A vaccine cold chain transportation method with improved neural network

Abstract

The present invention discloses a vaccine cold chain transportation method with an improved neural network, which relates to the technical field of cross-integration of logistics management, including: using hyperspectral imaging and machine vision technology to perceive the environment around a cold chain logistics box to form environmental perception data; extracting key environmental features from the environmental perception data, and identifying obstacles based on the environmental features; using a deep learning model SSD to build a classification model, and performing object detection and classification on obstacles and terrain features; using a path planning algorithm to plan the driving path of the cold chain logistics box to obtain an obstacle distribution data set; combining a neural network to build a transportation obstacle avoidance model, inputting the obstacle distribution data set into the transportation obstacle avoidance model to optimize path selection, and outputting an optimal path; during driving, using the results of the real-time optimal path and the real-time obstacle distribution data set to avoid obstacles.

Description

Technical Field

[0001] The present invention relates to the technical field of cross-integration of logistics management, and in particular to a vaccine cold chain transportation method with an improved neural network. Background Art

[0002] With the growing global public health needs, vaccines, as a key means of preventing and controlling infectious diseases, have become an indispensable part of the public health system in their safe and effective transportation. Vaccine cold chain transportation technology aims to ensure that vaccines maintain a constant low temperature from production to vaccination sites to maintain the biological activity and safety of vaccines. In recent years, with the rapid development of the Internet of Things, big data, and artificial intelligence technologies, vaccine cold chain transportation systems are gradually transforming towards intelligence and refinement to achieve more efficient and accurate temperature control management and real-time monitoring.

[0003] However, current vaccine cold chain transportation technology still faces many challenges, especially in complex and changeable transportation environments. On the one hand, traditional cold chain transportation monitoring systems mostly rely on GPS positioning and temperature sensors. They do not respond promptly to sudden obstacles during transportation, including traffic jams and sections affected by natural disasters, and lack effective dynamic path planning and adjustment mechanisms. On the other hand, the real-time perception and analysis capabilities of environmental factors are limited, making it difficult to accurately assess the potential threats of the surrounding environment to vaccine storage conditions, such as the impact of changes in light intensity on the temperature inside cold chain logistics boxes. In addition, the shortcomings of existing technologies in intelligent decision-making also limit transportation efficiency and safety, especially when facing complex terrain and densely populated urban areas. Relying solely on preset routes often cannot effectively respond to emergencies. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a vaccine cold chain transportation method with an improved neural network to solve the problem that most traditional cold chain transportation monitoring systems rely on GPS positioning and temperature sensors, and do not respond promptly to sudden obstacles during transportation, including traffic congestion and sections affected by natural disasters, and lack effective dynamic path planning and adjustment mechanisms. On the other hand, the real-time perception and analysis capabilities of environmental factors are limited, and it is difficult to accurately assess the potential threats of the surrounding environment to vaccine storage conditions, such as the impact of changes in light intensity on the temperature in cold chain logistics boxes. In addition, the deficiencies of existing technologies at the intelligent decision-making level also limit transportation efficiency and safety, especially when facing complex terrain and densely populated urban areas, relying solely on preset routes often cannot effectively respond to emergencies.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a vaccine cold chain transportation method using an improved neural network, which includes using hyperspectral imaging and machine vision technology to perceive the environment around the cold chain logistics box to form environmental perception data;

[0008] Extract key environmental features from environmental perception data and identify obstacles based on the environmental features;

[0009] Use the deep learning model SSD to build a classification model to detect and classify obstacles and terrain features;

[0010] Use the path planning algorithm to plan the driving path of the cold chain logistics box and obtain the obstacle distribution data set;

[0011] Combined with neural network to build a transportation obstacle avoidance model, the obstacle distribution data set is input into the transportation obstacle avoidance model to optimize path selection and output the optimal path;

[0012] During driving, obstacle avoidance is performed using the results of the real-time optimal path and real-time obstacle distribution dataset.

[0013] As a preferred solution of the vaccine cold chain transportation method with improved neural network described in the present invention, wherein: the hyperspectral imaging and machine vision technology are used to perceive the environment around the cold chain logistics box to form environmental perception data, and the specific steps are as follows:

[0014] Hyperspectral imaging sensors are installed on the top and sides of the cold chain logistics box to collect hyperspectral data in multiple bands and obtain environmental perception data;

[0015] The environmental perception data, the environmental perception data expression is:

[0016]

[0017] R spec =[R 1 , R 2 , ..., R n "

[0018] Where D represents the environmental perception data, R spec Represents the spectral reflectance array of different bands, R n represents the spectral reflectance of the nth band, V index represents vegetation index, T is ambient temperature, H is ambient humidity, and W is ambient wind speed;

[0019] The spectral reflectance R in D spec Perform radiation correction to eliminate system errors;

[0020] The radiation correction expression is:

[0021]

[0022] Among them, R cal Represents the corrected spectral reflectance array, R 1 , R 2 , ..., R n Represents the original spectral reflectance, R dark，1 , R dark，2 , ..., R dark，n represents the corresponding dark current, R ref，1 , R ref，2 , ..., R ref，n Indicates the corresponding reference spectral radiation value.

