Real-time monitoring and management system and method for tobacco transportation

Abstract

The invention provides a tobacco transportation real-time monitoring and management system and method. The system comprises an environment sensor array, a tobacco transportation state analysis model and a state analysis module. The environment sensor collects and preprocesses transportation environment data to generate standardized environment parameters. The tobacco transportation state analysis model is composed of three data processing modules which are connected in sequence, and each module comprises a data fusion unit and a plurality of multi-dimensional parameter analysis units. The data fusion unit generates fusion features from the input data and transmits the fusion features to the multi-dimensional parameter analysis unit. And the multi-dimensional parameter analysis unit generates weighted feature data through the environmental parameter attention module, and then outputs enhanced feature data through the data enhancement module. And the state analysis module performs parameter integration and decision mapping on the model output data to generate a transportation state evaluation result. According to the invention, high-precision dynamic monitoring and state evaluation of environment abnormity in the tobacco transportation process can be realized, and the intelligent level of transportation quality management and control is improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent logistics monitoring technology, and more specifically, to a real-time monitoring and management system and method for tobacco leaf transportation. Background Art

[0002] In the tobacco transportation sector, existing monitoring methods rely heavily on simple sensor data collection, providing only basic environmental parameters such as temperature and humidity, but lacking in-depth analysis and comprehensive processing of this data. While these traditional methods can provide a basic understanding of the transportation environment, their inability to effectively integrate multi-dimensional data and their lack of ability to capture spatiotemporal correlations result in inaccurate overall assessments of transportation status. Furthermore, existing monitoring systems often overlook the dynamic nature of data during data processing, making it difficult to adapt to the complex and changing environmental conditions during transportation, resulting in an inability to promptly and accurately identify potential environmental anomalies.

[0003] In the process of implementing the embodiments of the present invention, the inventors found that there are at least the following problems or defects in the existing technology: it is impossible to fully perceive the spatiotemporal correlation characteristics of the transportation environment, the data processing accuracy is insufficient, it is difficult to achieve real-time and accurate assessment of the transportation status, and there is a lack of effective model optimization mechanism, and it is unable to adapt to complex and changeable transportation environment conditions. These problems have seriously affected the quality assurance and safety management during the transportation of tobacco leaves. Summary of the Invention

[0004] The present invention provides a tobacco leaf transportation real-time monitoring and management system and method.

[0005] In a first aspect of the present invention, a tobacco leaf transportation real-time monitoring and management system is provided, comprising:

[0006] an environmental sensor array configured to collect transportation environment data and output standardized environmental parameters;

[0007] The tobacco transport status analysis model is composed of three sequentially connected data processing modules, each of which includes a sequentially connected data fusion unit and at least two multidimensional parameter analysis units, wherein the data fusion unit of the first data processing module receives the standardized environmental parameters;

[0008] The data fusion unit is configured to perform feature fusion processing on the input data and output the result to the corresponding multi-dimensional parameter analysis unit;

[0009] The multidimensional parameter analysis unit includes an environmental parameter attention module and a data enhancement module, wherein: the environmental parameter attention module is configured to generate a feature vector through multidimensional filtering, generate superposition features after performing dynamic normalization and spatiotemporal correlation analysis on the feature vector, generate aggregation parameters based on extreme value sampling and mean sampling, and generate weighted feature data through weight allocation;

[0010] The data enhancement module is configured to sequentially perform feature dimensionality reduction, multi-dimensional filtering, nonlinear activation, noise suppression, and feature dimensionality increase operations on the input data to generate enhanced feature data;

[0011] The output of the multi-dimensional parameter analysis unit is determined by the weighted feature data and the enhanced feature data and serves as the input of the next processing unit;

[0012] The state analysis module is connected to the terminal data processing module and is configured to perform feature integration and decision mapping on the output data of the tobacco transportation state analysis model to generate a transportation state evaluation result.

[0013] Furthermore, the data fusion unit in the first data processing module is a basic data fusion unit, and the data fusion units in the other data processing modules are multi-dimensional data fusion units;

[0014] The basic data fusion unit includes using a first filter network to process the standardized environmental parameters to obtain primary features, and performing data standardization processing on the primary features to obtain output data of the basic data fusion unit;

[0015] The multidimensional data fusion unit includes performing a second filtering network and feature dimensionality reduction processing on the output data of the previous data processing module to obtain a first processing result and a second processing result; performing channel calibration processing on the first processing result to obtain a third processing result; and inputting the third processing result and the second processing result into the feature splicing module to obtain the output data of the multidimensional data fusion unit.

[0016] Furthermore, the multidimensional parameter analysis unit further includes a first data normalization layer, a first feedback network, a second data normalization layer, a second feedback network and a feature normalization layer;

[0017] Processing the first input data input to the multidimensional parameter analysis unit using the first data normalization layer to obtain second input data;

[0018] The environmental parameter attention module processes the second input data to obtain weighted feature data;

[0019] Adding the weighted feature data to the first input data through the first feedback network, and adding the addition result to the second data normalization layer;

[0020] Processing the input data of the layer using the second data normalization layer to obtain third input data;

[0021] The data enhancement module processes the third input data to obtain enhanced feature data;

[0022] The enhanced feature data and the addition result are added to the feature normalization layer through the second feedback network to obtain the output data of the multidimensional parameter analysis unit.

[0023] Furthermore, the tobacco transport status analysis model includes three data processing modules connected in sequence, the second data processing module is provided with three multidimensional parameter analysis units connected in sequence, and the remaining data processing modules are provided with two multidimensional parameter analysis units connected in sequence.

[0024] Furthermore, the state analysis module includes a parameter integration layer and a decision mapping layer. The parameter integration layer is used to perform 1×1 parameter aggregation on the input data of the state analysis module in the feature dimension; and the decision mapping layer is used to perform nonlinear mapping on the output result of the parameter integration layer to obtain the transportation state assessment result.

[0025] Furthermore, the tobacco transport state analysis model is optimized using the constructed environmental anomaly discrimination loss function. The expression of the environmental anomaly discrimination loss function is:

[0026]

[0027] in, represents the environmental anomaly factor of the i-th sample, Indicates that the model correctly judges sample i as the actual state y i Confidence, s i,j represents the feature score of the model predicting sample i as state j, C represents the number of transportation environment state categories, α j represents the frequency factor of state category j, n j represents the number of samples in category j, represents the number of samples in the category to which sample i belongs, μ = 0, σ = 0.1 represents the distribution parameters of environmental parameters.

[0028] Furthermore, the window size of the first multidimensional filtering processing is 5×5×3, a 7×7×5 three-dimensional convolution kernel is used in the spatiotemporal correlation analysis processing, the expansion interval is 2, a fully connected layer is used in the linear transformation of the aggregation parameters, the window size of the second multidimensional filtering processing is 3×3×2, and the nonlinear activation layer applies the ReLU function.

