Urban cold chain freight rate index prediction method and system based on remote sensing image

Abstract

The invention discloses an urban cold chain freight rate index prediction method and system based on a remote sensing image, and the method comprises the steps: obtaining historical cold chain transport vehicle data, historical urban meteorological data, historical economic data and a corresponding historical cold chain monthly freight rate index, and taking the data as a data set; preprocessing the data set, and dividing the data set into a training set and a test set; constructing a Bayesian neural network model, and based on the training set, using variational inference to iterate Bayesian neural network model parameters for training to obtain an optimized Bayesian neural network model; defining a loss function, and in combination with the test set, measuring the difference between the monthly freight rate index of the urban cold chain predicted by the Bayesian neural network model and a true value; and if the difference does not meet the precision requirement, returning to the previous step to continue training, and if the precision meets the requirement, taking the optimized Bayesian neural network model as an optimal Bayesian neural network model so as to predict the monthly freight rate index of the urban cold chain and provide guidance for the transportation price.

Description

Technical Field

[0001] The present invention relates to the technical field of cold chain transportation price prediction, and in particular to a method and system for predicting a city cold chain freight rate index based on remote sensing images. Background Art

[0002] As an important part of modern supply chain management, cold chain logistics is crucial to ensuring the quality and safety of temperature-sensitive goods such as food and medicine. The temperature control and timeliness requirements involved in cold chain transportation make its cost relatively high and are affected by many factors, such as weather conditions, traffic flow, seasonal demand, etc. Therefore, accurately predicting the cold chain transportation price index is of great significance for optimizing resource allocation, reducing operating costs, and improving market competitiveness.

[0003] At present, forecasting models for freight prices are relatively common, but there is still a lack of accurate forecasting methods for cold chain transportation prices with small data volumes and difficult monitoring.

[0004] In the prior art, the invention patent with the patent publication number CN118095999A discloses a cold chain logistics data management method, device, equipment and storage medium, performs multi-dimensional data monitoring and data preprocessing on the target cold chain logistics, obtains the target multi-dimensional cold chain logistics data set and generates the first cold chain logistics cost information; extracts short-term time correlation features to obtain multiple short-term time correlation features; extracts long-term time correlation features to obtain multiple long-term time correlation features; performs feature weight analysis and feature fusion to obtain fused time correlation features and optimizes the cold chain logistics cargo storage cost and damage risk cost to obtain the first cold chain logistics cost optimization index; performs cold chain logistics transportation task mode cost optimization to obtain the second cold chain logistics cost optimization index; performs cost control strategy optimization to generate the second cold chain logistics cost information. The prior art improves the logistics efficiency of cold chain logistics, but achieves accurate cold chain logistics cost control, rather than cold chain transportation price prediction. Summary of the invention

[0005] The technical problem to be solved by the present invention is to solve the current problem of lack of accurate prediction of cold chain transportation prices.

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

[0007] A method for predicting urban cold chain freight rate index based on remote sensing images, comprising:

[0008] Obtain historical cold chain transport vehicle data, historical city meteorological data, historical economic data, and the corresponding historical cold chain monthly freight rate index as a data set;

[0009] After preprocessing the data set, it is divided into training set and test set;

[0010] Build a Bayesian neural network model based on the training set and use variational inference to iterate the Bayesian neural network model parameters to obtain an optimized Bayesian neural network model;

[0011] Define the loss function, combine with the test set, and measure the difference between the city cold chain monthly freight index predicted by the optimized Bayesian neural network model and the true value; if the difference does not meet the accuracy requirements, return to the previous step to continue training; if the accuracy meets the requirements, use this optimized Bayesian neural network model as the best Bayesian neural network model to predict the city cold chain monthly freight index.

