Aviation logistics management methods for cross-border e-commerce

Abstract

The present application discloses an air logistics management method for cross-border e-commerce. In the process of air logistics warehouse management for cross-border e-commerce, the surface state image of the goods to be sorted is collected by a camera, and an image processing and analysis algorithm based on artificial intelligence and machine vision is introduced at the back end to analyze the surface state image of the goods to be sorted, so as to learn and capture the surface state characteristics of the goods to be sorted, such as size, shape, color and semantics, so as to identify the type of the goods. In this way, efficient and automated sorting and packaging of goods can be carried out, which helps to achieve more efficient, reliable and sustainable air logistics management.

Description

Technical Field

[0001] The present application relates to the field of intelligent logistics management, and specifically to an aviation logistics management method for cross-border e-commerce. Background Art

[0002] With the growth of global trade, cross-border e-commerce has become the choice of more and more companies, which requires the logistics industry to provide more efficient and reliable services to meet the growing market demand. As a fast and efficient mode of transportation, air logistics can shorten the transportation time of goods, improve logistics efficiency, and meet consumers' demand for fast delivery. Therefore, air logistics plays a vital role in cross-border e-commerce.

[0003] Warehouse management is a key link in the air logistics management of cross-border e-commerce. However, traditional warehouses usually need to rely on a lot of manual operations when sorting goods, which not only increases labor costs, but also is prone to human errors under high work intensity. In addition, the speed of manual sorting is relatively slow, especially during peak hours, and it is impossible to quickly process a large amount of goods, resulting in backlogs and delays. With the diversification of market demand, the types and specifications of goods are increasing. The traditional sorting method has a low degree of intelligence and is difficult to quickly adapt to such changes, which will lead to a high error rate in the division and sorting of cross-border e-commerce goods.

[0004] Therefore, an optimized aviation logistics management solution for cross-border e-commerce is desired. Summary of the invention

[0005] This application is made in view of the above problems. One purpose of this application is to provide an aviation logistics management method for cross-border e-commerce.

[0006] The embodiment of the present application provides an aviation logistics management method for cross-border e-commerce, which includes:

[0007] Acquire the surface state image of the goods to be sorted collected by the camera;

[0008] Performing multi-level feature extraction on the surface state image of the goods to be sorted by a goods surface state feature extractor to obtain a shallow feature map of the goods surface state and a deep feature map of the goods surface state;

[0009] Performing feature multi-scale perception enhancement analysis on the cargo surface state deep feature map to obtain an enhanced cargo surface state deep feature map;

[0010] Inputting the shallow feature map of the cargo surface state and the enhanced deep feature map of the cargo surface state into a semantic mask association fusion module to obtain a multi-scale fusion feature map of the cargo surface state;

[0011] Determining the type label of the goods to be sorted based on the multi-scale fusion feature map of the surface state of the goods;

[0012] Based on the type labels of the goods to be sorted, the goods to be sorted are divided into corresponding capacity pools.

[0013] For example, according to the embodiment of the present application, an air logistics management method for cross-border e-commerce, wherein a surface state image of the goods to be sorted is subjected to multi-level feature extraction by a cargo surface state feature extractor to obtain a cargo surface state shallow feature map and a cargo surface state deep feature map, comprising: inputting the surface state image of the goods to be sorted into a cargo surface state feature extractor based on a bidirectional pyramid network to obtain the cargo surface state shallow feature map and the cargo surface state deep feature map.

[0014] For example, according to the air logistics management method for cross-border e-commerce of the embodiment of the present application, a feature multi-scale perception enhancement analysis is performed on the deep feature map of the surface state of the cargo to obtain an enhanced deep feature map of the surface state of the cargo, including: inputting the deep feature map of the surface state of the cargo into a feature multi-scale perception enhancement module to obtain the enhanced deep feature map of the surface state of the cargo.

[0015] For example, according to the aviation logistics management method for cross-border e-commerce in an embodiment of the present application, the deep feature map of the surface state of the cargo is input into the feature multi-scale perception enhancement module to obtain the enhanced deep feature map of the surface state of the cargo, including:

[0016] In the first branch, point convolution processing is performed on the deep feature map of the surface state of the goods to obtain a first deep channel compression feature map of the surface state of the goods; global mean pooling is performed on the first deep channel compression feature map of the surface state of the goods to obtain a first deep channel compression feature vector of the surface state of the goods; nonlinear activation processing is performed on the first deep channel compression feature vector of the surface state of the goods to obtain a first deep channel local activation feature vector of the surface state of the goods;

[0017] In the second branch, point convolution processing is performed on the deep feature map of the surface state of the goods to obtain a second deep channel compression feature map of the surface state of the goods;

[0018] In the third branch, the deep feature map of the surface state of the goods is subjected to hole convolution coding to obtain a deep receptive field expansion feature map of the surface state of the goods; the deep receptive field expansion feature map of the surface state of the goods is subjected to point convolution processing to obtain a global deep feature matrix of the surface state of the goods with expanded receptive field; the global deep feature matrix of the surface state of the goods with expanded receptive field is subjected to nonlinear activation to obtain a global deep activation feature matrix of the surface state of the goods with expanded receptive field;

[0019] Multiply the global deep activation feature matrix of the receptive field expanded cargo surface state by each feature matrix along the channel dimension of the second cargo surface state deep channel compression feature map by position points to obtain the second channel compression cargo surface state global deep activation feature map;

[0020] Taking each position feature value in the local activation feature vector of the deep channel of the first cargo surface state as a weighted weight, weighting each feature matrix along the channel dimension of the deep channel compression feature map of the second cargo surface state to obtain the deep channel compression local activation feature map of the second cargo surface state;

[0021] The second channel compressed cargo surface state global deep activation feature map and the second cargo surface state deep channel compressed local activation feature map are added by position to obtain the second channel compressed cargo surface state deep multi-scale fusion activation feature map;

[0022] The deep multi-scale fusion activation feature map of the compressed cargo surface state of the second channel is subjected to hole convolution encoding to obtain a deep feature map of the enhanced cargo surface state.

