browning phenomenon. According to the method of Holcroft et al., a comparative analysis of color saturation (C*) is performed. The higher the C* value, the brighter the color and the higher the commodity value.

[0088] The calculation formula is as follows:

[0089]

[0090] like Figure 7 , 8As shown in Figures 9, with the increase of storage time, the brightness, color saturation and chroma of the fruits in the experimental group and the control group showed a downward trend. Specifically, the brightness and color saturation of the fruit decreased more gently, while the chroma decreased more rapidly within the first 18 days of storage, and then the downward trend slowed down from 18 days to 36 days. It is worth noting that the decline in various indicators of the fruit in the experimental group was smaller than that in the control group. Specific data show that on the 35th day, the brightness of the experimental group decreased by 14.41% and the chroma decreased by 40.34%, which is significantly better than the 29.97% and 52.70% of the control group. Compared with ordinary cold storage, the AI ​​fresh-keeping storage described in this application performs better in the storage effect of cherry fruit, and can more effectively maintain the original color of the fruit, thereby maintaining its commercial value.

[0091] 3. Fruit hardness Select one point on each side of the fruit, and use a fruit hardness tester with a 5mm diameter probe to perform a uniform penetration test. The hardness of the fruit is expressed by the average maximum resistance of the two sides at the moment of penetration. The unit of hardness is Newton per square centimeter (N / cm²). Each sample needs to be tested three times, with 10 fruits tested each time, and the average value is calculated. Fig.10 As shown in the figure, during the storage period, the cherry fruit firmness of all experimental groups was higher than that of the control group. From 0 days to 9 days, the cherry fruit firmness of both the control group and the experimental group increased. However, the fruit firmness of the control group began to decrease from 9 days to 18 days, then increased from 18 days to 27 days, but showed a downward trend from 27 days to 36 days. In contrast, the fruit firmness of the experimental group reached a peak at 27 days. Compared with ordinary cold storage, AI fresh-keeping storage performs better in maintaining the firmness of cherry fruit.

[0092] 4. Breathing intensity Gas collection and measurement: Open the measuring container and place it upside down to balance the gas in the container with the air. Then put in about 150g of fruit and place it in a sealed storage environment for 1h. When sampling, shake the container gently to evenly distribute the gas in the container. Use an injector to take 1.0ml of gas from the top and measure the chromatogram obtained after detection by the chromatograph. The CO2 fraction of the injected gas can be obtained by the peak area. The respiration intensity is expressed as milliliters of CO2 released per kilogram of fruit per hour, and the unit is mLCO2 / kg-h. Gas chromatography configuration and working conditions: The gas chromatograph is equipped with a stainless steel packed chromatographic column (8700 model) and a thermal conductivity detector (TCD). The carrier gas is He, and the flow rate is 30ml / min. The injection temperature is 120℃, the column temperature is 35℃, the detection temperature is 120℃, the pressure is 75.6kPa, the injection volume is 1000μl, and the current is 30mA.

[0093] like Fig.11As shown, in the first 28 days of storage, the respiration intensity of cherry fruits showed a gradually increasing trend in both the control group and the experimental group. From the 28th day to the 35th day, the respiration intensity of cherry fruits began to gradually decrease. From the 0th day to the 14th day, the difference in the respiration intensity of cherry fruits between the treatment groups was not obvious and was relatively close. However, starting from the 21st day, the rate of increase in the respiration intensity of cherry fruits in the control group was significantly accelerated, significantly higher than that in the experimental group. This phenomenon indicates that at this stage, the respiration activity of cherry fruits in the control group is more active. From the 7th day to the 35th day, the respiration intensity of sweet cherry fruits in the experimental group continued to be lower than that in the control group. This result shows that the respiration intensity of cherry fruits in the experimental group has been effectively controlled and suppressed. By comparing the data of the experimental group and the control group, it can be seen that the AI ​​preservation technology described in this application has a significant effect in suppressing the respiration intensity of cherry fruits.