[0012] In one embodiment of the present invention, historical cold chain transport vehicle data is obtained by the following steps:

[0013] Obtain KML geographic information files of the city’s main cold chain logistics parks and large farmers’ market parking lots;

[0014] Based on the KML geographic information files obtained, remote sensing images of parking lots in major cold chain logistics parks and large farmers' markets in the city are collected using image satellites. The collection cycle is daily, and the shooting time points are five time points during the peak delivery and departure periods to obtain remote sensing image data;

[0015] Use the preset local variance curve to segment the outline and background of the refrigerated truck in the remote sensing image data to obtain the vehicle boundary box;

[0016] A convolutional neural network was used to extract the number of vehicle bounding boxes, and all identified vehicle bounding boxes were counted as the number of cold chain transport vehicles to obtain the average daily parking number on the filming date.

[0017] In one embodiment of the present invention, the average daily parking quantity on the shooting date is obtained by the following formula:

[0018]

[0019] Where Total_trucks represents the average daily parking number on the shooting date, h represents each shooting time point, k represents the count of cold chain logistics parks and large farmers' markets in the current city, n represents the number of identified vehicle bounding boxes, and N represents the total number of major cold chain logistics parks and large farmers' markets in the city.

[0020] In one embodiment of the present invention, historical city meteorological data is obtained through a public meteorological website, including temperature data, humidity data, and precipitation data of the city;

[0021] Historical economic data are obtained from government information websites or online trading platforms, including the added value of the primary industry, total retail sales of consumer goods, consumer price index, total fixed asset investment, and highway freight rate index.

[0022] In one embodiment of the present invention, the average freight rates of cold chain transportation types, including fruits and vegetables, poultry, meat and eggs, and aquatic products, are collected based on agricultural product sources and wholesale markets. The collection cycle is monthly, and the sample cold chain vehicles collected are small and medium-sized refrigerated vehicles. The transportation type is distribution within the city, and the freight rates of the same type of cold chain food are averaged;

[0023] According to the proportion of total residents' consumption published in the official statistical yearbook, the proportion of total consumption of each type of cold chain transport products in the overall consumption is defined as the corresponding cold chain transport type consumption weight. Based on the cold chain transport type consumption weight, a cold chain monthly freight index is constructed to obtain the corresponding historical cold chain monthly freight index.

[0024] In one embodiment of the present invention, the monthly cold chain freight index is obtained by the following formula:

[0025] CFI=C 果蔬 ×ω 果蔬 +C 禽肉蛋 ×ω 禽肉蛋 +C 水产品 ×ω 水产品 ;

[0026] In the formula, CFI is the monthly cold chain freight index, C 果蔬 , C 禽肉蛋 , C 水产品 They are respectively the average freight rates of fruits and vegetables, poultry, meat and eggs, and aquatic products transported in the cold chain, ω 果蔬 ,ω 禽肉蛋 ,ω 水产品 They are respectively represented by the consumption weights of fruits and vegetables, poultry, meat and eggs, and aquatic products in cold chain transportation.

[0027] In one embodiment of the present invention, obtaining an optimized Bayesian neural network model includes the following steps:

[0028] Define the network topology of the initial Bayesian neural network model, which includes a three-layer neural network structure with one hidden layer. Each layer of the neural network structure uses RELU as the activation function, takes the data set as input, and takes the predicted city cold chain monthly freight rate index as output;

[0029] Set the prior distribution and select Gaussian distribution as the prior distribution of the weight W and bias b of the neural network;

[0030] Defining the output of the neural network For a Gaussian distribution, obtain the likelihood function: in, x represents the collection of historical cold chain transport vehicle data, historical city meteorological data, and historical economic data in the data set, y represents the corresponding historical cold chain monthly freight rate index, σ represents the variance, N represents the normal distribution, and f(·) represents the neuron of the neural network;

[0031] Infer the true distribution according to Bayes' theorem: In the formula, P(W,b|x,y) represents the posterior distribution, ∝ represents the positive proportion, and P(W,b) represents the prior distribution;