[0023] For example, according to the aviation logistics management method for cross-border e-commerce of an embodiment of the present application, the shallow feature map of the surface state of the cargo and the deep feature map of the enhanced surface state of the cargo are input into a semantic mask association fusion module to obtain a multi-scale fusion feature map of the surface state of the cargo, including:

[0024] The shallow characteristic map of the surface state of the cargo and the deep characteristic map of the surface state of the enhanced cargo are spliced ​​based on the channel dimension to obtain a shallow-deep spliced ​​characteristic map of the surface state of the cargo;

[0025] Performing convolution processing on the shallow-deep splicing feature map of the surface state of the goods to obtain a shallow-deep splicing hidden semantic feature map of the surface state of the goods;

[0026] Using a transformation matrix and a bias matrix, a linear transformation is performed on the shallow-deep concatenated hidden semantic feature map of the surface state of the goods to obtain a linearly transformed shallow-deep concatenated hidden semantic feature map of the surface state of the goods;

[0027] Using a softmax activation function to activate the shallow-deep concatenated hidden semantic feature map of the linearly transformed cargo surface state to obtain an activated shallow-deep concatenated hidden semantic feature map of the linearly transformed cargo surface state;

[0028] Taking the feature values ​​of each position in the shallow-deep concatenated hidden semantic feature map of the activated linear transformation cargo surface state as the threshold weight, the position-weighted sum of the shallow feature map of the cargo surface state and the enhanced deep feature map of the cargo surface state is calculated to obtain the multi-scale fusion feature map of the cargo surface state.

[0029] For example, according to an embodiment of the present application, an air logistics management method for cross-border e-commerce is used, wherein a transformation matrix and a bias matrix are used to perform a linear transformation on the shallow-deep splicing hidden semantic feature map of the surface state of the cargo to obtain a linearly transformed shallow-deep splicing hidden semantic feature map of the surface state of the cargo, including: calculating the matrix multiplication of each feature matrix along the channel dimension in the shallow-deep splicing hidden semantic feature map of the surface state of the cargo and the transformation matrix, and then adding each feature matrix along the channel dimension in the obtained feature map with the bias matrix by position to obtain the linearly transformed shallow-deep splicing hidden semantic feature map of the surface state of the cargo.

[0030] For example, according to the aviation logistics management method for cross-border e-commerce of the embodiment of the present application, each position feature value in the shallow-deep splicing hidden semantic feature map of the activated linear transformation cargo surface state is used as a threshold value weight, and the position-weighted sum of the shallow feature map of the cargo surface state and the enhanced deep feature map of the cargo surface state is calculated to obtain the multi-scale fusion feature map of the cargo surface state, including:

[0031] Calculate the product of the characteristic value of each position in the shallow characteristic map of the surface state of the goods and the threshold value weight to obtain a weighted shallow characteristic map of the surface state of the goods;

[0032] Calculate the product of the feature value of each position in the enhanced cargo surface state deep feature map and one minus the threshold value weight to obtain a weighted enhanced cargo surface state deep feature map;

[0033] The weighted item cargo surface state shallow feature map and the weighted item enhanced cargo surface state deep feature map are added according to the position points to obtain the cargo surface state multi-scale fusion feature map.

[0034] For example, according to the aviation logistics management method for cross-border e-commerce of the embodiment of the present application, the type label of the goods to be sorted is determined based on the multi-scale fusion feature map of the surface state of the goods, including:

[0035] The multi-scale fusion feature map of the surface state of the goods is input into an automatic sorting module based on a classifier to obtain a sorting result, and the sorting result is used to represent the type label of the goods to be sorted.

[0036] For example, according to the air logistics management method for cross-border e-commerce of the embodiment of the present application, the goods to be sorted are divided into corresponding capacity pools based on the type labels of the goods to be sorted, including: based on the sorting results, the goods to be sorted are divided into corresponding capacity pools by an AGV cart.

[0037] For example, according to the aviation logistics management method for cross-border e-commerce of the embodiment of the present application, it also includes a training step: for training the cargo surface state feature extractor based on the bidirectional pyramid network, the feature multi-scale perception enhancement module, the semantic mask association fusion module and the classifier-based automatic sorting module;

[0038] Wherein, the training step includes:

[0039] Acquire training data, wherein the training data includes a training surface state image of the goods to be sorted acquired by a camera and a real result representing a type label of the goods to be sorted;

[0040] Inputting the training surface state image of the goods to be sorted into the goods surface state feature extractor based on the bidirectional pyramid network to obtain a shallow feature map of the training goods surface state and a deep feature map of the training goods surface state;

[0041] Inputting the training cargo surface state deep feature map into the feature multi-scale perception enhancement module to obtain a training enhanced cargo surface state deep feature map;

[0042] Inputting the shallow feature map of the surface state of the training cargo and the deep feature map of the surface state of the training enhanced cargo into the semantic mask association fusion module to obtain a multi-scale fusion feature map of the surface state of the training cargo;

[0043] Inputting the multi-scale fusion feature map of the surface state of the training goods into the classifier-based automatic sorting module to obtain a training sorting result;

[0044] Calculating a cross entropy loss function value between the training sorting result and the true result as a classification loss function value;

[0045] The cargo surface state feature extractor based on the bidirectional pyramid network, the feature multi-scale perception enhancement module, the semantic mask association fusion module and the classifier-based automatic sorting module are trained based on the classification loss function value.