[0029] Furthermore, the feature vectors are subjected to dynamic normalization processing and spatiotemporal correlation analysis processing respectively, and the two processing results are superimposed to obtain superimposed features:

[0030]

[0031] Among them, F conv1 represents the result of the first multi-dimensional filtering process, Norm(·) represents dynamic normalization processing, Conv dilated (·) represents dilated convolution processing, Indicates an overlay operation.

[0032] Furthermore, the convolution kernel size of the first filtering network is 5×5×2, the convolution kernel size of the second filtering network is 5×5×1, and the sliding step size is 1; the feature dimensionality reduction processing uses a 2×2×1 convolution kernel with a step size of 1.

[0033] In a second aspect of the present invention, a method for real-time monitoring and management of tobacco transportation is provided, comprising:

[0034] Step 1: Collect transportation environment data and preprocess them to obtain standardized environmental parameters;

[0035] Step 2: Using the basic data fusion unit to perform feature fusion on the standardized environmental parameters to obtain the first output parameter;

[0036] Step 3: Input the first output parameter into a plurality of sequentially connected multidimensional parameter analysis units to obtain a second output parameter as input data for a multidimensional data fusion unit; and capture the spatiotemporal correlation characteristics of the transportation environment through the multidimensional perception capabilities of the multidimensional parameter analysis units;

[0037] Step 4: Using the multi-dimensional data fusion unit to perform multi-scale feature fusion processing on the input data of the unit to obtain the third output parameter;

[0038] Step 5: Inputting the third output parameter into a plurality of sequentially connected multidimensional parameter analysis units to obtain a fourth output parameter, and inputting the fourth output parameter into a multidimensional data fusion unit;

[0039] Step 6: Repeat steps 4 to 5 until a preset number of iterations is reached, and output the fifth output parameter;

[0040] Step 7: Input the fifth output parameter into the status analysis module to obtain a transportation environment status assessment result.

[0041] The above-described embodiments of the present invention have at least the following beneficial effects: the real-time monitoring and management system and method for tobacco transportation can improve the accuracy of environmental monitoring during tobacco transportation. By collecting transportation environment data through an environmental sensor array and preprocessing it to obtain standardized environmental parameters, and then performing in-depth analysis through multiple data processing modules in the tobacco transportation status analysis model, the temporal and spatial correlation characteristics and multi-scale characteristics of the transportation environment can be accurately captured. This multi-dimensional, multi-level data processing method ensures a comprehensive perception of the transportation environment status, providing accurate and reliable data support for subsequent status assessments, thereby better ensuring that the quality of tobacco during transportation is not adversely affected by environmental factors.

[0042] Furthermore, the system enables real-time assessment and optimization of tobacco transportation status. The status analysis module rapidly derives transportation status assessment results based on the output data of the tobacco transportation status analysis model. It then optimizes the model using a constructed environmental anomaly discrimination loss function, further improving the model's ability to discern environmental anomalies. This real-time monitoring and assessment mechanism promptly identifies environmental anomalies that may arise during transportation, allowing relevant personnel to quickly implement appropriate adjustments and address them, effectively reducing transportation risks and improving the overall efficiency and safety of tobacco transportation. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily apparent by reading the following detailed description with reference to the accompanying drawings, in which several embodiments of the present invention are shown by way of example and not limitation, in which:

[0044] Figure 1 A schematic diagram of the structure of a real-time monitoring and management system for tobacco transportation provided by one embodiment of the present invention;

[0045] Figure 2 A schematic flow chart of a method for real-time monitoring and management of tobacco transportation provided by one embodiment of the present invention;

[0046] Figure 3 The figure schematically shows the structure of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The principles and spirit of the present invention will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are provided solely to enable those skilled in the art to better understand and implement the present invention, and are not intended to limit the scope of the present invention in any way. Rather, these embodiments are provided to make the present invention more thorough and complete, and to fully convey the scope of the present invention to those skilled in the art.

[0048] Those skilled in the art will appreciate that the embodiments of the present invention may be implemented as a system, apparatus, device, method, or computer program product. Therefore, the present invention may be implemented in the following forms: entirely in hardware, entirely in software (including firmware, resident software, microcode, etc.), or in a combination of hardware and software.

[0049] It should be noted that any number of elements in the drawings is for illustration only and not for limitation, and any naming is only for distinction and does not have any limiting meaning.

[0050] Reference below Figure 1 , Figure 1 This is a schematic diagram of the structure of a tobacco transportation real-time monitoring and management system provided by one embodiment of the present invention. Figure 1 As shown, a tobacco leaf transportation real-time monitoring and management system includes:

[0051] An environmental sensor array 101 is configured to collect transportation environment data and output standardized environmental parameters;

[0052] The tobacco transport status analysis model 102 is composed of three sequentially connected data processing modules, each of which includes a sequentially connected data fusion unit and at least two multi-dimensional parameter analysis units, wherein the data fusion unit of the first data processing module receives the standardized environmental parameters;

[0053] The data fusion unit is configured to perform feature fusion processing on the input data and output the result to the corresponding multi-dimensional parameter analysis unit;

[0054] The multidimensional parameter analysis unit includes an environmental parameter attention module and a data enhancement module, wherein: the environmental parameter attention module is configured to generate a feature vector through multidimensional filtering, generate superposition features after performing dynamic normalization and spatiotemporal correlation analysis on the feature vector, generate aggregation parameters based on extreme value sampling and mean sampling, and generate weighted feature data through weight allocation;

[0055] The data enhancement module is configured to sequentially perform feature dimensionality reduction, multi-dimensional filtering, nonlinear activation, noise suppression, and feature dimensionality increase operations on the input data to generate enhanced feature data;

[0056] The output of the multi-dimensional parameter analysis unit is determined by the weighted feature data and the enhanced feature data and serves as the input of the next processing unit;

[0057] The state analysis module 103 is connected to the terminal data processing module and is configured to perform feature integration and decision mapping on the output data of the tobacco transport state analysis model to generate a transport state evaluation result.

[0058] Specifically, the environmental sensor array 101 collects transportation environment data and pre-processes it to obtain standardized environmental parameters.

[0059] The tobacco transport state analysis model 102 includes three sequentially connected data processing modules. The standardized environmental parameters are input to the front end of the first data processing module of the tobacco transport state analysis model. The output of any data processing module serves as the input of the next data processing module connected thereto. Each data processing module includes a sequentially connected data fusion unit and multiple multidimensional parameter analysis units.