[0032] Use variational inference to solve the posterior distribution, so that the relative entropy of the approximation between the variational distribution Q(W,b) and the true distribution P(W,b|x,y) is described by the KL divergence; where the KL divergence is obtained by the following formula:

[0033]

[0034] Among them, KL[Q(W,b)||P(W,b|x,y)] represents the relative entropy of the approximation between the variational distribution Q(W,b) and the true distribution P(W,b|x,y), and log represents the logarithmic function;

[0035] Minimizing the KL divergence is transformed into maximizing the lower bound of evidence ELBO, and training is performed using the gradient descent method to obtain the true distribution P(W,b|x,y); wherein maximizing the lower bound of evidence ELBO is obtained by the following formula:

[0036] ELBO=E Q(W,b) [logP(y|x,W,b)]-KL(Q(W,b))||P(W,b));

[0037] Among them, E Q(W,b) It is expressed as the expected value under the variational distribution Q(W,b), and P(y|x,W,b) is expressed as the joint probability of the observed data (x,y) and the latent variable (W,b);

[0038] After the initial Bayesian neural network model training is completed, the optimized Bayesian neural network model is obtained.

[0039] In one embodiment of the present invention, the Bayesian neural network model is optimized and obtained by the following formula:

[0040]

[0041] in, Expressed as the predicted distribution, x * It represents the input data, inputting the current monthly cold chain transportation vehicle data, urban meteorological data, and economic data. t represents the period. It is expressed as the predicted monthly freight rate index of cold chain in the next city.

[0042] In one embodiment of the present invention, the loss function is expressed by the following formula:

[0043]

[0044] In the formula, L(θ) represents the loss function, M represents the number of samples, and y t It is expressed as the actual monthly freight rate index of urban cold chain. It is expressed as the predicted monthly freight rate index of urban cold chain, and t represents the number of periods.

[0045] The present invention also provides a city cold chain freight index prediction system based on remote sensing images, which uses the above-mentioned city cold chain freight index prediction method based on remote sensing images, including:

[0046] The data set module is used to obtain historical cold chain transport vehicle data, historical city meteorological data, historical economic data, and the corresponding historical cold chain monthly freight index as a data set;

[0047] The preprocessing module is used to preprocess the data set and divide it into training set and test set;

[0048] A modeling module is used to build a Bayesian neural network model based on a training set and to train the parameters of the Bayesian neural network model using variational inference to obtain an optimized Bayesian neural network model.

[0049] The prediction module is used to define the loss function and measure the difference between the city cold chain monthly freight index predicted by the optimized Bayesian neural network model and the true value in combination with the test set; if the difference does not meet the accuracy requirements, return to the previous step to continue training; if the accuracy meets the requirements, the optimized Bayesian neural network model is used as the best Bayesian neural network model to predict the city cold chain monthly freight index.

[0050] Compared with the prior art, the beneficial effects of the present invention are: combining satellite remote sensing images to provide real-time and comprehensive climate monitoring data and the number of transport vehicles in the cold chain logistics park parking lot, and combining economic data obtained from statistical yearbooks and transport price index platforms to pre-process the data to obtain a basic data set that can be used for Bayesian neural network model training. The probabilistic inference framework based on Bayesian statistics uses variational inference to iterate model parameters, which helps to make more accurate predictions on changes in cold chain transportation prices. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 The present invention is a flowchart of a method for predicting a city cold chain freight rate index based on remote sensing images according to an embodiment of the present invention.

[0052] Figure 2 This is a schematic diagram of a tailored cold chain logistics park according to an embodiment of the present invention.

[0053] Figure 3 The present invention is a block diagram of a system for predicting urban cold chain freight rate index based on remote sensing images. DETAILED DESCRIPTION

[0054] In order to facilitate those skilled in the art to understand the technical solution of the present invention, the technical solution of the present invention is further described in conjunction with the accompanying drawings of the specification.