[0046] According to the air logistics management method for cross-border e-commerce of the embodiment of the present application, in the process of air logistics warehouse management for cross-border e-commerce, the surface state image of the goods to be sorted is collected by a camera, and an image processing and analysis algorithm based on artificial intelligence and machine vision is introduced at the back end to analyze the surface state image of the goods to be sorted, so as to learn and capture the surface state features of the goods to be sorted, such as size, shape, color and semantics, so as to identify the type of the goods. In this way, efficient and automated sorting and packaging of goods can be carried out, which helps to achieve more efficient, reliable and sustainable air logistics management. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings of the embodiments of the present application are briefly introduced below. Obviously, the drawings described below only relate to some embodiments of the present application, and are not intended to limit the present application.

[0048] Figure 1 A schematic diagram of the application architecture of the aviation logistics management method for cross-border e-commerce in an embodiment of the present application is shown;

[0049] Figure 2 A flow chart of an aviation logistics management method for cross-border e-commerce in an embodiment of the present application is shown;

[0050] Figure 3 A flowchart of sub-step S540 of the aviation logistics management method for cross-border e-commerce in an embodiment of the present application is shown;

[0051] Figure 4 A schematic diagram of the structure of an aviation logistics management system for cross-border e-commerce in an embodiment of the present application is shown;

[0052] Figure 5 An application scenario diagram of the aviation logistics management method for cross-border e-commerce in an embodiment of the present application is shown;

[0053] Figure 6 The structure diagram of the ERP intelligent operation platform in the embodiment of the present application is shown; and

[0054] Figure 7 A schematic diagram of a storage medium according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0055] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work also fall within the scope of protection of the present application.

[0056] The terms used in this specification are those common terms currently widely used in the art in consideration of the functions of the present application, but these terms may vary according to the intention of a person of ordinary skill in the art, precedents, or new technologies in the art. In addition, specific terms may be selected by the applicant, and in this case, their detailed meanings will be described in the detailed description of the present application. Therefore, the terms used in the specification should not be understood as simple names, but rather as a general description based on the meaning of the terms and the present application.

[0057] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0058] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps may be processed in reverse order or simultaneously as required. At the same time, other operations may also be added to these processes, or a certain step or several steps of operations may be removed from these processes.

[0059] Figure 1 A schematic diagram of the application architecture of the aviation logistics management method for cross-border e-commerce in an embodiment of the present application is shown, including a server 100 and a terminal device 200.

[0060] The terminal device 200 and the server 100 can be connected via the Internet to achieve mutual communication. Optionally, the above-mentioned Internet uses standard communication technology and / or protocol. The Internet is usually the Internet, but it can also be any network, including but not limited to any combination of a local area network (LAN), a metropolitan area network (MAN), a wide area network (WAN), a mobile, wired or wireless network, a dedicated network or a virtual private network. In some embodiments, the data exchanged through the network is represented by technologies and / or formats including Hyper Text Markup Language (HTML), Extensible Markup Language (XML), etc. In addition, conventional encryption technologies such as Secure Socket Layer (SSL), Transport Layer Security (TLS), Virtual Private Network (VPN), Internet Protocol Security (IPsec) can also be used to encrypt all or some links. In other embodiments, customized and / or dedicated data communication technologies can also be used to replace or supplement the above-mentioned data communication technologies.

[0061] The server 100 can provide various network services for the terminal device 200, wherein the server 100 can be a single server, a server cluster consisting of several servers, or a cloud computing center. Specifically, the server 100 may include a processor 110 (Center Processing Unit, CPU), a memory 120, an input device 130, and an output device 140, etc. The input device 130 may include a keyboard, a mouse, a touch screen, etc., and the output device 140 may include a display device, such as a liquid crystal display (Liquid Crystal Display, LCD), a cathode ray tube (Cathode Ray Tube, CRT), etc.

[0062] The memory 120 may include a read-only memory (ROM) and a random access memory (RAM), and provides the processor 110 with program instructions and data stored in the memory 120. In an embodiment of the present application, the memory 120 may be used to store a program of the aviation logistics management method for cross-border e-commerce in an embodiment of the present application.

[0063] The processor 110 calls the program instructions stored in the memory 120, and the processor 110 is used to execute the steps of any one of the air logistics management methods for cross-border e-commerce in the embodiments of the present application according to the obtained program instructions.

[0064] In addition, the application architecture diagram in the embodiment of the present application is intended to more clearly illustrate the technical solution in the embodiment of the present application, and does not constitute a limitation on the technical solution provided in the embodiment of the present application. Of course, for other application architectures and business applications, the technical solution provided in the embodiment of the present application is also applicable to similar problems.

[0065] The following is a non-restrictive description of the aviation logistics management method for cross-border e-commerce provided according to at least one embodiment of the present application through several examples or embodiments. As described below, different features in these specific examples or embodiments can be combined with each other without conflicting with each other to obtain new examples or embodiments, and these new examples or embodiments also fall within the scope of protection of the present application.

[0066] In response to the above technical problems, the technical concept of this application is to collect the surface state images of the goods to be sorted through cameras during the air logistics warehouse management of cross-border e-commerce, and introduce image processing and analysis algorithms based on artificial intelligence and machine vision at the back end to analyze the surface state images of the goods to be sorted, so as to learn and capture the surface state characteristics of the goods to be sorted, such as size, shape, color and semantics, so as to identify the type of the goods. In this way, it is possible to achieve rapid and accurate classification of goods in the air logistics warehouse management of cross-border e-commerce, and divide the goods into corresponding capacity pools. In this way, efficient and automated sorting and packaging of goods can be carried out, which helps to achieve more efficient, reliable and sustainable air logistics management.