[0023] As a preferred solution of the vaccine cold chain transportation method of the improved neural network of the present invention, wherein: the key environmental features are extracted from the environmental perception data, and the specific steps are:

[0024] Set the environmental perception data after radiation correction to D l , the expression is:

[0025]

[0026] Using standardized methods, D l Processing, the standardized expression is:

[0027]

[0028] Among them, D P represents the environmental perception data after standardization, μ represents the average value of each band data value, and σ represents the standard deviation of each band data value;

[0029] Use principal component analysis (PCA) to analyze D P Analyze and get the covariance matrix, which is expressed as:

[0030]

[0031] Among them, ∑ represents the covariance matrix, N represents the number of samples, Indicates D P The transposed matrix of

[0032] The covariance matrix ∑ is subjected to eigendecomposition, and the expression is:

[0033] ∑v=λv

[0034] Among them, v represents the eigenvector, λ represents the eigenvalue;

[0035] The eigenvalues ​​are calculated according to λ 1 ,λ2 , ..., λ p Sort from largest to smallest, where p is the total number of eigenvalues;

[0036] Select the largest k eigenvalues ​​after sorting, denoted as λ 1 ,λ 2 , ..., λ k ;

[0037] Finally, select the eigenvector v corresponding to these k eigenvalues 1 , v 2 , ..., v k , these eigenvectors are combined to obtain the eigenvector matrix F, which is expressed as:

[0038] F=D P v 1 , D P v 2 , ..., D P v k

[0039] Among them, F is the eigenvector matrix, representing the environmental characteristics, v 1 , v 2 , ..., v k Represents the eigenvectors corresponding to the k eigenvalues.

[0040] As a preferred solution of the vaccine cold chain transportation method with improved neural network of the present invention, the specific steps of identifying obstacles based on environmental characteristics are as follows:

[0041] A hyperspectral imaging sensor installed on the top of the cold chain logistics box is used to obtain a forward-looking image.

[0042] The obstacles in the front view image are identified by using the extracted environmental features F(x, y) combined with machine vision technology;

[0043] The Canny edge detection algorithm is used to detect the edges in the image. The expression is:

[0044] E'(x, y)=α·F(x, y)+(1-α)·E(x, y)

[0045] Wherein, E′(x, y) represents the edge intensity image after environmental feature adjustment, E(x, y) represents the original edge intensity image, and α is a weight factor used to adjust the influence of environmental feature F(x, y) on edge intensity image E(x, y);

[0046] Further analysis is performed on the detected edges, including calculating the number of edges per unit area of ​​the detected edges and setting a density threshold P, where the value range of P is 0 to 1;

[0047] The P includes a low-density interval, a medium-density interval and a high-density interval;

[0048] When the number of edges per unit area is less than 0.2, it means that there are no obstacles and micro obstacles in the area;

[0049] When 0.2≤the number of edges per unit area<0.5, it means that there are small or medium-sized obstacles in the area;

[0050] When the number of edges per unit area is ≥ 0.5, it indicates that there are large and dense obstacles in the area;

[0051] Support vector machine (SVM) is used to classify obstacles. The classification expression is:

[0052] cf = SVM(F, Y)

[0053] Among them, cf is the classification result, indicating the category assigned to each sample, and Y represents the category label.

[0054] As a preferred solution of the vaccine cold chain transportation method of the improved neural network of the present invention, wherein: the classification model is constructed by using the deep learning model SSD to detect and classify obstacles and terrain features, and the specific steps are:

[0055] Build a classification model using the deep learning framework TensorFlow and the SSD model architecture;

[0056] The input image size is adjusted to a standard size of 300x300 pixels suitable for the classification model, so that images from different sources can be processed uniformly;

[0057] Use ResNet as the basic network structure;

[0058] Construct multi-scale feature maps for detecting objects of different sizes;

[0059] Design prediction boxes of different scales for each feature map, which are used for object detection and classification;

[0060] The classification model includes an input layer, a basic network structure, a multi-scale feature map generation layer, a prediction box generation layer, a classification layer, a regression layer and an output layer;

[0061] Input the classification result cf into the constructed classification model for training and obtain the prediction result;

[0062] The cross entropy loss function is used to measure the difference between the model prediction result and the true value. The expression is:

[0063] loss = CELoss(y, y)

[0064] Among them, loss is the loss value, which indicates the difference between the predicted result and the true value, CELoss represents the cross entropy loss function, and y represents the true category label. Represents the predicted category label;

[0065] The SGD optimizer is used to tune the parameters of the classification model. The tuning expression is:

[0066] θ = SGD(θ, loss)

[0067] Among them, θ represents the weight and bias of the classification model;

[0068] Deploy the trained classification model to computing devices on cold chain logistics boxes for real-time object detection and classification;

[0069] The forward propagation process of the SSD model is used to perform object detection and classification on the front-view images collected in real time.