[0055] The terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0056] See also Figure 1 As shown, the present invention discloses a method for predicting a city cold chain freight rate index based on remote sensing images, comprising:

[0057] S10, obtaining historical cold chain transport vehicle data, historical city meteorological data, historical economic data and the corresponding historical cold chain monthly freight index as a data set.

[0058] In one embodiment of the present invention, historical cold chain transport vehicle data is obtained by the following steps:

[0059] S11, obtain the KML geographic information files of the main cold chain logistics parks and large farmers' market parking lots in the city.

[0060] In this embodiment, taking prefecture-level city A as an example, there are a total of 15 cold chain logistics parks and large farmers' markets, and the KML geographic information files of each cold chain logistics park and large farmers' market with parking lot are manually cut. KML (Keynote markup language) geographic information file, originally defined by Google, is used to describe key data in the map, such as path, marker location, overlay layer and other information. Figure 2 As shown, the cold chain logistics park is cropped and there is a parking lot around the factory building in the cold chain logistics park.

[0061] S12, based on the acquired KML geographic information file, uses image satellites to collect remote sensing images of parking lots in major cold chain logistics parks and large farmers' markets in the city. The collection cycle is daily, and the shooting time points are five time points during the peak delivery and departure periods to obtain remote sensing image data.

[0062] In this embodiment, using high-resolution imaging satellites, some better shooting angles can be selected for acquisition to reduce shadow interference and improve recognition accuracy. The acquired remote sensing image data can also be subjected to background filtering, grayscale conversion, and edge detection to improve image contrast and clarity.

[0063] S13, using a preset local variance curve to segment the outline of the refrigerated truck and the background in the remote sensing image data to obtain a vehicle boundary box.

[0064] In this embodiment, the present invention introduces the local variance curve LV (Local Variance) optimal segmentation scale and merging scale parameter to segment the outline and background of the refrigerated truck from the remote sensing image data and obtain the vehicle boundary box. The image segmentation technology used in this embodiment is a public conventional technology.

[0065] S14, using a convolutional neural network to extract the number of vehicle bounding boxes, and counting all the identified vehicle bounding boxes as the number of cold chain transportation vehicles to obtain the average daily parking number on the shooting date.

[0066] In this embodiment, a convolutional neural network is used to extract a series of vehicle bounding boxes, each of which is represented by the upper left corner coordinate (x i ,y i ) and the lower right corner coordinate (x i +w i ,y i +l i ) indicates that, where w i and l i are the width and length of the bounding box respectively, and i is the count of the bounding box. The number of all identified bounding boxes is counted as the number of cold chain transport vehicles n, that is, n=len(B), where B is the set of all detected bounding boxes, and len is a commonly used function, mainly used to calculate the length or size of an object. In Python, the len() function can be used to obtain the number of elements in data structures such as lists, strings, and dictionaries. Among them, the convolutional neural network in this embodiment has not been improved, but the conventional convolutional neural network is used in this embodiment.

[0067] The average daily parking quantity on the shooting date is obtained by the following formula:

[0068]

[0069] Where Total_trucks represents the average daily parking number on the shooting date, h represents each shooting time point, k represents the count of cold chain logistics parks and large farmers' markets in the current city, n represents the number of identified vehicle bounding boxes, and N represents the total number of major cold chain logistics parks and large farmers' markets in the city.

[0070] In one embodiment of the present invention, historical city meteorological data is obtained through public meteorological websites, including temperature data, humidity data, and precipitation data of the city. Historical economic data is obtained from government information websites or online trading platforms, including the added value of the primary industry, total retail sales of consumer goods, consumer price index, total fixed asset investment, and highway freight rate index.