[0067] Figure 2 The flowchart of the aviation logistics management method for cross-border e-commerce in the embodiment of the present application is shown. For example, the aviation logistics management method for cross-border e-commerce can be executed by a server, which can be Figure 1 The server 100 shown in FIG. Figure 2As shown, according to the embodiment of the present application, the air logistics management method for cross-border e-commerce includes the following steps: S510, acquiring a surface state image of the goods to be sorted collected by a camera; S520, performing multi-level feature extraction on the surface state image of the goods to be sorted by a goods surface state feature extractor to obtain a shallow feature map of the goods surface state and a deep feature map of the goods surface state; S530, performing feature multi-scale perception enhancement analysis on the deep feature map of the goods surface state to obtain an enhanced deep feature map of the goods surface state; S540, inputting the shallow feature map of the goods surface state and the enhanced deep feature map of the goods surface state into a semantic mask association fusion module to obtain a multi-scale fusion feature map of the goods surface state; S550, determining the type label of the goods to be sorted based on the multi-scale fusion feature map of the goods surface state; S560, dividing the goods to be sorted into corresponding capacity pools based on the type label of the goods to be sorted.

[0068] Specifically, in the technical solution of the present application, first, the surface state image of the goods to be sorted collected by the camera is obtained. It should be understood that since the surface state features of the goods to be sorted are distributed in different levels in the image, for example, in the shallow features of the image, they are manifested as basic features such as the edge, texture and color of the goods, while in the deep features of the image, they are manifested as more abstract semantic features of the goods. Based on this, in order to more comprehensively capture the surface state features of the goods in the surface state image of the goods to be sorted, so as to achieve more accurate classification and sorting of goods, in the technical solution of the present application, the surface state image of the goods to be sorted is input into the goods surface state feature extractor based on the bidirectional pyramid network to obtain the shallow feature map of the goods surface state and the deep feature map of the goods surface state. It should be understood that the bidirectional pyramid network can extract image features from different levels, including shallow features and deep features. The shallow features usually contain basic information such as the edge and texture of the target object in the image, while the deep features contain more abstract semantic information of the target object. Through the processing of the cargo surface state feature extractor based on the bidirectional pyramid network, the cargo surface state feature information in the surface state image can be extracted from different scales and levels. This can more comprehensively and accurately identify the surface state of the cargo, improve the generalization ability of the model to identify the cargo type in different scenes, cope with different complex scenes, such as illumination changes, occlusion, background interference, etc., and improve the robustness of the sorting system in different environments.

[0069] Accordingly, in step S520, a cargo surface state feature extractor performs multi-level feature extraction on the surface state image of the cargo to be sorted to obtain a cargo surface state shallow feature map and a cargo surface state deep feature map, including: inputting the surface state image of the cargo to be sorted into a cargo surface state feature extractor based on a bidirectional pyramid network to obtain the cargo surface state shallow feature map and the cargo surface state deep feature map.

[0070] Then, in the air logistics warehouse management of cross-border e-commerce, the goods may have different scales and sizes, and at the same time, the differences in the types of some goods are reflected in the slight differences in the surface state. Therefore, in order to further improve the characterization ability of the goods characteristics and the recognition accuracy of the system, in the technical solution of the present application, the deep feature map of the surface state of the goods is input into the feature multi-scale perception enhancement module to obtain the enhanced deep feature map of the surface state of the goods. Through the processing of the feature multi-scale perception enhancement module, the local and global deep semantics of the surface state of the goods can be learned and captured, so as to further process and fuse the deep features of the surface state of the goods at different scales, so that the system can perceive and understand the characteristics of the surface of the goods more comprehensively. This helps the system to better adapt to goods of different scales and sizes, and at the same time can better portray the subtle differences in the surface state of the goods, so as to more accurately distinguish different types of goods and improve the accuracy of classification and sorting.

[0071] Accordingly, in step S530, a feature multi-scale perception enhancement analysis is performed on the cargo surface state deep feature map to obtain an enhanced cargo surface state deep feature map, including: inputting the cargo surface state deep feature map into a feature multi-scale perception enhancement module to obtain the enhanced cargo surface state deep feature map.

[0072] Specifically, the deep feature map of the surface state of the goods is input into the feature multi-scale perception enhancement module to obtain the enhanced deep feature map of the surface state of the goods, including: in the first branch, performing point convolution processing on the deep feature map of the surface state of the goods to obtain a first deep channel compression feature map of the surface state of the goods; performing global mean pooling on the first deep channel compression feature map of the surface state of the goods to obtain a first deep channel compression feature vector of the surface state of the goods; performing nonlinear activation processing on the first deep channel compression feature vector of the surface state of the goods to obtain a first deep channel local activation feature vector of the surface state of the goods; in the second branch, performing point convolution processing on the deep feature map of the surface state of the goods to obtain a second deep channel compression feature map of the surface state of the goods; in the third branch, performing atrous convolution coding on the deep feature map of the surface state of the goods to obtain a deep receptive field expansion feature map of the surface state of the goods; performing point convolution processing on the deep receptive field expansion feature map of the surface state of the goods to obtain a global deep feature matrix of the surface state of the goods with expanded receptive field; The global deep feature matrix is ​​nonlinearly activated to obtain the global deep activation feature matrix of the receptive field expanded cargo surface state; the global deep activation feature matrix of the receptive field expanded cargo surface state is multiplied by each feature matrix along the channel dimension of the second cargo surface state deep channel compression feature map according to the position point to obtain the second channel compressed cargo surface state global deep activation feature map; each position feature value in the first cargo surface state deep channel local activation feature vector is used as a weighted weight to weight each feature matrix along the channel dimension of the second cargo surface state deep channel compression feature map to obtain the second cargo surface state deep channel compression local activation feature map; the second channel compressed cargo surface state global deep activation feature map and the second cargo surface state deep channel compression local activation feature map are added according to the position to obtain the second channel compressed cargo surface state deep multi-scale fusion activation feature map; the second channel compressed cargo surface state deep multi-scale fusion activation feature map is subjected to hole convolution coding to obtain the enhanced cargo surface state deep feature map.