[0070] As a preferred solution of the vaccine cold chain transportation method of the improved neural network of the present invention, wherein: the use of the path planning algorithm to plan the driving path of the cold chain logistics box to obtain the obstacle distribution data set includes setting the search space and identifying and marking the positions of obstacles in the environment;

[0071] The setting of the search space includes inputting the starting and ending positions of the cold chain logistics box, using machine vision technology to obtain an image of the current position of the cold chain logistics box, and determining the current position of the cold chain logistics box through a positioning algorithm;

[0072] Identify and mark the location of obstacles in the environment, including using the Canny edge detection algorithm to detect edges in the image, and further analyze the detected edges to determine the material, size and location of the obstacles.

[0073] As a preferred solution of the vaccine cold chain transportation method of the improved neural network of the present invention, wherein: the transportation obstacle avoidance model is constructed by combining the neural network, the obstacle distribution data set is input into the transportation obstacle avoidance model to optimize the path selection, and the factors of energy consumption and time efficiency are considered to output the optimal path. The specific steps are:

[0074] Use the deep learning framework TensorFlow and the convolutional neural network (CNN) architecture to build a transportation obstacle avoidance model, and set the model architecture, including the input layer, convolution layer, pooling layer, fully connected layer, and output layer;

[0075] Set the obstacle distribution dataset to B;

[0076] Input B into the transportation obstacle avoidance model to complete the training of the transportation obstacle avoidance model and obtain the predicted optimal path.

[0077] As a preferred solution of the vaccine cold chain transportation method with improved neural network of the present invention, in which: during the driving process, the results of the real-time optimal path and the real-time obstacle distribution data set are used to avoid obstacles, and the specific steps are as follows:

[0078] Real-time collection of hyperspectral data and machine vision data in front of the cold chain logistics box to form real-time environmental perception data;

[0079] Use machine vision technology to identify obstacles and extract key real-time environmental features from real-time environmental perception data;

[0080] Input the real-time environment features into the classification model for classification;

[0081] The path planning algorithm is used to calculate the classified real-time environmental features to obtain a real-time obstacle distribution data set;

[0082] The real-time obstacle distribution dataset is input into the transportation obstacle avoidance model to obtain the real-time optimal path.

[0083] In a second aspect, an embodiment of the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the vaccine cold chain transportation method with an improved neural network as described in the first aspect of the present invention is implemented.

[0084] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: the computer program, when executed by a processor, implements any step of the vaccine cold chain transportation method with an improved neural network as described in the first aspect of the present invention.

[0085] The beneficial effects of the present invention are: by integrating hyperspectral imaging, machine vision, deep learning and path planning algorithms, a more intelligent and flexible vaccine transportation management system is constructed. The invention can not only perceive and analyze the environmental characteristics around the cold chain logistics box in real time, identify and avoid obstacles, but also realize accurate object detection and classification through the deep learning model SSD, further improving the accuracy and adaptability of path planning. More importantly, the present invention innovatively applies neural networks to the transportation obstacle avoidance model, and dynamically optimizes the path selection by inputting the obstacle distribution data set to ensure the continuity and safety of the vaccine transportation process. This technological breakthrough not only solves the limitations of the existing technology in real-time response, environmental adaptability and intelligent decision-making, but also promotes the field of vaccine cold chain transportation to a higher level of intelligence. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0087] Figure 1 This is a flow chart of the vaccine cold chain transportation method with improved neural network in Example 1. DETAILED DESCRIPTION

[0088] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0089] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0090] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0091] Example 1, reference Figure 1 , which is the first embodiment of the present invention, and provides a vaccine cold chain transportation method with an improved neural network, comprising the following steps:

[0092] Use hyperspectral imaging and machine vision technology to perceive the environment around the cold chain logistics box to form environmental perception data;

[0093] Extract key environmental features from environmental perception data and identify obstacles based on the environmental features;

[0094] Use the deep learning model SSD to build a classification model to detect and classify obstacles and terrain features;

[0095] Use the path planning algorithm to plan the driving path of the cold chain logistics box and obtain the obstacle distribution data set;

[0096] Combined with neural network to build a transportation obstacle avoidance model, the obstacle distribution data set is input into the transportation obstacle avoidance model to optimize path selection and output the optimal path;

[0097] During driving, obstacle avoidance is performed using the results of the real-time optimal path and real-time obstacle distribution dataset.