[0094] 5. Malondialdehyde (MDA) content Refer to Gao Junfeng's method and make appropriate modifications. Take 50g of fruit, peel it, chop the pulp, stir it evenly, take 2g of pulp, add 0.2g of polyvinyl pyrrolidone in 10mL of 0.05mol / L phosphate buffer (pH7.8), grind it in an ice bath, centrifuge it at 4000r / min at 4℃ for 30min, and take the supernatant to measure the enzyme activity and MDA content. Add 1.5mL of crude enzyme extract to 2.5mL of 0.5% thiobarbituric acid (prepared with 20% trichloroacetic acid solution) solution, mix well, boil it in boiling water for 20min, cool it quickly, and centrifuge it at 5000r / min at 4℃ for 15min. Take the supernatant and measure the absorbance at wavelengths of 532nm and 600nm respectively. The calculation formula is as follows:

[0095] The increased MDA (malondialdehyde) content in the fruit indicates that the degree of oxidative damage has increased. Fig.12 As shown, during the storage period of cherry fruits after harvest, the MDA content gradually increased with the increase of storage time. From 0 days to 18 days, the MDA content of cherry fruits in different treatment groups changed relatively little. However, from 21 days to 28 days, the MDA content increased significantly, and both the control group and the experimental group reached a peak on the 28th day. Subsequently, from 28 days to 35 days, the MDA content began to decrease slowly. It is worth noting that from the 7th day, the MDA content of the control group continued to be higher than that of the experimental group. On the 35th day, the MDA contents of the control group and the experimental group were 22.49 μmol / g and 18.92 μmol / g, respectively. This shows that the AI ​​preservation technology described in the present application can effectively inhibit the increase of MDA content in cherry fruits to a certain extent.

[0096] 6. Fruit titratable acid (TA) content, browning rate, rot rate, good fruit rate, fresh food quality The titratable acid (TA) content was determined by acid-base titration: First, 10 g of pulp was ground, then 60 ml of distilled water was added, and the mixture was washed into a 100 ml conical flask. Next, the conical flask was placed in an 80°C water bath for 30 minutes of leaching, during which it was stirred every 5 minutes. After the leaching was completed, the mixture was taken out and cooled, and then filtered. Finally, 0.1 mol / L sodium hydroxide solution was used for titration, and 0.067 was used as the conversion factor (in terms of malic acid) to calculate the content of organic acids.

[0097] The browning index is divided into five levels: level 0, no browning, level 1, mild browning, level 2, less than half of the browning area, level 3, more than half of the browning area, and level 4, complete browning. The rot index is also divided into five levels: level 0, no rot, level 1, small area of ​​lesions, level 2, less than a quarter of the rotten area, level 3, between a quarter and half of the rotten area, and level 4, more than half of the rotten area. The good fruit rate refers to fruits without rot and browning. The quality of fresh food is scored by taste, which is divided into five levels: 5 points means the taste is the same or better than when it was picked, 4 points means the taste is normal, 3 points means the taste is slightly lower, 2 points means the taste is light, and 1 point means the flavor is extremely light or has an odor.

[0098] The calculation formula is as follows:

[0099]

[0100]

[0101] The specific measurement results are shown in Table 2.

[0102] Table 2 Titratable acid (TA) content, browning rate, rot rate, good fruit rate, and fresh food quality of the fruit

[0103] As shown in Table 2, within the first 21 days of storage, no rot occurred in the sweet cherry fruits of the control group and the experimental group. However, from the 28th day, the cherry fruits in the control group began to rot, and the rot rate gradually increased with the increase in storage time. In contrast, the rot rate of cherry fruits in the experimental group was significantly reduced. On the 35th day of storage, the rot rate of cherry fruits in the control group reached 23.4%, while the rot rate of the experimental group on the 49th day was only 2.8%. This shows that the AI ​​preservation technology described in this application can significantly inhibit the rot process of cherry fruits.

[0104] Browning can seriously affect the sensory quality of cherries, and in severe cases can even cause fruit rot. As can be seen from Table 2, in the first 21 days of storage, no browning occurred in the cherries in either group; but on the 35th day, the browning index of the control group reached 27.6%, while that of the experimental group was only 1.7%, with a significant difference between the two groups (P < 0.05). Fig.13 As shown, the AI ​​preservation technology described in this application can significantly reduce the browning phenomenon of cherry fruits.

[0105] In summary, the AI ​​preservation technology described in this application performs well in cherry storage. It not only significantly extends the shelf life of cherries, but also effectively reduces the rot and mildew of cherries during storage, ensures the freshness and taste of cherries, and provides consumers with a better fruit experience.