[0060] The data fusion unit generates fusion feature data from the received data, and inputs the processed data into the multi-dimensional parameter analysis unit connected thereto;

[0061] The multidimensional parameter analysis unit includes an environmental parameter attention module and a data enhancement module. The environmental parameter attention module includes performing a first multidimensional filtering process on the input data of the module to obtain a feature vector; performing dynamic normalization processing and spatiotemporal correlation analysis processing on the feature vector, and superimposing the two processing results to obtain a superposition feature; performing extreme value sampling and mean sampling processing on the superposition feature, and merging the results of the two samplings to obtain an aggregation parameter; performing linear transformation and activation function processing on the aggregation parameter to obtain an attention parameter, and averaging the attention parameter in the dimensional space to obtain a first weight coefficient and a second weight coefficient; performing dimensional segmentation on the superposition feature to obtain a first superposition feature and a second superposition feature, and processing and combining the first superposition feature and the second superposition feature based on the first weight coefficient and the second weight coefficient to obtain weighted feature data;

[0062] The data enhancement module includes sequentially subjecting the input data of the module to feature dimensionality reduction processing, second multi-dimensional filtering processing, nonlinear activation layer, noise suppression layer and feature dimensionality increase processing to obtain enhanced feature data;

[0063] Obtaining output data of the multidimensional parameter analysis unit based on the weighted feature data and the enhanced feature data as input data of the adjacent connected multidimensional parameter analysis unit;

[0064] The state analysis module 103 is connected to the end of the tobacco transport state analysis model. The output data of the tobacco transport state analysis model is input into the state analysis module to obtain a transport state evaluation result.

[0065] The real-time monitoring and management system for tobacco transportation of the present invention comprises an environmental sensor array, a tobacco transportation state analysis model, and a state analysis module. The environmental sensor array is used to collect and preprocess transportation environment data to obtain standardized environmental parameters. These parameters are processed numerical values that can be directly used for subsequent analysis and reflect various characteristics of the transportation environment, such as temperature, humidity, and light intensity. The tobacco transportation state analysis model is composed of multiple data processing modules, each of which includes a data fusion unit and a multidimensional parameter analysis unit for deep processing and feature extraction of input data. The data fusion unit integrates data from different sources to generate fused feature data to better reflect the comprehensive state of the transportation environment. The multidimensional parameter analysis unit further analyzes the data to extract more valuable feature information. The state analysis module, based on the output data of the tobacco transportation state analysis model, obtains a transportation state assessment result, which is used to determine whether the transportation process is normal and whether there are potential risks.

[0066] Specifically, the standardized environmental parameters obtained after preprocessing the transportation environment data collected by the environmental sensor array are numerical values converted from the raw data into a unified format and dimension. For example, temperature data is standardized to values within the Celsius range, and humidity data is standardized to percentages. The first data processing module in the tobacco transportation status analysis model receives these standardized environmental parameters and inputs them into the data fusion unit. The basic data fusion unit within the data fusion unit processes the standardized environmental parameters using a first filtering network. The filtering network is a mathematical model used to remove noise from the data and extract useful primary features. Primary features are data sets that reflect the basic characteristics of the data after preliminary processing. Subsequently, the primary features are normalized to meet the requirements of subsequent processing modules. The environmental parameter attention module in the multidimensional parameter analysis unit performs multidimensional filtering on the input data. This method analyzes data across multiple dimensions and extracts features across different dimensions. Dynamic normalization dynamically adjusts the data normalization parameters based on real-time data changes to better adapt to data changes. Spatiotemporal correlation analysis considers the temporal and spatial correlations of the data and analyzes the patterns of data variation across different temporal and spatial locations. Aggregation parameters combine multiple feature parameters to create a comprehensive parameter representation. The data enhancement module enhances the data's feature representation capabilities through a series of operations, including feature dimensionality reduction, multidimensional filtering, nonlinear activation layers, noise suppression layers, and feature dimensionality increase, making it more suitable for subsequent analysis and processing.

[0067] Preferably, the tobacco transportation status analysis model is constructed as follows: First, based on the types of environmental parameters that need to be monitored during tobacco transportation, appropriate sensors are selected to form an environmental sensor array and calibrated to ensure the accuracy and reliability of the collected data. Next, the specific structure and parameter settings of the data fusion unit and multidimensional parameter analysis unit in the data processing module are designed. For example, the first filtering network can adopt a convolutional neural network structure, with the convolution kernel size selected based on the characteristics of the actual data, such as 5×5×2, to extract local features of the data. The convolution kernel size of the second filtering network can be set to 5×5×1 to further refine the features. Feature dimensionality reduction can use methods such as principal component analysis to reduce the dimensionality of the data while retaining the key feature information. In the multidimensional parameter analysis unit, the multidimensional filtering processing of the environmental parameter attention module can extract features of the data in different dimensions by setting an appropriate window size, such as 5×5×3. Dynamic normalization can be performed in real time based on the mean and standard deviation of the data to ensure data stability. Spatiotemporal correlation analysis can use a three-dimensional convolution kernel, such as 7×7×5, with a dilation interval of 2, to capture temporal and spatial correlations in the data. Linear transformation of aggregated parameters can be achieved through a fully connected layer, integrating multiple feature parameters into a comprehensive parameter representation. The nonlinear activation layer in the data enhancement module can use the ReLU function to enhance the nonlinear characteristics of the data. Finally, the parameter integration layer and decision mapping layer in the state analysis module can be designed according to actual needs. For example, the parameter integration layer can perform 1×1 parameter aggregation on the input data in the feature dimension, and the decision mapping layer can use nonlinear mapping methods such as the softmax function to map the aggregated parameters to the transportation status assessment results. During the model optimization process, the constructed environmental anomaly discrimination loss function can be used to adjust the model parameters to make the model more capable of discerning environmental anomalies.

[0068] In some embodiments, the data fusion unit in the first data processing module is a basic data fusion unit, and the data fusion units in the other data processing modules are multi-dimensional data fusion units;

[0069] The basic data fusion unit includes using a first filter network to process the standardized environmental parameters to obtain primary features, and performing data standardization processing on the primary features to obtain output data of the basic data fusion unit;

[0070] The multidimensional data fusion unit includes performing a second filtering network and feature dimensionality reduction processing on the output data of the previous data processing module to obtain a first processing result and a second processing result; performing channel calibration processing on the first processing result to obtain a third processing result; and inputting the third processing result and the second processing result into the feature splicing module to obtain the output data of the multidimensional data fusion unit.

[0071] It should be noted that the real-time monitoring and management system for tobacco transportation mentioned in the present invention has a clear hierarchical division in the design of the data processing module, wherein the data fusion unit in the first data processing module is called the basic data fusion unit, while the data fusion units in other data processing modules are multidimensional data fusion units. The main function of the basic data fusion unit is to perform preliminary processing on the standardized environmental parameters initially input, and generate primary feature data that can be used for subsequent analysis. The multidimensional data fusion unit, based on the basic data fusion unit, further integrates and optimizes the data in multiple dimensions to extract more representative and accurate feature information, providing richer data support for subsequent transportation status assessment. This hierarchical data fusion method can better adapt to the data processing requirements at different stages and improve the data processing efficiency and accuracy of the entire system.