[0098] The hyperspectral imaging and machine vision technology are used to sense the environment around the cold chain logistics box to form environmental perception data. The specific steps are as follows:

[0099] Hyperspectral imaging sensors are installed on the top and sides of the cold chain logistics box to collect hyperspectral data in multiple bands and obtain environmental perception data;

[0100] The environmental perception data includes road reflectivity, vegetation index, temperature, humidity and wind speed. The environmental perception data expression is:

[0101]

[0102] R spec =[R 1 , R 2 , …, R n "

[0103] Where D represents the environmental perception data, R spec Represents the spectral reflectance array of different bands, R n represents the spectral reflectance of the nth band, V index represents vegetation index, T is ambient temperature, H is ambient humidity, and W is ambient wind speed;

[0104] The spectral reflectance R in D spec Perform radiation correction to eliminate system errors;

[0105] The radiation correction expression is:

[0106]

[0107] Among them, R cal Represents the corrected spectral reflectance array, R 1 , R 2 , ..., R n Represents the original spectral reflectance, R dark，1 , R dark，2 , ..., R dark，n represents the corresponding dark current, R ref，1 , R ref，2 , ..., R ref，n Indicates the corresponding reference spectral radiation value;

[0108] The dark current refers to the current generated inside the sensor when there is no external light source;

[0109] The reference spectral radiation value R ref, which means using a known spectral radiation value as a reference, is a standard light source.

[0110] The specific steps of extracting key environmental features from environmental perception data are as follows:

[0111] Set the environmental perception data after radiation correction to D l , the expression is:

[0112] D l =[R cal , V index , T, H, W]

[0113] Using standardized methods, D l The data value of each band is subtracted from the mean value of the band, and then divided by the square root of the variance of the band, so that the data value of each band has zero mean and unit variance. The standardized expression is:

[0114]

[0115] Among them, D P represents the environmental perception data after standardization, μ represents the average value of each band data value, and σ represents the standard deviation of each band data value;

[0116] Use principal component analysis (PCA) to analyze D P Analyze and get the covariance matrix, which is expressed as:

[0117]

[0118] Among them, ∑ represents the covariance matrix, N represents the number of samples, Indicates D P The transposed matrix of

[0119] The covariance matrix ∑ is subjected to eigendecomposition, and the expression is:

[0120] ∑v=λv

[0121] Among them, v represents the eigenvector, λ represents the eigenvalue;

[0122] The eigenvalues ​​are calculated according to λ 1 ,λ 2 , ..., λ p Sort from largest to smallest, where p is the total number of eigenvalues;

[0123] Select the largest k eigenvalues ​​after sorting, denoted as λ 1 ,λ 2 , ..., λ k ;

[0124] Finally, select the eigenvector v corresponding to these k eigenvalues 1 , v 2 , ..., v k , these eigenvectors are combined to obtain the eigenvector matrix F, which is expressed as:

[0125] F=D P v 1 , D P v 2 , ..., D P v k

[0126] Among them, F is the eigenvector matrix, representing the environmental characteristics, v 1 , v 2 , ..., v k Represents the eigenvectors corresponding to the k eigenvalues.

[0127] The specific steps of identifying obstacles based on environmental features are as follows:

[0128] A hyperspectral imaging sensor installed on the top of the cold chain logistics box is used to obtain a forward-looking image.

[0129] The obstacles in the front view image are identified by using the extracted environmental features F(x, y) combined with machine vision technology;

[0130] The Canny edge detection algorithm is used to detect the edges in the image, and the environmental features are combined to enhance the recognition of obstacles. The edge detection algorithm can detect the brightness changes in the image, and the environmental features can help distinguish different types of obstacles. The expression is:

[0131] E'(x, y)=α·F(x, y)+(1-α)·E(x, y)

[0132] Wherein, E′(x, y) represents the edge intensity image after environmental feature adjustment, E(x, y) represents the original edge intensity image, and α is a weight factor used to adjust the influence of environmental feature F(x, y) on edge intensity image E(x, y);

[0133] Further analysis is performed on the detected edges, including calculating the number of edges per unit area of ​​the detected edges and setting a density threshold P, where the value range of P is 0 to 1;

[0134] The P includes a low-density interval, a medium-density interval and a high-density interval;

[0135] When the number of edges per unit area is less than 0.2, it means that there are no obstacles and micro obstacles in the area;

[0136] When 0.2≤the number of edges per unit area<0.5, it means that there are small or medium-sized obstacles in the area;

[0137] When the number of edges per unit area is ≥ 0.5, it indicates that there are large and dense obstacles in the area;

[0138] Support vector machine (SVM) is used to classify obstacles. The classification expression is:

[0139] cf = SVM(F, Y)

[0140] in, c f is the classification result, indicating the category assigned to each sample, and Y represents the category label;

[0141] The category labels are carefully marked by professionals based on environmental characteristics, including low temperature areas, normal temperature areas, high temperature areas, obstacle categories, flat areas, slopes, and uneven areas;

[0142] The obstacle category, including obstacle material, size and location;

[0143] The obstacle materials include metal, plastic and wood.