[0071] In one embodiment of the present invention, the average freight rates of cold chain transport types, including fruits and vegetables, poultry, meat and eggs, and aquatic products, are collected based on agricultural product sources and wholesale markets. The collection cycle is monthly, and the sample cold chain vehicles collected are small and medium-sized refrigerated vehicles. The transportation type is distribution within the city, and the freight rates of the same type of cold chain food are averaged. According to the proportion of total resident consumption published in the official statistical yearbook, the proportion of total consumption of each cold chain transport type of product in the overall total consumption is defined as the corresponding cold chain transport type consumption weight, and a cold chain monthly freight index is constructed based on the cold chain transport type consumption weight to obtain the corresponding historical cold chain monthly freight index.

[0072] In this embodiment, the monthly cold chain freight index is obtained by the following formula:

[0073] CFI=C 果蔬 ×ω 果蔬 +C 禽肉蛋 ×ω 禽肉蛋 +C 水产品 ×ω 水产品 ;

[0074] In the formula, CFI is the monthly cold chain freight index, C 果蔬 , C 禽肉蛋 , C 水产品 They are respectively the average freight rates of fruits and vegetables, poultry, meat and eggs, and aquatic products transported in the cold chain, ω 果蔬 ,ω 禽肉蛋 ,ω 水产品 They are respectively represented by the consumption weights of fruits and vegetables, poultry, meat and eggs, and aquatic products in cold chain transportation.

[0075] In this embodiment, in order to ensure the consistency of data, historical cold chain transport vehicle data, historical city meteorological data, and historical economic data are collected from January 1, 2016, and the deadline is June 30, 2024, when the corresponding historical cold chain monthly freight index is obtained. From the perspective of the cold chain monthly freight index, the cold chain monthly freight index is only related to the average freight rate and consumption weight of the cold chain transport type. However, from the perspective of macro data, cold chain transport vehicle data, city meteorological data, and economic data affect the average freight rate and consumption weight of the cold chain transport type.

[0076] S20, preprocessing the data set and dividing it into a training set and a test set.

[0077] In this embodiment, the data set is uniformly converted into monthly data or monthly average data by performing outlier test and missing value processing, wherein the data set after data sorting is as shown in Table 1:

[0078] Table 1 The sorted data set

[0079]

[0080] In this embodiment, the combed data set is normalized and the result is mapped to [0.1], which is converted into a data set that can be used for Bayesian neural network model training, wherein 80% of the data set is used as a training set and 20% as a test set.

[0081] S30, constructing a Bayesian neural network model based on the training set, and using variational inference to iterate the Bayesian neural network model parameters for training to obtain an optimized Bayesian neural network model.

[0082] In this embodiment, obtaining an optimized Bayesian neural network model includes the following steps:

[0083] S31, defines the network topology of the initial Bayesian neural network model, which includes a three-layer neural network structure with one hidden layer, wherein each layer of the neural network structure adopts RELU as the activation function, takes the data set as input, and takes the predicted city cold chain monthly freight index as output.

[0084] S32, set the prior distribution, and select Gaussian distribution as the prior distribution of the weight W and bias b of the neural network, that is,

[0085] S33, define the output of the neural network For a Gaussian distribution, obtain the likelihood function: in, x represents the collection of historical cold chain transport vehicle data, historical city meteorological data, and historical economic data in the data set, y represents the corresponding historical cold chain monthly freight rate index, σ represents the variance, N represents the normal distribution, and f(·) represents the neuron of the neural network;

[0086] S34, infer the true distribution based on Bayes' theorem: In the formula, P(W,b|x,y) represents the posterior distribution, ∝ represents the positive proportion, and P(W,b) represents the prior distribution;

[0087] S35, using variational inference to solve the posterior distribution, so that the relative entropy of the approximation between the variational distribution Q(W,b) and the true distribution P(W,b|x,y) is described by KL divergence; wherein, KL divergence is obtained by the following formula:

[0088]

[0089] Among them, KL[Q(W,b)||P(W,b|x,y)] represents the relative entropy of the approximation between the variational distribution Q(W,b) and the true distribution P(W,b|x,y), and log represents the logarithmic function.