[0073] In a specific example, inputting the deep feature map of the surface state of the cargo into the feature multi-scale perception enhancement module to obtain the enhanced deep feature map of the surface state of the cargo includes: inputting the deep feature map of the surface state of the cargo into the feature multi-scale perception enhancement module to process the deep feature map of the surface state of the cargo using the following multi-scale perception enhancement formula to obtain the enhanced deep feature map of the surface state of the cargo; wherein the multi-scale perception enhancement formula is:

[0074]

[0075] Among them, F is the deep characteristic map of the surface state of the goods, C 3,3 and C3,2 is a 3×3 dilated convolution operation with dilation numbers of 3 and 2, C1 is a 1×1 convolution operation, Avg(·) represents the global mean pooling of each feature matrix along the channel dimension in the feature map, δ(·) is a nonlinear activation process, ⊙ represents the positional point multiplication, It means that the feature map is processed by weighted multiplication along the channel dimension using the feature vector, and F2 is the deep feature map of the surface state of the enhanced goods.

[0076] Furthermore, since the shallow feature map of the cargo surface state and the enhanced deep feature map of the cargo surface state respectively capture different levels of feature information of the cargo surface, the shallow feature map mainly captures the basic features of the cargo surface, while the deep feature map contains more abstract and advanced feature information. Therefore, in order to effectively fuse these two different levels of feature information about the cargo surface state, so as to comprehensively utilize the information at different levels to more comprehensively characterize the surface state of the cargo, in the technical solution of the present application, the shallow feature map of the cargo surface state and the enhanced deep feature map of the cargo surface state are further input into the semantic mask association fusion module to obtain the multi-scale fusion feature map of the cargo surface state. Through the processing of the semantic mask association fusion module, the correlation relationship and hidden association pattern in the shallow feature map of the cargo surface state and the enhanced deep feature map of the cargo surface state can be learned and captured, so that the feature information at different levels can be associated with each other, and the key features related to the cargo type recognition task under the association mode of the two can be identified, so as to associate and fuse the shallow and deep semantic features of the cargo surface state, and give higher weight and attention to the key features in the fusion process, so that the fused features are more representative and discriminative. This helps the system better characterize the surface state of the goods and improves the accuracy of the system's identification and classification of goods.

[0077] Accordingly, in step S540, if Figure 3As shown, the shallow feature map of the cargo surface state and the enhanced deep feature map of the cargo surface state are input into the semantic mask association fusion module to obtain a multi-scale fusion feature map of the cargo surface state, including: S541, the shallow feature map of the cargo surface state and the enhanced deep feature map of the cargo surface state are spliced ​​based on the channel dimension to obtain a shallow-deep splicing feature map of the cargo surface state; S542, the shallow-deep splicing feature map of the cargo surface state is convoluted to obtain a shallow-deep splicing hidden semantic feature map of the cargo surface state; S543, the shallow-deep splicing hidden semantic feature map of the cargo surface state is convolved using a transformation matrix and a bias matrix The graph is linearly transformed to obtain a linearly transformed shallow-deep spliced ​​hidden semantic feature map of the surface state of the goods; S544, the linearly transformed shallow-deep spliced ​​hidden semantic feature map of the surface state of the goods is activated using a softmax activation function to obtain an activated linearly transformed shallow-deep spliced ​​hidden semantic feature map of the surface state of the goods; S545, each position feature value in the activated shallow-deep spliced ​​hidden semantic feature map of the surface state of the goods is used as a threshold value weight, and the position-weighted sum of the shallow feature map of the surface state of the goods and the enhanced deep feature map of the surface state of the goods is calculated to obtain the multi-scale fusion feature map of the surface state of the goods.

[0078] Among them, in step S543, the shallow-deep splicing hidden semantic feature map of the surface state of the goods is linearly transformed using the transformation matrix and the bias matrix to obtain the linearly transformed shallow-deep splicing hidden semantic feature map of the surface state of the goods, including: calculating the matrix multiplication of each feature matrix along the channel dimension in the shallow-deep splicing hidden semantic feature map of the surface state of the goods and the transformation matrix, and then adding each feature matrix along the channel dimension in the obtained feature map with the bias matrix by position to obtain the linearly transformed shallow-deep splicing hidden semantic feature map of the surface state of the goods.

[0079] Among them, in step S545, the feature values ​​of each position in the activated linear transformation shallow-deep concatenated hidden semantic feature map of the cargo surface state are used as threshold weights, and the position-weighted sum of the shallow feature map of the cargo surface state and the enhanced deep feature map of the cargo surface state is calculated to obtain the multi-scale fusion feature map of the cargo surface state, including: calculating the product of the feature value of each position in the shallow feature map of the cargo surface state and the threshold weight to obtain a weighted shallow feature map of the cargo surface state; calculating the product of the feature value of each position in the enhanced deep feature map of the cargo surface state and one minus the threshold weight to obtain a weighted enhanced deep feature map of the cargo surface state; adding the weighted shallow feature map of the cargo surface state and the weighted enhanced deep feature map of the cargo surface state by position points to obtain the multi-scale fusion feature map of the cargo surface state.

[0080] In a specific example, the shallow feature map of the cargo surface state and the enhanced deep feature map of the cargo surface state are input into a semantic mask association fusion module to obtain a multi-scale fusion feature map of the cargo surface state, including: inputting the shallow feature map of the cargo surface state and the enhanced deep feature map of the cargo surface state into the semantic mask association fusion module to process them according to the following semantic mask association fusion formula to obtain the multi-scale fusion feature map of the cargo surface state; wherein the semantic mask association fusion formula is:

[0081] f c =t·f1+(1-t)f2

[0082]

[0083] Among them, F1 and F2 are respectively the shallow feature map of the surface state of the goods and the deep feature map of the surface state of the enhanced goods, concat(·,·) represents the feature matrix concatenation operation along the channel dimension of the feature map, W i is the transformation matrix, f N×N represents the N×N convolution kernel, b i is the bias matrix, represents matrix multiplication, softmax(·) represents the softmax activation function, T is the activated linear transformation shallow-deep correlation fusion feature map of the surface state of the goods, t is the threshold value of each position in the activated linear transformation shallow-deep correlation fusion feature map T of the surface state of the goods, f1 and f2 are the feature values ​​of each position in the shallow feature map of the surface state of the goods and the deep feature map of the enhanced surface state of the goods, respectively, c It is the feature value of each position in the multi-scale fusion feature map of the surface state of the cargo.