[0144] The deep learning model SSD is used to build a classification model to detect and classify obstacles and terrain features. The specific steps are as follows:

[0145] Build a classification model using the deep learning framework TensorFlow and the SSD model architecture;

[0146] The input image size is adjusted to a standard size of 300x300 pixels suitable for the classification model, so that images from different sources can be processed uniformly;

[0147] Use ResNet as the basic network structure;

[0148] Construct multi-scale feature maps for detecting objects of different sizes;

[0149] Design prediction boxes of different scales for each feature map, which are used for object detection and classification;

[0150] The classification model includes an input layer, a basic network structure, a multi-scale feature map generation layer, a prediction box generation layer, a classification layer, a regression layer and an output layer;

[0151] The input layer is used to receive the resized image data, the size of which is 300x300 pixels;

[0152] The basic network structure is used to extract deep features of the image, including multiple convolutional layers, pooling layers and activation layers;

[0153] The multi-scale feature map generation layer includes adding multiple convolutional layers on the basis of ResNet to detect objects of different sizes, and each convolutional layer is responsible for generating a 300x300 pixel feature map;

[0154] The prediction box generation layer is a preset bounding box generated based on the position and scale of each feature map;

[0155] The classification layer, for each prediction box, is responsible for predicting the category of the object it contains;

[0156] The regression layer, for each predicted box, is responsible for predicting the exact position of the bounding box;

[0157] The output layer is used to output the category probability and bounding box coordinate offset of each prediction box;

[0158] Input the classification result cf into the constructed classification model for training and obtain the prediction result;

[0159] The cross entropy loss function is used to measure the difference between the model prediction result and the true value. The expression is:

[0160] loss = CELoss(y, y)

[0161] Among them, loss is the loss value, which indicates the difference between the predicted result and the true value, CELoss represents the cross entropy loss function, and y represents the true category label. Represents the predicted category label;

[0162] The SGD optimizer is used to tune the parameters of the classification model. The tuning expression is:

[0163] θ = SGD(θ, loss)

[0164] Among them, θ represents the weight and bias of the classification model;

[0165] Deploy the trained classification model to computing devices on cold chain logistics boxes for real-time object detection and classification;

[0166] The forward propagation process of the SSD model is used to perform object detection and classification on the front-view images collected in real time.

[0167] The method of using a path planning algorithm to plan the driving path of the cold chain logistics box and obtain obstacle distribution data includes setting a search space and identifying and marking the locations of obstacles in the environment;

[0168] The setting of the search space includes inputting the starting and ending positions of the cold chain logistics box, using machine vision technology to obtain an image of the current position of the cold chain logistics box, and determining the current position of the cold chain logistics box through a positioning algorithm;

[0169] Identify and mark the location of obstacles in the environment, including using the Canny edge detection algorithm to detect edges in the image, and further analyze the detected edges to determine the material, size and location of the obstacles.

[0170] The transportation obstacle avoidance model is constructed by combining a neural network, and the obstacle distribution data set is input into the transportation obstacle avoidance model to optimize the path selection, consider the factors of energy consumption and time efficiency, and output the optimal path. The specific steps are as follows:

[0171] Use the deep learning framework TensorFlow and the convolutional neural network (CNN) architecture to build a transportation obstacle avoidance model, and set the model architecture, including the input layer, convolution layer, pooling layer, fully connected layer, and output layer;

[0172] Set the obstacle distribution dataset to B;

[0173] Input B into the transportation obstacle avoidance model to complete the training of the transportation obstacle avoidance model and obtain the predicted optimal path.

[0174] During the driving process, real-time hyperspectral data and machine vision analysis are used to avoid obstacles. The specific steps are as follows:

[0175] Real-time collection of hyperspectral data and machine vision data in front of the cold chain logistics box to form real-time environmental perception data;

[0176] Use machine vision technology to identify obstacles and extract key real-time environmental features from real-time environmental perception data;

[0177] Input the real-time environment features into the classification model for classification;

[0178] The path planning algorithm is used to calculate the classified real-time environmental features to obtain a real-time obstacle distribution data set;

[0179] Input the real-time obstacle distribution data set into the transportation obstacle avoidance model to obtain the real-time optimal path;

[0180] The obstacle avoidance strategy for the density threshold P and obstacle characteristics is:

[0181] When the density threshold P is in the low-density interval, it means that there are no obstacles or micro-obstacles in the area, and the cold chain logistics box can travel at a normal speed without special attention;

[0182] When the density threshold P is in the medium density range, it means that there are small and medium obstacles in the area, and the cold chain logistics box needs to reduce speed and remain highly vigilant to the environment;

[0183] When the density threshold P is in the high-density interval, it means that there are large and dense obstacles in the area. The cold chain logistics box needs to slow down immediately and send an alarm to the driver to inform him to drive carefully.