[0090] S36, transform the minimization of KL divergence into maximizing the lower bound of evidence ELBO, use the gradient descent method for training, and obtain the true distribution P(W,b|x,y). Thus, the solution of P(W,b|x,y) is transformed into the optimal solution of the optimization problem, where the maximization of the lower bound of evidence ELBO is obtained by the following formula:

[0091] ELBO=E Q(W,b) [logP(y|x,W,b)]-KL(Q(W,b))||P(W,b));

[0092] Among them, E Q(W,b) It is expressed as the expected value under the variational distribution Q(W,b), and P(y|x,W,b) is expressed as the joint probability of the observed data (x,y) and the latent variables (W,b).

[0093] S36, after the initial Bayesian neural network model training is completed, an optimized Bayesian neural network model is obtained. The inferred posterior distribution is used to predict the probability distribution of the future cold chain freight index. The optimized Bayesian neural network model is obtained by the following formula:

[0094]

[0095] in, Expressed as the predicted distribution, x * It represents the input data, inputting the current monthly cold chain transportation vehicle data, urban meteorological data, and economic data. t represents the period. It is represented as the predicted monthly freight rate index of the cold chain in the next city. In this implementation, 800 samples are generated in the predicted distribution of y, and the mean is taken to represent the final predicted value.

[0096] S40, define the loss function, combine with the test set, measure the difference between the city cold chain monthly freight index predicted by the optimized Bayesian neural network model and the true value; if the difference does not meet the accuracy requirement, return to the previous step to continue training, if the accuracy meets the requirements, then this optimized Bayesian neural network model is used as the best Bayesian neural network model to predict the city cold chain monthly freight index.

[0097] In this embodiment, the loss function is expressed by the following formula:

[0098]

[0099] In the formula, L(θ) represents the loss function, M represents the number of samples, and y t It is expressed as the actual monthly freight rate index of urban cold chain. The predicted monthly cold chain freight rate index of the city is expressed as t, and the period is expressed as t. The input feature data set of prefecture-level city A in June 2024 was collected through the above data collection method and input into the above optimal Bayesian neural network model, and the monthly cold chain freight rate index of prefecture-level city A in July was obtained as 87.

[0100] In one embodiment of the present invention, the present invention can store the predicted city cold chain monthly freight rate index in real time and display it to the user in the form of a graph or table, supporting interactive operations.

[0101] See also Figure 3 As shown, the present invention also provides a city cold chain freight index prediction system based on remote sensing images, which uses the above-mentioned city cold chain freight index prediction method based on remote sensing images, including:

[0102] The data set module 10 is used to obtain historical cold chain transport vehicle data, historical city meteorological data, historical economic data and the corresponding historical cold chain monthly freight index as a data set.

[0103] The preprocessing module 20 is used to preprocess the data set and divide it into a training set and a test set.

[0104] The modeling module 30 is used to construct a Bayesian neural network model based on a training set and to train the parameters of the Bayesian neural network model using variational inference to obtain an optimized Bayesian neural network model.

[0105] The prediction module 40 is used to define the loss function and measure the difference between the city cold chain monthly freight index predicted by the optimized Bayesian neural network model and the true value in combination with the test set; if the difference does not meet the accuracy requirement, return to the previous step to continue training; if the accuracy meets the requirement, the optimized Bayesian neural network model is used as the best Bayesian neural network model to predict the city cold chain monthly freight index.

[0106] It is obvious to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, the embodiments should be regarded as exemplary and non-limiting from any point of view, and the scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention, and any reference numerals in the claims should not be regarded as limiting the claims involved.