[0106] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand 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. An AI cold storage intelligent control method based on state and fresh-keeping temperature coordination, characterized in that: include: Obtain the surface state image of cherries in the warehouse collected by the camera and the time queue of the real-time preservation temperature value collected by the temperature sensor; Extracting biological state features of the surface state image of the cherries in the warehouse to obtain a biological state feature map of the cherries in the warehouse; Extracting the time series correlation features of the time queue of the real-time fresh-keeping temperature value to obtain the fresh-keeping temperature time series correlation implicit feature vector; Cross-modal fusion is performed on the biological state feature graph of cherries in the warehouse and the time series associated implicit feature vector of the fresh-keeping temperature to obtain a deep cross-modal fusion coding feature graph of cherry biological feature-fresh-keeping temperature; Based on the cherry biological feature-fresh-keeping temperature deep cross-modal fusion encoding feature map, determine whether the fresh-keeping temperature value is within the recommended fresh-keeping temperature range; Among them, the biological state characteristic graph of cherries in the warehouse and the implicit feature vector associated with the fresh-keeping temperature time series are cross-modally fused, including: extracting the intrinsic common features between the implicit feature vector associated with the fresh-keeping temperature time series and the biological state characteristic graph of cherries in the warehouse to obtain the intrinsic common feature vector of cherry biological feature-fresh-keeping temperature; based on the intrinsic common feature vector of cherry biological feature-fresh-keeping temperature, performing dual feature modulation on the biological state characteristic graph of cherries in the warehouse to obtain the deep cross-modal fusion coding feature graph of cherry biological feature-fresh-keeping temperature.

2. The AI ​​cold storage intelligent control method based on state and fresh-keeping temperature coordination according to claim 1 is characterized in that: Extracting biological state features of the surface state image of cherries in the warehouse to obtain a biological state feature map of cherries in the warehouse includes: The surface state image of the cherries in the warehouse is input into a cherry biological feature extractor based on a deep separable convolutional neural network model to obtain a biological state feature map of the cherries in the warehouse.

3. The AI ​​cold storage intelligent control method based on state and fresh-keeping temperature coordination according to claim 2 is characterized in that: Extracting the time series correlation features of the time queue of the real-time fresh-keeping temperature value to obtain the fresh-keeping temperature time series correlation implicit feature vector includes: The time queue of the real-time fresh-keeping temperature value is input into the fresh-keeping temperature time series feature extractor based on the forward LSTM model to obtain the fresh-keeping temperature time series associated implicit feature vector.

4. The AI ​​cold storage intelligent control method based on state and fresh-keeping temperature coordination according to claim 3 is characterized in that: Extracting the inherent common features between the time series associated implicit feature vector of the fresh-keeping temperature and the biological state feature graph of cherries in the warehouse to obtain the inherent common feature vector of cherry biological feature-fresh-keeping temperature includes: Performing feature distillation on the biological state feature graph of cherries in the warehouse to obtain a biological state distillation feature vector of cherries in the warehouse; The temporal association implicit feature vector of the fresh-keeping temperature and the biological state distillation feature vector of the cherries in the library are input into a multimodal implicit association feature capture network to obtain the intrinsic commonality feature vector of the cherry biological feature-fresh-keeping temperature.

5. The AI ​​cold storage intelligent control method based on state and fresh-keeping temperature coordination according to claim 4 is characterized in that: Performing feature distillation on the biological state feature graph of cherries in the warehouse to obtain a biological state distillation feature vector of cherries in the warehouse, including: Performing point convolution encoding on the biological state characteristic graph of cherries in the storage to obtain a channel modulation biological state characteristic graph of cherries in the storage; The biological state feature map of cherries in the channel modulation library is subjected to global mean pooling processing along the channel dimension and then input into a Sigmoid function for activation processing to obtain a distilled feature vector of the biological state of cherries in the library.