[0184] The obstacle avoidance strategy based on the material, size, location and characteristics of the obstacle is:

[0185] When the obstacle is made of metal: the cold chain logistics box needs to be kept at a greater distance to avoid direct contact;

[0186] When the obstacle is made of plastic: the cold chain logistics box can travel at a lower speed and remain vigilant;

[0187] When the obstacle is made of wood: the cold chain logistics box can travel at a lower speed and remain alert;

[0188] When the obstacle is a large one: the cold chain logistics box needs to slow down immediately and send an alarm to the driver to inform him to drive carefully;

[0189] When the obstacle is a medium-sized obstacle: the cold chain logistics box needs to reduce speed and remain alert;

[0190] When the obstacle is a small one: the cold chain logistics box can travel at normal speed without paying special attention;

[0191] When an obstacle appears in the middle of the road: the cold chain logistics box needs to slow down immediately and advise the driver to detour on both sides;

[0192] When obstacles appear in the roadside area: the cold chain logistics box can travel at a lower speed and convey suggestions for intermediate driving to the driver.

[0193] This embodiment also provides a computer device suitable for improving the cold chain transportation method of vaccines using a neural network, comprising: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the cold chain transportation method of vaccines using an improved neural network as proposed in the above embodiment.

[0194] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0195] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by the processor, the vaccine cold chain transportation method for realizing an improved neural network as proposed in the above embodiment is implemented; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic storage, flash memory, disk or optical disk.

[0196] In summary, the present invention integrates hyperspectral imaging, machine vision, deep learning and path planning algorithms to build a more intelligent and flexible vaccine transportation management system. The invention can not only perceive and analyze the environmental characteristics around the cold chain logistics box in real time, identify and avoid obstacles, but also realize accurate object detection and classification through the deep learning model SSD, further improving the accuracy and adaptability of path planning. More importantly, the present invention innovatively applies neural networks to the transportation obstacle avoidance model, dynamically optimizes path selection by inputting obstacle distribution data sets, and ensures the continuity and safety of vaccine transportation. This technological breakthrough not only solves the limitations of existing technologies in real-time response, environmental adaptability and intelligent decision-making, but also promotes the field of vaccine cold chain transportation to a higher level of intelligence.

[0197] Example 2

[0198] Referring to Table 1, which is the second embodiment of the present invention, in order to further verify the advancement of the present invention, experimental simulation data of the vaccine cold chain transportation method based on the improved neural network are provided.

[0199] The experiment was conducted in a complex suburban environment, which contains various types of obstacles (such as trees, vehicles, pedestrians) and different terrain features (flat ground, slopes). The experiment selected 5 paths from different starting points to the end points, each path was about 20 kilometers long, to ensure that various possible obstacles were covered.

[0200] Implementation steps

[0201] Hyperspectral imaging sensors and conventional cameras are installed on the top and sides of the cold chain logistics box. The collected hyperspectral data are combined with machine vision images after radiation correction to form a comprehensive environmental perception dataset. Key environmental features are extracted through standardization and principal component analysis (PCA).

[0202] Obstacles were identified and classified using Canny edge detection and support vector machine (SVM). Three main types of obstacles were identified in the experiment: small static obstacles (such as trash cans), medium-sized mobile obstacles (such as pedestrians) and large fixed obstacles (such as buildings). Through feature adjustment, the recognition accuracy was improved and the false alarm rate was reduced to 5%.

[0203] An object detection and classification model based on the deep learning model SSD was constructed, and a transportation obstacle avoidance model based on the convolutional neural network (CNN) was combined. The A* algorithm was used as the basic path planning algorithm, combined with the neural network to optimize the path selection, ensuring that the path avoids obstacles while minimizing energy consumption and transportation time.

[0204] During driving, real-time hyperspectral data and machine vision information are used to update environmental perception data, instantly identify new obstacles through classification models, and adjust the driving path in real time through the transport obstacle avoidance model, ensuring rapid response even in unknown or rapidly changing environments.

[0205] The details are shown in the following table:

[0206] Table 1 Experimental record table

[0207]

[0208] It can be seen from the above experimental data that the improved neural network vaccine cold chain transportation method of the present invention shows significant advantages in practical applications. First, in terms of obstacle recognition, through the integration of hyperspectral imaging and machine vision technology, it can achieve rapid and accurate recognition of various obstacles, especially the handling of mixed obstacles, and reduce the false alarm rate, ensuring the continuity and safety of transportation. For example, although path P4 contains the most obstacles, due to the accuracy of real-time recognition and classification, the obstacle avoidance response time is still controlled within an acceptable range.