[0107] The above-described embodiments merely represent implementation methods of the invention. The protection scope of the present invention is not limited to the above-described embodiments. For those skilled in the art, several modifications and improvements may be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. A method for predicting urban cold chain freight rate index based on remote sensing images, characterized in that: include: Obtain historical cold chain transport vehicle data, historical city meteorological data, historical economic data, and the corresponding historical cold chain monthly freight rate index as a data set; After preprocessing the data set, it is divided into training set and test set; Build a Bayesian neural network model based on the training set and use variational inference to iterate the Bayesian neural network model parameters to obtain an optimized Bayesian neural network model; Define the loss function, combine with the test set, and measure the difference between the city cold chain monthly freight index predicted by the optimized Bayesian neural network model and the true value; if the difference does not meet the accuracy requirements, return to the previous step to continue training; if the accuracy meets the requirements, use this optimized Bayesian neural network model as the best Bayesian neural network model to predict the city cold chain monthly freight index.

2. The method for predicting urban cold chain freight rate index based on remote sensing images according to claim 1 is characterized in that: Historical cold chain transport vehicle data can be obtained through the following steps: Obtain KML geographic information files of the city’s main cold chain logistics parks and large farmers’ market parking lots; Based on the KML geographic information files obtained, remote sensing images of parking lots in major cold chain logistics parks and large farmers' markets in the city are collected using image satellites. The collection cycle is daily, and the shooting time points are five time points during the peak delivery and departure periods to obtain remote sensing image data; Use the preset local variance curve to segment the outline and background of the refrigerated truck in the remote sensing image data to obtain the vehicle boundary box; A convolutional neural network was used to extract the number of vehicle bounding boxes, and all identified vehicle bounding boxes were counted as the number of cold chain transport vehicles to obtain the average daily parking number on the filming date.

3. The method for predicting urban cold chain freight rate index based on remote sensing images according to claim 2 is characterized in that: Get the average daily parking quantity on the shooting date using the following formula: Where Total_trucks represents the average daily parking number on the shooting date, h represents each shooting time point, k represents the count of cold chain logistics parks and large farmers' markets in the current city, n represents the number of identified vehicle bounding boxes, and N represents the total number of major cold chain logistics parks and large farmers' markets in the city.

4. The method for predicting urban cold chain freight rate index based on remote sensing images according to claim 1 is characterized in that: Historical urban meteorological data are obtained through public meteorological websites, including the city’s temperature data, humidity data, and precipitation data; historical economic data are obtained from government information websites or online trading platforms, including the added value of the primary industry, total retail sales of consumer goods, consumer price index, total fixed asset investment, and highway freight rate index.

5. The method for predicting urban cold chain freight rate index based on remote sensing images according to claim 1 is characterized in that: Relying on agricultural product sources and wholesale markets, we collected the average freight rates of cold chain transportation, including fruits and vegetables, poultry, meat and eggs, and aquatic products. The collection cycle is monthly. The sample cold chain vehicles collected are small and medium-sized refrigerated vehicles. The transportation type is distribution within the city. The freight rates of the same type of cold chain food are averaged. According to the proportion of total residents' consumption published in the official statistical yearbook, the proportion of total consumption of each type of cold chain transport products in the overall consumption is defined as the corresponding cold chain transport type consumption weight. Based on the cold chain transport type consumption weight, a cold chain monthly freight index is constructed to obtain the corresponding historical cold chain monthly freight index.

6. The method for predicting urban cold chain freight rate index based on remote sensing images according to claim 5 is characterized in that: The monthly cold chain freight index is obtained by the following formula: CFI=C 果蔬 ×ω 果蔬 +C 禽肉蛋 ×ω 禽肉蛋 +C 水产品 ×ω 水产品 ; In the formula, CFI is the monthly cold chain freight index, C 果蔬 , C 禽肉蛋 , C 水产品 They are respectively the average freight rates of fruits and vegetables, poultry, meat and eggs, and aquatic products transported in the cold chain, ω 果蔬 ,ω 禽肉蛋 ,ω 水产品 They are respectively represented by the consumption weights of fruits and vegetables, poultry, meat and eggs, and aquatic products in cold chain transportation.