[0084] Then, the multi-scale fusion feature map of the surface state of the goods is input into the automatic sorting module based on the classifier to obtain the sorting result, and the sorting result is used to represent the type label of the goods to be sorted. In other words, the multi-scale and multi-level fusion features of the surface state of the goods are used for classification processing to automatically identify the type of goods for sorting, and then, based on the sorting result, the goods to be sorted are divided into corresponding capacity pools. In this way, it is possible to achieve rapid and accurate classification of goods in the air logistics warehouse management of cross-border e-commerce, and divide the goods into corresponding capacity pools. In this way, efficient and automatic sorting and packaging of goods can be carried out, which helps to achieve more efficient, reliable and sustainable air logistics management.

[0085] Accordingly, in step S550, based on the multi-scale fusion feature map of the surface state of the goods, the type label of the goods to be sorted is determined, including: inputting the multi-scale fusion feature map of the surface state of the goods into the automatic sorting module based on the classifier to obtain a sorting result, and the sorting result is used to represent the type label of the goods to be sorted. Then, in step S560, based on the type label of the goods to be sorted, the goods to be sorted are divided into corresponding capacity pools, including: based on the sorting result, the goods to be sorted are divided into corresponding capacity pools by the AGV trolley.

[0086] It should be understood that the role of the classifier is to use the given categories and known training data to learn classification rules and classifiers, and then classify (or predict) unknown data. Logistic regression and SVM are often used to solve binary classification problems. For multi-class classification problems, logistic regression or SVM can also be used, but multiple binary classifications are required to form a multi-classification. However, this is prone to errors and is not efficient. Commonly used multi-classification methods include the Softmax classification function.

[0087] Furthermore, in the technical solution of the present application, the aviation logistics management method for cross-border e-commerce also includes a training step: for training the cargo surface state feature extractor based on the bidirectional pyramid network, the feature multi-scale perception enhancement module, the semantic mask association fusion module and the classifier-based automatic sorting module.

[0088] The training step includes: obtaining training data, wherein the training data includes a training surface state image of the goods to be sorted collected by a camera and a real result representing the type label of the goods to be sorted; inputting the training surface state image of the goods to be sorted into the goods surface state feature extractor based on the bidirectional pyramid network to obtain a shallow feature map of the surface state of the training goods and a deep feature map of the surface state of the training goods; inputting the deep feature map of the surface state of the training goods into the feature multi-scale perception enhancement module to obtain a deep feature map of the surface state of the training enhanced goods; and inputting the shallow feature map of the surface state of the training goods into the feature multi-scale perception enhancement module to obtain a deep feature map of the surface state of the training enhanced goods; and inputting the shallow feature map of the surface state of the training goods into the feature multi-scale perception enhancement module to obtain a deep feature map of the surface state of the training enhanced goods. The deep feature map of the surface state of the goods is trained and enhanced and inputted into the semantic mask association fusion module to obtain a multi-scale fusion feature map of the surface state of the training goods; the multi-scale fusion feature map of the surface state of the training goods is inputted into the automatic sorting module based on the classifier to obtain a training sorting result; the cross entropy loss function value between the training sorting result and the real result is calculated as the classification loss function value; based on the classification loss function value, the cargo surface state feature extractor based on the bidirectional pyramid network, the feature multi-scale perception enhancement module, the semantic mask association fusion module and the automatic sorting module based on the classifier are trained.

[0089] In an embodiment of the present application, for the cross-scale and cross-depth image semantic feature distribution differences of the training surface state images of the goods to be sorted represented by the training goods surface state shallow feature map and the training goods surface state deep feature map respectively, after performing feature multi-scale perception enhancement through the training goods surface state deep feature map and performing semantic mask association fusion on the training goods surface state shallow feature map and the training enhanced goods surface state deep feature map, the obtained training goods surface state multi-scale fusion feature map will also have the problem of insufficient distribution aggregation caused by the differentiated image semantic feature channel change distribution, thereby affecting its classification convergence efficiency through the classifier, that is, affecting the efficiency of classification training and the accuracy of classification results.

[0090] Therefore, preferably, when the multi-scale fusion feature map of the surface state of the training goods is input into the classifier-based automatic sorting module to obtain the sorting result, the multi-scale fusion feature map of the surface state of the goods is optimized. Specifically, the multi-scale fusion feature map of the surface state of the training cargo is input into the classifier-based automatic sorting module to obtain the training sorting result, including the following optimization steps: after dot-multiplying the multi-scale fusion feature map of the surface state of the training cargo with the scale of the multi-scale fusion feature map of the surface state of the training cargo, the multi-scale fusion feature map of the surface state of the training cargo is subtracted from the sum of the absolute values ​​of each eigenvalue of the multi-scale fusion feature map of the surface state of the training cargo to obtain the multi-scale fusion full semantic feature map of the surface state of the training cargo, wherein the scale of the multi-scale fusion feature map of the surface state of the training cargo is the width of the feature matrix of the multi-scale fusion feature map of the surface state of the training cargo multi-scale fusion feature map multi-scale multi-channel number of the multi-scale fusion feature map of the surface state of the training cargo; after dot-multiplying the multi-scale fusion feature map of the surface state of the training cargo with the square root of the scale of the multi-scale fusion feature map of the surface state of the training cargo, the multi-scale fusion feature map of the surface state of the training cargo is subtracted from the square root of the sum of the squares of each eigenvalue of the multi-scale fusion feature map of the surface state of the training cargo to obtain The method comprises the following steps: obtaining a multi-scale fusion semi-semantic feature map of the surface state of training goods; calculating the square root of the absolute value of each eigenvalue of the multi-scale fusion full semantic feature map of the surface state of training goods and the multi-scale fusion semi-semantic feature map of the surface state of training goods respectively to obtain a multi-scale fusion full semantic modulation feature map of the surface state of training goods and a multi-scale fusion semi-semantic modulation feature map of the surface state of training goods; calculating the logarithm of each eigenvalue of the multi-scale fusion full semantic modulation feature map of the surface state of training goods and the multi-scale fusion semi-semantic modulation feature map of the surface state of training goods respectively to obtain a multi-scale fusion full semantic information feature map of the surface state of training goods and a multi-scale fusion semi-semantic information feature map of the surface state of training goods; and calculating the weighted sum of the multi-scale fusion full semantic information feature map of the surface state of training goods and the multi-scale fusion semi-semantic information feature map of the surface state of training goods with the weighting coefficient as a hyperparameter to obtain an optimized multi-scale fusion feature map of the surface state of training goods.