[0072] Specifically, the first filtering network in the basic data fusion unit is a mathematical model used for preliminary processing of standardized environmental parameters. Its function is to remove noise from the data and extract primary features. Primary features are data sets that, after preliminary processing, reflect the basic characteristics of the data, such as temperature trends extracted from temperature data. Data standardization converts the primary feature data into a unified format and dimension, enabling better data processing by subsequent processing modules. The second filtering network and feature dimensionality reduction in the multidimensional data fusion unit are key steps for further data processing. The second filtering network is used to further filter the output data of the previous data processing module to extract more refined features. Feature dimensionality reduction reduces the data dimension to remove redundant information while retaining the key feature information, for example, using methods such as principal component analysis. Channel calibration calibrates the different channels of the data to ensure data consistency and accuracy. The feature splicing module splices the data after channel calibration and feature dimensionality reduction to generate the output data of the multidimensional data fusion unit, providing richer input data for the subsequent multidimensional parameter analysis unit.

[0073] Preferably, the first filter network in the basic data fusion unit can employ a convolutional neural network architecture, with its convolution kernel size selected based on the characteristics of the actual data, for example, 5×5×2, to extract local features. Data normalization can employ the Z-score method, converting the data to standardized data with a mean of 0 and a standard deviation of 1 by calculating the mean and standard deviation. The convolution kernel size of the second filter network in the multidimensional data fusion unit can be set to 5×5×1 to further refine features. Feature dimensionality reduction can employ principal component analysis, calculating the data covariance matrix and eigenvalues to extract the principal components and reduce the data dimension. Channel calibration can calculate and adjust the differences between channels to ensure data consistency. The feature splicing module splices the data after channel calibration and feature dimensionality reduction according to specific rules, such as by chronological order or spatial location, to generate the output data for the multidimensional data fusion unit. This hierarchical data fusion approach can better adapt to the data processing requirements of different stages, improve the data processing efficiency and accuracy of the entire system, and provide more reliable data support for real-time monitoring and assessment of tobacco transportation status.

[0074] In some embodiments, the multidimensional parameter analysis unit further comprises a first data normalization layer, a first feedback network, a second data normalization layer, a second feedback network, and a feature normalization layer;

[0075] Processing the first input data input to the multidimensional parameter analysis unit using the first data normalization layer to obtain second input data;

[0076] The environmental parameter attention module processes the second input data to obtain weighted feature data;

[0077] Adding the weighted feature data to the first input data through the first feedback network, and adding the addition result to the second data normalization layer;

[0078] Processing the input data of the layer using the second data normalization layer to obtain third input data;

[0079] The data enhancement module processes the third input data to obtain enhanced feature data;

[0080] The enhanced feature data and the addition result are added to the feature normalization layer through the second feedback network to obtain the output data of the multidimensional parameter analysis unit.

[0081] It should be noted that the multidimensional parameter analysis unit mentioned in the present invention introduces multiple key modules in the data processing process, including a first data normalization layer, a first feedback network, a second data normalization layer, a second feedback network and a feature normalization layer. The setting of these modules is to further optimize the data processing process and improve the accuracy and stability of feature extraction. The function of the first data normalization layer is to perform preliminary standardization processing on the input data so that subsequent modules can better process the data. The environmental parameter attention module uses weighted feature data to highlight important features and enhance the model's perception of key information. The first feedback network and the second feedback network further enhance the feature representation capability of the data by combining the processed data with the original data. The feature normalization layer normalizes the final output data to ensure the consistency and comparability of the output results. This multi-layer feedback and normalization processing method can effectively improve the robustness and accuracy of the system.

[0082] Specifically, the first data normalization layer normalizes the input data, converting it into a unified format and dimension for better processing by subsequent modules. For example, Z-score normalization can be used to adjust the data's mean to 0 and its standard deviation to 1. The environmental parameter attention module weights the input data to highlight important feature information. Weighted feature data assigns different weights based on its importance, thereby enhancing the model's ability to perceive key information. The first and second feedback networks combine the processed data with the original data, further enhancing the data's feature representation capabilities through feedback mechanisms. The feedback network can be implemented through methods such as weighted summation to fuse the processed and original data. The feature normalization layer normalizes the final output data to ensure consistency and comparability. Normalization can use the maximum and minimum normalization method to adjust the data range to between 0 and 1. The configuration of these modules enables the multidimensional parameter analysis unit to more efficiently process data and extract more valuable feature information.

[0083] Preferably, the construction process of the multidimensional parameter analysis unit can be further refined. First, the first data standardization layer can adopt the Z-score standardization method to convert the data into standardized data with a mean of 0 and a standard deviation of 1 by calculating the mean and standard deviation of the data. When processing data, the environmental parameter attention module can dynamically adjust the weight according to the importance and relevance of the data, for example, by calculating the confidence and correlation coefficient of the data to determine the weight. The first feedback network and the second feedback network can be implemented by weighted summation to fuse the processed data with the original data to enhance the feature representation ability of the data. The weight of the feedback network can be dynamically adjusted according to the feature importance of the data to ensure the accuracy and stability of the data. The feature normalization layer can adopt the maximum and minimum normalization method to adjust the range of the data to between 0 and 1 to ensure the consistency and comparability of the output results. Through these optimization steps, the multidimensional parameter analysis unit can process data more efficiently, extract more valuable feature information, and provide more reliable data support for real-time monitoring and evaluation of tobacco transportation status.

[0084] In some embodiments, the tobacco transport status analysis model includes three data processing modules connected in sequence, the second data processing module is provided with three multidimensional parameter analysis units connected in sequence, and the remaining data processing modules are provided with two multidimensional parameter analysis units connected in sequence.

[0085] It should be noted that the tobacco transport status analysis model mentioned in the present invention is composed of three data processing modules connected in sequence, of which the second data processing module contains three multidimensional parameter analysis units, and the remaining data processing modules contain two multidimensional parameter analysis units connected in sequence. This structural design is to process data more carefully. By increasing the number of analysis units in the second module, it is possible to more deeply explore the characteristic information in the data, thereby improving the accuracy and reliability of the assessment of the tobacco transport status. The data processing module is the core part of the system, which is used to perform multi-level processing and feature extraction on the input standardized environmental parameters, while the multidimensional parameter analysis unit is the module in each module used for specific analysis and processing of data. Through the collaborative work of multiple units, it is possible to better capture the complex relationships and patterns in the data.

[0086] Specifically, the data processing module is the basic building block of the tobacco transport status analysis model. Each module consists of a data fusion unit and a multidimensional parameter analysis unit. The data fusion unit in the first data processing module is the basic data fusion unit, primarily responsible for preliminary processing of the input standardized environmental parameters to generate primary feature data. The multidimensional parameter analysis unit, on the other hand, further analyzes and processes this primary feature data, extracting more valuable feature information through components such as the environmental parameter attention module and the data augmentation module. The second data processing module contains three multidimensional parameter analysis units, meaning that at this stage, the data undergoes three stages of detailed analysis and processing, enabling deeper exploration of the data's feature information. The remaining data processing modules contain two multidimensional parameter analysis units. This design ensures in-depth data processing while avoiding unnecessary computational complexity. Each multidimensional parameter analysis unit incorporates multiple processing steps, such as feature dimensionality reduction, multidimensional filtering, and nonlinear activation. These steps work together to ensure that the data is effectively processed and optimized after passing through each unit.