[0209] Secondly, the path optimization strategy combines deep learning with traditional path planning algorithms, effectively reducing energy consumption and transportation time. The comparison between paths P1 and P5 is particularly obvious. Although neither encounters large obstacles, the improved method achieves lower energy consumption and shorter transportation time on the P1 path, indicating the effectiveness of real-time path optimization.

[0210] In addition, the real-time obstacle avoidance function performs outstandingly in complex environments (such as P3). Although the number of obstacles is small, the presence of large obstacles increases the proportion of path deviation from the original plan. However, due to the real-time response of the system, direct collisions are avoided and vaccine safety is guaranteed.

[0211] To sum up, the present invention not only improves the efficiency and safety of vaccine cold chain transportation through highly integrated intelligent technology, but also demonstrates excellent adaptability and flexibility in the face of complex environmental changes. Compared with traditional methods, it has achieved obvious innovation and novel effects in reducing energy consumption, shortening transportation time and improving obstacle avoidance accuracy.

[0212] It should be noted that the above embodiments are only used to illustrate the technical solutions 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 solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A vaccine cold chain transportation method with improved neural network, characterized in that: include, Use hyperspectral imaging and machine vision technology to perceive the environment around the cold chain logistics box to form environmental perception data; Extract key environmental features from environmental perception data and identify obstacles based on the environmental features; Use the deep learning model SSD to build a classification model to detect and classify obstacles and terrain features; Use the path planning algorithm to plan the driving path of the cold chain logistics box and obtain the obstacle distribution data set; The specific steps of extracting key environmental features from environmental perception data are as follows: Set the environmental perception data after radiation correction to D l , the expression is: D l =[R cal ,V index ,T,H,W] Among them, R cal Represents the corrected spectral reflectance array, V index represents vegetation index, T is ambient temperature, H is ambient humidity, and W is ambient wind speed; Using a standardized method, D l Processing, the standardized expression is: Among them, D P represents the environmental perception data after standardization, μ represents the average value of each band data value, and σ represents the standard deviation of each band data value; Using principal component analysis (PCA) to analyze D P Analyze and get the covariance matrix, which is expressed as: Among them, ∑ represents the covariance matrix, N represents the number of samples, Indicates D P The transposed matrix of The covariance matrix ∑ is subjected to eigendecomposition, and the expression is: ∑v=λv Among them, v represents the eigenvector, λ represents the eigenvalue; The eigenvalues ​​are divided into λ1, λ2, ..., λ p Sort from largest to smallest, where p is the total number of eigenvalues; Select the largest k eigenvalues ​​after sorting and record them as λ1, λ2, ..., λ k ; Finally, we select the eigenvectors v1, v2, ..., v corresponding to these k eigenvalues. k , these eigenvectors are combined to obtain the eigenvector matrix F, which is expressed as: F=D P v1,D P v2,...,D P v k Among them, F is the feature vector matrix, representing the environmental characteristics, v1, v2, ..., v k represents the eigenvectors corresponding to k eigenvalues; The specific steps of identifying obstacles based on environmental features are as follows: A hyperspectral imaging sensor installed on the top of the cold chain logistics box is used to obtain a forward-looking image. The obstacles in the front view image are identified by using the extracted environmental features F(x, y) combined with machine vision technology; The Canny edge detection algorithm is used to detect the edges in the image. The expression is: E'(x, y)=α·F(x, y)+(1-α)·E(x, y) Wherein, E′(x, y) represents the edge intensity image after environmental feature adjustment, E(x, y) represents the original edge intensity image, and α is a weight factor used to adjust the influence of environmental feature F(x, y) on edge intensity image E(x, y); Further analysis is performed on the detected edges, including calculating the number of edges per unit area of ​​the detected edges and setting a density threshold P, where the value range of P is 0 to 1; The P includes a low-density interval, a medium-density interval and a high-density interval; When the number of edges per unit area is <0.2, it means that there are no obstacles and micro obstacles in the area; When 0.2≤the number of edges per unit area<0.5, it means that there are small or medium-sized obstacles in the area; When the number of edges per unit area is ≥ 0.5, it indicates that there are large and dense obstacles in the area; Support vector machine (SVM) is used to classify obstacles. The classification expression is: cf = SVM(F, Y) Among them, cf is the classification result, indicating the category assigned to each sample, and Y represents the category label; The method of using a path planning algorithm to plan the driving path of the cold chain logistics box and obtain obstacle distribution data includes setting a search space and identifying and marking the locations of obstacles in the environment; The setting of the search space includes inputting the starting and ending positions of the cold chain logistics box, using machine vision technology to obtain an image of the current position of the cold chain logistics box, and determining the current position of the cold chain logistics box through a positioning algorithm; Identify and mark the location of obstacles in the environment, including using the Canny edge detection algorithm to detect edges in the image and further analyze the detected edges to determine the material, size and location of the obstacles; Combined with neural network to build a transportation obstacle avoidance model, the obstacle distribution data set is input into the transportation obstacle avoidance model to optimize path selection and output the optimal path; During driving, obstacle avoidance is performed using the results of the real-time optimal path and real-time obstacle distribution dataset.