7. The method for predicting urban cold chain freight rate index based on remote sensing images according to claim 1 is characterized in that: Obtaining an optimized Bayesian neural network model includes the following steps: Define the network topology of the initial Bayesian neural network model, which includes a three-layer neural network structure with one hidden layer. Each layer of the neural network structure uses RELU as the activation function, takes the data set as input, and takes the predicted city cold chain monthly freight rate index as output; Set the prior distribution and select Gaussian distribution as the prior distribution of the weight W and bias b of the neural network; Defining the output of the neural network For a Gaussian distribution, obtain the likelihood function: in, x represents the collection of historical cold chain transport vehicle data, historical city meteorological data, and historical economic data in the data set, y represents the corresponding historical cold chain monthly freight rate index, σ represents the variance, N represents the normal distribution, and f(·) represents the neuron of the neural network; Infer the true distribution according to Bayes' theorem: In the formula, P(W, b|x, y) represents the posterior distribution, ∝ represents the positive proportion, and P(W, b) represents the prior distribution; Use variational inference to solve the posterior distribution, so that the relative entropy of the approximation between the variational distribution Q(W, b) and the true distribution P(W, b|x, y) is described by the KL divergence; where the KL divergence is obtained by the following formula: Among them, KL[Q(W, b)||P(W, b|x, y)] represents the relative entropy of the approximation between the variational distribution Q(W, b) and the true distribution P(W, b|x, y), and log represents the logarithmic function; Minimizing the KL divergence is transformed into maximizing the lower bound of evidence ELBO, and training is performed using the gradient descent method to obtain the true distribution P(W, b |x, y); wherein maximizing the lower bound of evidence ELBO is obtained by the following formula: ELBO=E Q(W,b) [logP(y|x,W,b)]-KL(Q(W,b))||P(W,b)); Among them, E Q(W,b) It is expressed as the expected value under the variational distribution Q(W, b), and P(y|x, W, b) is expressed as the joint probability of the observed data (x, y) and the latent variable (W, b); After the initial Bayesian neural network model training is completed, the optimized Bayesian neural network model is obtained.

8. The method for predicting urban cold chain freight rate index based on remote sensing images according to claim 7 is characterized in that: The optimized Bayesian neural network model is obtained by the following formula: in, Expressed as the predicted distribution, x * It represents the input data, inputting the current monthly cold chain transport vehicle data, urban meteorological data, and economic data. t represents the period. It is expressed as the predicted monthly freight rate index of cold chain in the next city.

9. The method for predicting urban cold chain freight rate index based on remote sensing images according to claim 8 is characterized in that: The loss function is expressed by the following formula: In the formula, L(θ) represents the loss function, M represents the number of samples, and y t It is expressed as the actual monthly freight rate index of urban cold chain. It is expressed as the predicted monthly freight rate index of urban cold chain, and t represents the period.

10. A city cold chain freight rate index prediction system based on remote sensing images, characterized in that: The method for predicting the urban cold chain freight rate index based on remote sensing images according to any one of claims 1 to 9 comprises: The data set module is used to obtain historical cold chain transport vehicle data, historical city meteorological data, historical economic data, and the corresponding historical cold chain monthly freight index as a data set; The preprocessing module is used to preprocess the data set and divide it into training set and test set; A modeling module is used to build a Bayesian neural network model based on a training set and to train the parameters of the Bayesian neural network model using variational inference to obtain an optimized Bayesian neural network model. The prediction module is used to define the loss function and measure the difference between the city cold chain monthly freight index predicted by the optimized Bayesian neural network model and the true value in combination with the test set; if the difference does not meet the accuracy requirements, return to the previous step to continue training; if the accuracy meets the requirements, the optimized Bayesian neural network model is used as the best Bayesian neural network model to predict the city cold chain monthly freight index.

Images