[0087] Preferably, the construction process of the tobacco transportation status analysis model can be further refined. First, based on the types of environmental parameters that need to be monitored during tobacco transportation, appropriate sensors are selected to form an environmental sensor array and calibrated to ensure the accuracy and reliability of the collected data. Next, the specific structure and parameter settings of the data processing module are designed. For example, the basic data fusion unit of the first data processing module can adopt a convolutional neural network architecture, with the convolution kernel size selected based on the characteristics of the actual data, such as 5×5×2, to extract local features of the data. In the second data processing module, the three multidimensional parameter analysis units can each be configured with different parameters to meet the data processing requirements at different stages. For example, the first multidimensional parameter analysis unit can focus on extracting short-term data features, the second unit can analyze medium-term data trends, and the third unit can explore long-term data correlations. The environmental parameter attention module within each multidimensional parameter analysis unit can determine weights by calculating the data's confidence and correlation coefficients to highlight important features. The data enhancement module enhances the data's feature representation capabilities through a series of operations, including feature dimensionality reduction, multidimensional filtering, nonlinear activation layers, noise suppression layers, and feature dimensionality increase. Through this hierarchical, multi-stage data processing method, the tobacco transportation status analysis model can process data more effectively, extract more valuable feature information, and provide more reliable data support for real-time monitoring and evaluation of tobacco transportation status.

[0088] In some embodiments, the state analysis module includes a parameter integration layer and a decision mapping layer. The parameter integration layer is used to perform 1×1 parameter aggregation on the input data of the state analysis module in the feature dimension; and the decision mapping layer is used to perform nonlinear mapping on the output result of the parameter integration layer to obtain the transportation state assessment result.

[0089] It should be noted that the status analysis module is a key component of the real-time monitoring and management system for tobacco transportation. Its primary function is to perform a final status assessment on data processed by multiple data processing modules. The status analysis module consists of a parameter integration layer and a decision mapping layer. The parameter integration layer aggregates input data along the feature dimension to extract more representative feature information. The decision mapping layer maps this aggregated feature information into specific transportation status assessment results, thereby enabling real-time monitoring and assessment of tobacco transportation status. This modular design effectively improves the system's decision-making efficiency and accuracy, ensuring reliable quality assurance of tobacco during transportation.

[0090] Specifically, the parameter integration layer and the decision mapping layer are the two core components of the state analysis module. The parameter integration layer performs 1×1 parameter aggregation on the input data along the feature dimension. This 1×1 parameter aggregation refers to a special aggregation method that compresses and integrates multidimensional feature data along the feature dimension to generate a more representative low-dimensional feature vector. This aggregation method effectively reduces data redundancy while retaining key information, enabling the subsequent decision mapping layer to perform state assessment more efficiently. The decision mapping layer uses nonlinear mapping methods to map the low-dimensional feature vector output by the parameter integration layer into a specific transportation state assessment result. Nonlinear mapping is the process of converting input data into output results through complex nonlinear functions. It can better capture complex relationships and patterns in the data, thereby improving the accuracy of state assessment. In practical applications, nonlinear mapping can be implemented through various methods, such as activation functions in neural networks or complex mathematical models.

[0091] Preferably, the construction and optimization process of the state analysis module can be further refined. In the parameter integration layer, the 1×1 convolution operation in the convolutional neural network can be used to achieve parameter aggregation. 1×1 convolution is a special convolution operation that can perform weighted summation of feature channels without changing the spatial dimension of the data, thereby achieving feature aggregation and compression. For example, when the input data is a multi-channel feature map, the 1×1 convolution can weightedly combine the features of these channels to generate a new feature representation. In the decision mapping layer, the softmax function can be used to achieve nonlinear mapping. The softmax function is a commonly used activation function that can map the input feature vector to a probability distribution, making the output result more interpretable. For example, for the assessment of the transportation status of tobacco leaves, the softmax function can map the input feature vector to a probability distribution of states such as normal and abnormal, thereby achieving accurate classification of the transportation status. In addition, to further improve the performance of the state analysis module, regularization techniques, such as L2 regularization, can be introduced to prevent model overfitting and improve the model's generalization ability. Through these optimization measures, the status analysis module can process input data more efficiently, achieve accurate assessment of tobacco transportation status, and provide reliable technical support for tobacco transportation management.

[0092] In some embodiments, the tobacco transport state analysis model is optimized using the constructed environmental anomaly discrimination loss function, and the expression of the environmental anomaly discrimination loss function is:

[0093]

[0094] in, represents the environmental anomaly factor of the i-th sample, Indicates that the model correctly judges sample i as the actual state y i Confidence, s i,j represents the feature score of the model predicting sample i as state j, C represents the number of transportation environment state categories, α j represents the frequency factor of state category j, n j represents the number of samples of category j, represents the number of samples in the category to which sample i belongs, μ = 0, σ = 0.1 represents the distribution parameters of environmental parameters.

[0095] It should be noted that the environmental anomaly discrimination loss function mentioned in the present invention is a key technical means for optimizing the tobacco transportation status analysis model. This loss function evaluates and adjusts the output of the model by comprehensively considering multiple factors such as the environmental anomaly factor, confidence, feature score, and frequency factor of the state category of the sample. Its purpose is to enable the model to better learn the characteristic patterns of environmental anomalies during the training process, thereby improving the ability to identify abnormal situations, and then improving the reliability and accuracy of the entire monitoring system. The environmental anomaly factor reflects the possibility of abnormalities in the sample in the transportation environment, the confidence level indicates the certainty of the model's sample classification, the feature score is used to measure the feature performance of the sample in different states, and the frequency factor is used to balance the weights of samples of different categories during the training process.

[0096] Specifically, the environmental anomaly factor in the environmental anomaly discrimination loss function is calculated using parameters such as the sample's confidence and state category frequency factor. Confidence refers to the model's degree of certainty that a sample is correctly identified as its actual state, reflecting the model's accuracy in classifying the sample. The feature score is the characteristic representation of a sample when the model predicts it to be in a certain state, and is used to measure the strength of the sample's features under different states. The state category frequency factor is related to the number of samples in the state category and the total number of samples. It is used to adjust the weights of samples of different categories during training to avoid bias in favor of the majority category due to class imbalance. Through the comprehensive calculation of these parameters, the loss function can comprehensively evaluate the model's output and guide the model's continuous optimization during training, thereby improving its ability to discern environmental anomalies.