6. The AI ​​cold storage intelligent control method based on state and fresh-keeping temperature coordination according to claim 5 is characterized in that: Based on the intrinsic common feature vector of the cherry biological feature-fresh-keeping temperature, dual feature modulation is performed on the biological state feature map of cherries in the warehouse to obtain the cherry biological feature-fresh-keeping temperature deep cross-modal fusion coding feature map, including: Input the common feature vector of the cherry biological feature-keeping temperature into the feature preliminary screening module based on the gate control unit to obtain the biological state feature preliminary screening weight vector of cherries in the primary library; Based on the primary screening weight vector of the biological state characteristics of cherries in the first-level warehouse, the biological state characteristic diagram of cherries in the warehouse is preliminarily screened to obtain a primary screening biological state characteristic diagram of cherries in the warehouse; Inputting the common feature vector of the cherry biological feature-keeping temperature into the feature secondary screening module based on the multi-level gating unit to obtain the cherry biological state feature screening weight vector in the secondary library; Based on the biological state feature screening weight vector of cherries in the secondary warehouse, the biological state feature map of cherries in the primary screening warehouse is subjected to secondary feature screening to obtain the cherry biological feature-shelf temperature deep cross-modal fusion coding feature map.

7. The AI ​​cold storage intelligent control method based on state and fresh-keeping temperature coordination according to claim 6 is characterized in that: The common feature vector of the biological feature of cherries and the preservation temperature is input into the feature initial screening module based on the gate control unit to obtain the biological state feature initial screening weight vector of cherries in the first-level library, including: After multiplying the cherry biological characteristic-keeping temperature intrinsic commonality feature vector by the first weight parameter matrix, the first bias term is added to obtain a first-level cherry biological characteristic-keeping temperature intrinsic commonality modulation vector; Inputting the first-level cherry biological characteristic-keeping temperature intrinsic common modulation vector into the Sigmoid function for nonlinear activation processing to obtain the first-level cherry biological characteristic-keeping temperature intrinsic common modulation activation vector; The first-level cherry biological characteristic-fresh-keeping temperature intrinsic common modulation activation vector is input into the mask unit to obtain the initial screening weight vector of the biological state characteristics of cherries in the first-level library.

8. The AI ​​cold storage intelligent control method based on state and fresh-keeping temperature coordination according to claim 7 is characterized in that: The common feature vector of the biological feature of cherries and the preservation temperature is input into the feature secondary screening module based on the multi-level gating unit to obtain the biological state feature screening weight vector of cherries in the secondary library, including: After multiplying the first-level cherry biometric-keeping temperature intrinsic common modulation vector by the second weight parameter matrix, the second-level cherry biometric-keeping temperature intrinsic common modulation vector is obtained by performing dot addition with the second bias term; Inputting the secondary cherry biological characteristic-keeping temperature intrinsic common modulation vector into a Sigmoid function for nonlinear activation processing to obtain a secondary cherry biological characteristic-keeping temperature intrinsic common modulation activation vector; The secondary cherry biological characteristic-fresh-keeping temperature intrinsic common modulation activation vector is input into the mask unit to obtain the cherry biological state characteristic screening weight vector in the secondary library.

9. The AI ​​cold storage intelligent control method based on state and fresh-keeping temperature coordination according to claim 8 is characterized in that: Based on the cherry biological feature-fresh-keeping temperature deep cross-modal fusion encoding feature map, determining whether the fresh-keeping temperature value is within the recommended fresh-keeping temperature range includes: The cherry biological feature-keeping temperature deep cross-modal fusion encoding feature map is input into the classifier-based keeping temperature optimization module to obtain an optimization result, and the optimization result is used to indicate whether the keeping temperature value is within the recommended keeping temperature range.

10. An AI cold storage intelligent control system based on state and fresh-keeping temperature coordination, characterized in that: include: The cherry preservation information acquisition module is used to obtain the surface state image of the cherries in the warehouse collected by the camera and the time queue of the real-time preservation temperature value collected by the temperature sensor; A biological state feature extraction module, used for extracting biological state features of the surface state image of the cherries in the warehouse to obtain a biological state feature map of the cherries in the warehouse; A fresh-keeping temperature time series encoding module is used to extract the time series correlation features of the time queue of the real-time fresh-keeping temperature value to obtain a fresh-keeping temperature time series correlation implicit feature vector; A cross-modal fusion module is used to perform cross-modal fusion on the biological state feature map of cherries in the warehouse and the time series associated implicit feature vector of the fresh-keeping temperature to obtain a deep cross-modal fusion coding feature map of cherry biological feature-fresh-keeping temperature; The fresh-keeping temperature suitability assessment module is used to determine whether the fresh-keeping temperature value is within the recommended fresh-keeping temperature range based on the cherry biological feature-fresh-keeping temperature deep cross-modal fusion encoding feature map.

Images