[0091] That is, the applicant of the present application takes the multi-scale fusion feature map of the surface state of the training cargo as a feature set, wherein the semantic representation of the change in the unit of the feature value of the multi-scale fusion feature map of the surface state of the training cargo is used. In order to dynamically aggregate the semantic set composed of different change semantics of the multi-scale fusion feature map of the surface state of the training cargo without ignoring the individual semantic change information, the individual features of the multi-scale fusion feature map of the surface state of the training cargo and the collective expression of its feature map scale are used as full-width and half-width feature representations, and the low-rank norms of different dimensions of the overall semantics of the feature set of the multi-scale fusion feature map of the surface state of the training cargo are used as phases through negative correlation calculation and amplitude-phase information representation modulation is performed to dynamically adjust the semantic information change relationship of the multi-scale fusion feature map of the surface state of the training cargo, thereby improving the aggregation of the overall semantic information expression of the multi-scale fusion feature map of the surface state of the training cargo, and improving the classification convergence efficiency of the multi-scale fusion feature map of the surface state of the training cargo when it is classified by the automatic sorting module based on the classifier. In this way, the classification of goods can be realized more accurately, and the goods can be divided into corresponding capacity pools, which is conducive to realizing more efficient, reliable and sustainable aviation logistics management.

[0092] Based on the above embodiments, see Figure 4 As shown, it is a structural schematic diagram of an air logistics management system 800 for cross-border e-commerce in an embodiment of the present application. The air logistics management system 800 for cross-border e-commerce includes: a surface state image acquisition module 810, which is used to obtain the surface state image of the goods to be sorted collected by the camera; a multi-level feature extraction module 820, which is used to perform multi-level feature extraction on the surface state image of the goods to be sorted through a goods surface state feature extractor to obtain a shallow feature map of the goods surface state and a deep feature map of the goods surface state; a multi-scale perception enhancement analysis module 830, which is used to perform feature multi-scale perception enhancement analysis on the deep feature map of the goods surface state to obtain an enhanced deep feature map of the goods surface state; an association fusion module 840, which is used to input the shallow feature map of the goods surface state and the enhanced deep feature map of the goods surface state into a semantic mask association fusion module to obtain a multi-scale fusion feature map of the goods surface state; a classification module 850, which is used to determine the type label of the goods to be sorted based on the multi-scale fusion feature map of the goods surface state; and a division module 860, which is used to divide the goods to be sorted into corresponding capacity pools based on the type label of the goods to be sorted.

[0093] Here, those skilled in the art can understand that the specific functions and operations of each module in the above-mentioned aviation logistics management system 800 for cross-border e-commerce have been referred to above. Figure 2 to Figure 3It has been introduced in detail in the description of the aviation logistics management method for cross-border e-commerce, and therefore, its repeated description will be omitted.

[0094] Figure 5 FIG. 1 is an application scenario diagram of the aviation logistics management method for cross-border e-commerce according to an embodiment of the present application. Figure 5 As shown, in this application scenario, first, the surface state image of the goods to be sorted collected by the camera is obtained (for example, Figure 5 D), then, the surface state image of the goods to be sorted is input to a server (for example, Figure 5 In S) as shown in , the server can use the aviation logistics management algorithm for cross-border e-commerce to process the surface state image of the goods to be sorted to obtain a sorting result for representing the type label of the goods to be sorted, and then, based on the sorting result, divide the goods to be sorted into corresponding capacity pools.

[0095] Furthermore, this application also provides an ERP intelligent operation platform. Figure 6 As shown, the ERP intelligent operation platform includes: product module, order module, warehouse module, report module and user module.

[0096] Based on the above embodiments, another exemplary embodiment of an electronic device is also provided in the embodiments of the present application. In some possible implementations, the electronic device in the embodiments of the present application may include a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor can implement the steps of the aviation logistics management method for cross-border e-commerce in the above embodiments when executing the program.

[0097] For example, in the case of electronic equipment Figure 1 Taking the server 100 in the example as an example, the processor in the electronic device is the processor 110 in the server 100, and the memory in the electronic device is the memory 120 in the server 100.

[0098] An embodiment of the present application also provides a computer-readable storage medium. Figure 7 1 shows a schematic diagram of a computer-readable storage medium 1000 according to an embodiment of the present application. Figure 7As shown, the computer-readable storage medium 1000 stores computer-executable instructions 1001. When the computer-executable instructions 1001 are executed by the processor, the aviation logistics management method for cross-border e-commerce according to the embodiment of the present application described with reference to the above figures can be executed. The computer-readable storage medium includes, but is not limited to, for example, volatile memory and / or non-volatile memory. The volatile memory may include, for example, a random access memory (RAM) and / or a cache memory (cache), etc. The non-volatile memory may include, for example, a read-only memory (ROM), a hard disk, a flash memory, etc.

[0099] The embodiment of the present application also provides a computer program product or a computer program, which includes computer executable instructions, and the computer executable instructions are stored in a computer readable storage medium. The processor of the computer device reads the computer executable instructions from the computer readable storage medium, and the processor executes the computer executable instructions, so that the computer device executes the aviation logistics management method for cross-border e-commerce according to the embodiment of the present application.