[0097] Preferably, the use of an environmental anomaly discrimination loss function in constructing and optimizing a tobacco transportation status analysis model can be further refined. First, the model's input data includes standardized environmental parameters collected and preprocessed from an environmental sensor array. These parameters are processed by a multi-stage data processing module to generate feature vectors for status assessment. During model training, the model output is evaluated using the environmental anomaly discrimination loss function by calculating the environmental anomaly factor for each sample and combining its confidence and feature score. The loss function's calculated result reflects the model's classification performance for the current sample, particularly its ability to identify environmental anomalies. Based on the value of the loss function, the model adjusts its internal parameters through a backpropagation algorithm, continuously optimizing the model structure and improving the model's accuracy in identifying environmental anomalies. For example, during training, if the model's confidence in a particular anomaly sample is low, the loss function's value will increase accordingly, prompting the model to adjust its parameters and improve its ability to identify similar anomalies. Through this optimization process based on the environmental anomaly discrimination loss function, the tobacco transportation status analysis model can better adapt to complex transportation environments, accurately identify and warn of environmental anomalies, and provide strong support for tobacco transportation safety and quality assurance.

[0098] In some embodiments, the window size of the first multidimensional filtering processing is 5×5×3, a 7×7×5 three-dimensional convolution kernel is used in the spatiotemporal correlation analysis processing, the expansion interval is 2, a fully connected layer is used in the linear transformation of the aggregation parameters, the window size of the second multidimensional filtering processing is 3×3×2, and the nonlinear activation layer applies the ReLU function.

[0099] It should be noted that the first multidimensional filtering process and spatiotemporal correlation analysis process in the multidimensional parameter analysis unit mentioned in the present invention are key data processing steps. The first multidimensional filtering process is used to extract data features in multiple dimensions, while the spatiotemporal correlation analysis process is used to capture the correlation between data in time and space. These processing steps can effectively enhance the model's ability to perceive data features, thereby improving the accuracy of tobacco transportation status assessment. By properly setting the window size of the filtering process and the convolution kernel parameters of the spatiotemporal correlation analysis, it can better adapt to complex transportation environment data and ensure that the model can accurately identify potential anomalies.

[0100] Specifically, the window size for the first multidimensional filtering process is 5×5×3, meaning that filtering operations are performed using window sizes of 5, 5, and 3 in three dimensions, respectively. This choice of window size effectively extracts feature information within the local region of the data while balancing computational efficiency and feature representation. A 7×7×5 three-dimensional convolution kernel with a dilation interval of 2 is used for spatiotemporal correlation analysis. This setting captures long-range temporal and spatial correlations in the data. The size of the 3D convolution kernel determines the range covered in both temporal and spatial dimensions, while the dilation interval is used to expand the kernel's receptive field, enabling the model to perceive a wider range of spatiotemporal information. Furthermore, a fully connected layer, a common neural network structure used to linearly transform the aggregation parameters, is used to transform the input feature vector into a higher-level feature representation. By learning the linear relationship between input and output features, the fully connected layer effectively integrates feature information from different dimensions. The window size for the second multidimensional filtering process is 3×3×2. This smaller window size is suitable for extracting detailed features of the data, further enriching the model's data perception capabilities. The nonlinear activation layer applies the ReLU function, which is a commonly used activation function that can introduce nonlinear factors, enabling the model to learn complex feature patterns.

[0101] Preferably, when constructing the multidimensional parameter analysis unit, these operational steps can be further refined. For example, in the first multidimensional filtering process, a convolutional layer within a convolutional neural network can be used, with the input data being preprocessed standardized environmental parameters and the output being a filtered feature vector. In the spatiotemporal correlation analysis process, a three-dimensional convolution operation with a dilation interval can be used, with the input being the output feature vector of the first multidimensional filtering process and the output being the feature vector after spatiotemporal correlation analysis. For the linear transformation of the aggregation parameters, a fully connected layer can be constructed, with the input being the feature vector after spatiotemporal correlation analysis and the output being the feature vector after linear transformation. In the second multidimensional filtering process, a convolutional layer can also be used, with the input being the feature vector after linear transformation and the output being the feature vector after the second multidimensional filtering process. Finally, a ReLU function is applied in a nonlinear activation layer, with the input being the feature vector after the second multidimensional filtering process and the output being the feature vector after nonlinear activation. Through these specific construction steps and parameter settings, the multidimensional parameter analysis unit can effectively process the input data and extract feature information with spatiotemporal correlation, providing more accurate data support for subsequent tobacco transportation status assessment.

[0102] In some embodiments, the feature vectors are subjected to dynamic normalization processing and spatiotemporal correlation analysis processing respectively, and the two processing results are superimposed to obtain superimposed features:

[0103]

[0104] Among them, F conv1 represents the result of the first multi-dimensional filtering process, Norm(·) represents dynamic normalization processing, Conv dilated (·) represents dilated convolution processing, Indicates an overlay operation.

[0105] It should be noted that the dynamic normalization and spatiotemporal correlation analysis of feature vectors mentioned in this invention are important steps in the multidimensional parameter analysis unit. Their purpose is to better integrate and optimize data features, thereby improving the model's perception of the transportation environment status. Dynamic normalization can adjust the distribution of data based on real-time changes in the data, making it more consistent with the requirements of model processing; while spatiotemporal correlation analysis can capture the correlation of data in the time and space dimensions, further enhancing the expressive power of features. By combining these two processing methods, the model's adaptability and accuracy to complex transportation environments can be effectively improved.

[0106] Specifically, dynamic normalization is a method that adjusts data distribution based on real-time data changes. It converts the data into standardized data with a mean of zero and a standard deviation of one by calculating its mean and standard deviation. This process removes scale differences in the data, making data of different dimensions comparable and helping to improve model convergence. Spatiotemporal correlation analysis considers the temporal and spatial correlations of data. It convolves the data with a three-dimensional convolution kernel to capture long-range dependencies in both temporal and spatial dimensions. For example, using a 7×7×5 three-dimensional convolution kernel with a dilation interval of 2 expands the kernel's receptive field, enabling the model to perceive a wider range of spatiotemporal information. Feature stacking combines the results of two processing operations to generate a richer feature representation. This operation preserves the details of the original data while introducing spatiotemporal correlations, enhancing the expressive power of features.

[0107] Preferably, when implementing dynamic normalization processing and spatiotemporal correlation analysis processing, the operation steps can be further refined. For example, in dynamic normalization processing, the mean and standard deviation of the input data can be calculated in real time, and the data can be standardized based on these statistics. Specifically, for each input feature vector, its mean and standard deviation are first calculated, and then standardized by subtracting the mean and dividing by the standard deviation. In spatiotemporal correlation analysis processing, a three-dimensional convolution layer can be constructed, with the input being the feature vector after dynamic normalization processing, the convolution kernel size set to 7×7×5, and the dilation interval set to 2. With this setting, the convolution layer can effectively capture the correlation of the data in time and space. The feature superposition operation can be implemented through simple addition or splicing operations, merging the feature vector after dynamic normalization processing with the feature vector after spatiotemporal correlation analysis processing to generate the final fused feature. This fused feature can better reflect the feature information of the data in different dimensions, providing richer data support for subsequent model processing.