2. The vaccine cold chain transportation method with improved neural network as claimed in claim 1, characterized in that: The hyperspectral imaging and machine vision technology are used to sense the environment around the cold chain logistics box to form environmental perception data. The specific steps are as follows: Hyperspectral imaging sensors are installed on the top and sides of the cold chain logistics box to collect hyperspectral data in multiple bands and obtain environmental perception data; The environmental perception data, the environmental perception data expression is: D=[R spec ,V index ,T,H,W] R spec =[R1,R2,…,R n ] Where D represents the environmental perception data, R spec Represents the spectral reflectance array of different bands, R n represents the spectral reflectance of the nth band, V index represents vegetation index, T is ambient temperature, H is ambient humidity, and W is ambient wind speed; The spectral reflectance R in D spec Perform radiation correction to eliminate system errors; The radiation correction expression is: Among them, R cal represents the corrected spectral reflectance array, R1, R2, ..., R n Represents the original spectral reflectance, R dark，1 , R dark，2 , ..., R dark，n represents the corresponding dark current, R ref，1 , R ref，2 , ..., R ref，n Indicates the corresponding reference spectral radiation value.

3. The vaccine cold chain transportation method with improved neural network as claimed in claim 2, characterized in that: The deep learning model SSD is used to build a classification model to detect and classify obstacles and terrain features. The specific steps are as follows: Build a classification model using the deep learning framework TensorFlow and the SSD model architecture; The input image size is adjusted to a standard size of 300x300 pixels suitable for the classification model, so that images from different sources can be processed uniformly; Use ResNet as the basic network structure; Construct multi-scale feature maps for detecting objects of different sizes; Design prediction boxes of different scales for each feature map, which are used for object detection and classification; The classification model includes an input layer, a basic network structure, a multi-scale feature map generation layer, a prediction box generation layer, a classification layer, a regression layer and an output layer; Input the classification result cf into the constructed classification model for training and obtain the prediction result; The cross entropy loss function is used to measure the difference between the model prediction result and the true value. The expression is: Among them, loss is the loss value, which indicates the difference between the predicted result and the true value, CELoss represents the cross entropy loss function, and y represents the true category label. Represents the predicted category label; The SGD optimizer is used to tune the parameters of the classification model. The tuning expression is: θ = SGD(θ, loss) Among them, θ represents the weight and bias of the classification model; Deploy the trained classification model to computing devices on cold chain logistics boxes for real-time object detection and classification; The forward propagation process of the SSD model is used to perform object detection and classification on the front-view images collected in real time.

4. The vaccine cold chain transportation method with improved neural network as claimed in claim 3, characterized in that: The transportation obstacle avoidance model is constructed by combining a neural network, and the obstacle distribution data set is input into the transportation obstacle avoidance model to optimize the path selection, consider the factors of energy consumption and time efficiency, and output the optimal path. The specific steps are as follows: Use the deep learning framework TensorFlow and the convolutional neural network (CNN) architecture to build a transportation obstacle avoidance model, and set the model architecture, including the input layer, convolution layer, pooling layer, fully connected layer, and output layer; Set the obstacle distribution dataset to B; Input B into the transportation obstacle avoidance model to complete the training of the transportation obstacle avoidance model and obtain the predicted optimal path.

5. The vaccine cold chain transportation method with improved neural network as claimed in claim 4, characterized in that: During the driving process, the real-time optimal path and the results of the real-time obstacle distribution data set are used to avoid obstacles. The specific steps are as follows: Real-time collection of hyperspectral data and machine vision data in front of the cold chain logistics box to form real-time environmental perception data; Use machine vision technology to identify obstacles and extract key real-time environmental features from real-time environmental perception data; Input the real-time environment features into the classification model for classification; The path planning algorithm is used to calculate the classified real-time environmental features to obtain a real-time obstacle distribution data set; The real-time obstacle distribution dataset is input into the transportation obstacle avoidance model to obtain the real-time optimal path.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the vaccine cold chain transportation method with improved neural network as described in any one of claims 1 to 5 are implemented.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the vaccine cold chain transportation method with improved neural network described in any one of claims 1 to 5 are implemented.

Images