[0108] In some embodiments, the convolution kernel size of the first filtering network is 5×5×2, the convolution kernel size of the second filtering network is 5×5×1, and the sliding step size is 1; a 2×2×1 convolution kernel is used in the feature dimensionality reduction processing, and the step size is 1.

[0109] It should be noted that the first and second filtering networks and feature dimensionality reduction processing mentioned in the present invention are key components of the basic data fusion unit and the multidimensional data fusion unit. The first and second filtering networks are used to filter the input data to extract useful features and remove noise. Feature dimensionality reduction is used to reduce the dimensionality of the data while retaining the main feature information to improve data processing efficiency and model performance. By properly setting the convolution kernel size and sliding step size, it can better adapt to the environmental data collected during tobacco transportation, ensuring the accuracy and efficiency of data processing.

[0110] Specifically, the convolution kernel size of the first filtering network is 5×5×2, which means that window sizes of 5, 5, and 2 are used for filtering operations in three dimensions respectively. This size selection can effectively extract the feature information of the data in the local area, while taking into account both computational efficiency and feature expression capabilities. The convolution kernel size of the second filtering network is 5×5×1, and the sliding step size is 1. This setting is suitable for further refining the data and extracting finer features. A 2×2×1 convolution kernel with a step size of 1 is used in the feature dimensionality reduction processing. This setting can effectively reduce the dimension of the data, remove redundant information, and retain the main feature information. Through the reasonable setting of these parameters, it can be ensured that the data can better reflect the characteristics of the tobacco transportation environment after filtering and dimensionality reduction processing, providing high-quality data support for subsequent data fusion and analysis.

[0111] Preferably, when constructing the basic data fusion unit and the multidimensional data fusion unit, these operation steps can be further refined. For example, in the first filtering network, the convolution layer in the convolutional neural network can be used to implement it, and the input data is the pre-processed standardized environmental parameters, and the output is the primary features after filtering. In the second filtering network, the convolution layer is also used, the input is the output data of the previous data processing module, and the output is the feature data after further filtering. In the feature dimensionality reduction processing, a convolution layer can be constructed, the input is the feature data processed by the second filtering network, and a 2×2×1 convolution kernel and a step size of 1 are used, and the output is the feature data after dimensionality reduction. Through these specific construction steps and parameter settings, the basic data fusion unit and the multidimensional data fusion unit can effectively process the input data, extract representative feature information, and provide more accurate data support for the subsequent tobacco transportation status assessment.

[0112] The above-mentioned embodiments of the present invention have the following beneficial effects: the present invention can achieve all-round dynamic monitoring of the tobacco transportation environment through the collaborative work of the environmental sensor array and the multi-level data processing module. The layered processing architecture of the basic data fusion unit and the multi-dimensional data fusion unit can construct a multi-scale feature expression. In conjunction with the environmental parameter attention mechanism and data enhancement module in the multi-dimensional parameter analysis unit, it can effectively capture the spatiotemporal correlation characteristics of the transportation process and improve the sensitivity of anomaly detection. Through the cascade processing of the three data processing modules, the transportation environment characteristics can be gradually extracted and optimized. Finally, through the parameter integration and decision mapping of the state analysis module, an accurate transportation status assessment result can be output.

[0113] The system optimizes model performance through a specific loss function for identifying environmental anomalies. Combined with processing techniques such as dynamic normalization and multidimensional filtering, it enhances the model's adaptability to complex transportation environments. Convolution kernels of varying sizes extract multi-granularity features, while feature dimensionality reduction and upscaling operations balance computational efficiency and expressiveness. The introduction of frequency factors and confidence weighting mechanisms mitigates data imbalances and improves the reliability of assessment results. The overall modular architecture allows for flexible configuration adjustments to meet monitoring needs in diverse transportation scenarios.

[0114] like Figure 2 As shown, some embodiments provide a method for real-time monitoring and management of tobacco leaf transportation, the method comprising:

[0115] Step 1: Collect transportation environment data and preprocess them to obtain standardized environmental parameters;

[0116] Step 2: Using the basic data fusion unit to perform feature fusion on the standardized environmental parameters to obtain the first output parameter;

[0117] Step 3: Input the first output parameter into a plurality of sequentially connected multidimensional parameter analysis units to obtain a second output parameter as input data for a multidimensional data fusion unit; and capture the spatiotemporal correlation characteristics of the transportation environment through the multidimensional perception capabilities of the multidimensional parameter analysis units;

[0118] Step 4: Using the multi-dimensional data fusion unit to perform multi-scale feature fusion processing on the input data of the unit to obtain the third output parameter;

[0119] Step 5: Inputting the third output parameter into a plurality of sequentially connected multidimensional parameter analysis units to obtain a fourth output parameter, and inputting the fourth output parameter into a multidimensional data fusion unit;

[0120] Step 6: Repeat steps 4 to 5 until a preset number of iterations is reached, and output the fifth output parameter;

[0121] Step 7: Input the fifth output parameter into the status analysis module to obtain a transportation environment status assessment result.

[0122] It is understandable that the steps and references in the tobacco transportation real-time monitoring and management method are Figure 1 The modules, features and beneficial effects described above for the tobacco transport real-time monitoring and management system also apply to the tobacco transport real-time monitoring and management method and the operations contained therein, and will not be repeated here.

[0123] Reference below Figure 3 , which shows a schematic structural diagram of an electronic device 300 suitable for implementing some embodiments of the present invention. The electronic devices in some embodiments of the present invention may include, but are not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 3 The terminal device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0124] like Figure 3 As shown, the electronic device 300 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage device 308 into a random access memory (RAM) 303. Various programs and data required for the operation of the electronic device 300 are also stored in the RAM 303. The processing device 301, the ROM 302, and the RAM 303 are connected to each other via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0125] Typically, the following devices may be connected to the I / O interface 305: an input device 306 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 307 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 308 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 309. The communication device 309 may allow the electronic device 300 to communicate with other devices wirelessly or by wire to exchange data. Figure 3 The electronic device 300 is shown with various devices, but it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed instead. Figure 3Each block shown in the figure may represent one device, or may represent multiple devices as needed.

[0126] Furthermore, the storage medium of the embodiment of the present application stores program instructions that can implement all the above methods, wherein the program instructions can be stored in the above storage medium in the form of a software product, including a number of instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) or a processor to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, or a terminal device such as a computer, a server, a mobile phone, or a tablet.

[0127] The above descriptions are merely some preferred embodiments of the present invention and an illustration of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present invention is not limited to the technical solutions formed by a specific combination of the above-mentioned technical features, but should also encompass other technical solutions formed by any combination of the above-mentioned technical features or their equivalents without departing from the above-mentioned inventive concept. For example, a technical solution formed by mutually replacing the above-mentioned features with (but not limited to) technical features having similar functions disclosed in the embodiments of the present invention.

Claims

1. A tobacco leaf transportation real-time monitoring and management system, characterized in that: include: an environmental sensor array configured to collect transportation environment data and output standardized environmental parameters; The tobacco transport status analysis model is composed of three sequentially connected data processing modules, each of which includes a sequentially connected data fusion unit and at least two multidimensional parameter analysis units, wherein the data fusion unit of the first data processing module receives the standardized environmental parameters; The data fusion unit is configured to perform feature fusion processing on the input data and output the result to the corresponding multi-dimensional parameter analysis unit; The multidimensional parameter analysis unit includes an environmental parameter attention module and a data enhancement module, wherein: the environmental parameter attention module is configured to generate a feature vector through multidimensional filtering, generate superposition features after performing dynamic normalization and spatiotemporal correlation analysis on the feature vector, generate aggregation parameters based on extreme value sampling and mean sampling, and generate weighted feature data through weight allocation; The data enhancement module is configured to sequentially perform feature dimensionality reduction, multi-dimensional filtering, nonlinear activation, noise suppression, and feature dimensionality increase operations on the input data to generate enhanced feature data; The output of the multi-dimensional parameter analysis unit is determined by the weighted feature data and the enhanced feature data and serves as the input of the next processing unit; The state analysis module is connected to the terminal data processing module and is configured to perform feature integration and decision mapping on the output data of the tobacco transportation state analysis model to generate a transportation state evaluation result.

2. A tobacco transportation real-time monitoring and management system according to claim 1, characterized in that: The data fusion unit in the first data processing module is a basic data fusion unit, and the data fusion units in other data processing modules are multi-dimensional data fusion units; The basic data fusion unit includes using a first filter network to process the standardized environmental parameters to obtain primary features, and performing data standardization processing on the primary features to obtain output data of the basic data fusion unit; The multidimensional data fusion unit includes performing a second filtering network and feature dimensionality reduction processing on the output data of the previous data processing module to obtain a first processing result and a second processing result; performing channel calibration processing on the first processing result to obtain a third processing result; and inputting the third processing result and the second processing result into the feature splicing module to obtain the output data of the multidimensional data fusion unit.

3. The real-time monitoring and management system for tobacco transportation according to claim 1, characterized in that: The multidimensional parameter analysis unit further includes a first data normalization layer, a first feedback network, a second data normalization layer, a second feedback network and a feature normalization layer; Processing the first input data input to the multidimensional parameter analysis unit using the first data normalization layer to obtain second input data; The environmental parameter attention module processes the second input data to obtain weighted feature data; Adding the weighted feature data to the first input data through the first feedback network, and adding the addition result to the second data normalization layer; Processing the input data of the layer using the second data normalization layer to obtain third input data; The data enhancement module processes the third input data to obtain enhanced feature data; The enhanced feature data and the addition result are added to the feature normalization layer through the second feedback network to obtain the output data of the multidimensional parameter analysis unit.

4. The tobacco leaf transportation real-time monitoring and management system according to claim 1, characterized in that: The tobacco transport status analysis model includes three sequentially connected data processing modules, the second data processing module is provided with three sequentially connected multidimensional parameter analysis units, and the remaining data processing modules are provided with two sequentially connected multidimensional parameter analysis units.

5. The tobacco leaf transportation real-time monitoring and management system according to claim 1, characterized in that: The state analysis module includes a parameter integration layer and a decision mapping layer, and the parameter integration layer is used to perform 1×1 parameter aggregation on the input data of the state analysis module in the feature dimension; The decision mapping layer is used to perform nonlinear mapping on the output of the parameter integration layer to obtain the transportation status assessment result.

6. The real-time monitoring and management system for tobacco transportation according to claim 1, characterized in that: The tobacco transport state analysis model is optimized using the constructed environmental anomaly discrimination loss function. The expression of the environmental anomaly discrimination loss function is: Among them, ω i represents the environmental anomaly factor of the i-th sample, Indicates that the model correctly judges sample i as the actual state y i Confidence, s i,j represents the feature score of the model predicting sample i as state j, C represents the number of transportation environment state categories, α j represents the frequency factor of state category j, n j represents the number of samples in category j, It represents the number of samples in the category to which sample i belongs, μ and σ represent the distribution parameters of environmental parameters.

7. The tobacco leaf transportation real-time monitoring and management system according to claim 1, characterized in that: The window size of the first multidimensional filtering processing is 5×5×3, a 7×7×5 three-dimensional convolution kernel is used in the spatiotemporal correlation analysis processing, the expansion interval is 2, and a fully connected layer is used in the linear transformation of the aggregation parameters. The window size of the second multidimensional filtering processing is 3×3×2, and the nonlinear activation layer applies the ReLU function.

8. The tobacco leaf transportation real-time monitoring and management system according to claim 7, characterized in that: The feature vectors are subjected to dynamic normalization processing and spatiotemporal correlation analysis processing respectively, and the two processing results are superimposed to obtain superimposed features: Among them, F fusion represents the superposition feature, F conv1 represents the result of the first multi-dimensional filtering process, Norm(·) represents dynamic normalization processing, Conv dilated (·) represents dilated convolution processing, Indicates an overlay operation.

9. The tobacco leaf transportation real-time monitoring and management system according to claim 2, characterized in that: The convolution kernel size of the first filter network is 5×5×2, the convolution kernel size of the second filter network is 5×5×1, and the sliding step size is 1; The feature dimensionality reduction process uses a 2×2×1 convolution kernel with a step size of 1.

10. A tobacco leaf transportation real-time monitoring method, which realizes the tobacco leaf transportation real-time monitoring and management system according to any one of claims 1 to 9, characterized in that: include: Step 1: Collect transportation environment data and preprocess them to obtain standardized environmental parameters; Step 2: Using the basic data fusion unit to perform feature fusion on the standardized environmental parameters to obtain the first output parameter; Step 3: Input the first output parameter into a plurality of sequentially connected multidimensional parameter analysis units to obtain a second output parameter as input data for a multidimensional data fusion unit; The multi-dimensional perception capability of the multi-dimensional parameter analysis unit is used to capture the spatiotemporal correlation characteristics of the transportation environment; Step 4: Using the multi-dimensional data fusion unit to perform multi-scale feature fusion processing on the input data of the unit to obtain the third output parameter; Step 5: Inputting the third output parameter into a plurality of sequentially connected multidimensional parameter analysis units to obtain a fourth output parameter, and inputting the fourth output parameter into a multidimensional data fusion unit; Step 6: Repeat steps 4 to 5 until a preset number of iterations is reached, and output the fifth output parameter; Step 7: Input the fifth output parameter into the status analysis module to obtain a transportation environment status assessment result.

Images