[0100] Those skilled in the art will appreciate that the contents disclosed in this application may be subject to various variations and improvements. For example, the various devices or components described above may be implemented by hardware, or by software, firmware, or a combination of some or all of the three.

[0101] In addition, although the present application makes various references to certain units in the system according to embodiments of the present application, any number of different units can be used and run on the client and / or server. The units are only illustrative, and different aspects of the system and method can use different units.

[0102] Those skilled in the art will appreciate that all or part of the steps in the above method can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, such as a read-only memory, a disk or an optical disk. Optionally, all or part of the steps in the above embodiment can also be implemented using one or more integrated circuits. Accordingly, each module / unit in the above embodiment can be implemented in the form of hardware or in the form of software function modules. The present application is not limited to any particular form of combination of hardware and software.

[0103] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those of ordinary skill in the art to which this application belongs. It should also be understood that terms such as those defined in common dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and should not be interpreted in an idealized or extremely formal sense, unless explicitly defined as such herein.

[0104] The above is an explanation of the present application and should not be considered as a limitation thereof. Although several exemplary embodiments of the present application have been described, it will be readily understood by those skilled in the art that many modifications may be made to the exemplary embodiments without departing from the novel teachings and advantages of the present application. Therefore, all of these modifications are intended to be included within the scope of the present application as defined by the claims. It should be understood that the above is an explanation of the present application and should not be considered to be limited to the specific embodiments disclosed, and modifications to the disclosed embodiments and other embodiments are intended to be included within the scope of the appended claims. The present application is defined by the claims and their equivalents.

Claims

1. A method for aviation logistics management for cross-border e-commerce, characterized in that: include: Acquire the surface state image of the goods to be sorted collected by the camera; Performing multi-level feature extraction on the surface state image of the goods to be sorted by a goods surface state feature extractor to obtain a shallow feature map of the goods surface state and a deep feature map of the goods surface state; Performing feature multi-scale perception enhancement analysis on the cargo surface state deep feature map to obtain an enhanced cargo surface state deep feature map; Inputting the shallow feature map of the cargo surface state and the enhanced deep feature map of the cargo surface state into a semantic mask association fusion module to obtain a multi-scale fusion feature map of the cargo surface state; Determining the type label of the goods to be sorted based on the multi-scale fusion feature map of the surface state of the goods; Based on the type labels of the goods to be sorted, the goods to be sorted are divided into corresponding capacity pools; The method comprises: performing multi-level feature extraction on the surface state image of the goods to be sorted by a goods surface state feature extractor to obtain a shallow feature map of the goods surface state and a deep feature map of the goods surface state, including: inputting the surface state image of the goods to be sorted into the goods surface state feature extractor based on a bidirectional pyramid network to obtain the shallow feature map of the goods surface state and the deep feature map of the goods surface state; Performing feature multi-scale perception enhancement analysis on the cargo surface state deep feature map to obtain an enhanced cargo surface state deep feature map, including: inputting the cargo surface state deep feature map into a feature multi-scale perception enhancement module to obtain the enhanced cargo surface state deep feature map; Inputting the cargo surface state deep feature map into the feature multi-scale perception enhancement module to obtain the enhanced cargo surface state deep feature map, including: In the first branch, point convolution processing is performed on the deep feature map of the surface state of the goods to obtain a first deep channel compression feature map of the surface state of the goods; global mean pooling is performed on the first deep channel compression feature map of the surface state of the goods to obtain a first deep channel compression feature vector of the surface state of the goods; nonlinear activation processing is performed on the first deep channel compression feature vector of the surface state of the goods to obtain a first deep channel local activation feature vector of the surface state of the goods; In the second branch, point convolution processing is performed on the deep feature map of the surface state of the goods to obtain a second deep channel compression feature map of the surface state of the goods; In the third branch, the deep feature map of the surface state of the goods is subjected to hole convolution coding to obtain a deep receptive field expansion feature map of the surface state of the goods; the deep receptive field expansion feature map of the surface state of the goods is subjected to point convolution processing to obtain a global deep feature matrix of the surface state of the goods with expanded receptive field; the global deep feature matrix of the surface state of the goods with expanded receptive field is subjected to nonlinear activation to obtain a global deep activation feature matrix of the surface state of the goods with expanded receptive field; Multiply the global deep activation feature matrix of the receptive field expanded cargo surface state by each feature matrix along the channel dimension of the second cargo surface state deep channel compression feature map by position points to obtain the second channel compression cargo surface state global deep activation feature map; Taking each position feature value in the local activation feature vector of the deep channel of the first cargo surface state as a weighted weight, weighting each feature matrix along the channel dimension of the deep channel compression feature map of the second cargo surface state to obtain the deep channel compression local activation feature map of the second cargo surface state; The second channel compressed cargo surface state global deep activation feature map and the second cargo surface state deep channel compressed local activation feature map are added by position to obtain the second channel compressed cargo surface state deep multi-scale fusion activation feature map; The deep multi-scale fusion activation feature map of the compressed cargo surface state of the second channel is subjected to hole convolution encoding to obtain a deep feature map of the enhanced cargo surface state.

2. The aviation logistics management method for cross-border e-commerce according to claim 1 is characterized in that: Inputting the shallow feature map of the cargo surface state and the enhanced deep feature map of the cargo surface state into a semantic mask association fusion module to obtain a multi-scale fusion feature map of the cargo surface state, including: The shallow characteristic map of the surface state of the cargo and the deep characteristic map of the surface state of the enhanced cargo are spliced ​​based on the channel dimension to obtain a shallow-deep spliced ​​characteristic map of the surface state of the cargo; Performing convolution processing on the shallow-deep splicing feature map of the surface state of the goods to obtain a shallow-deep splicing hidden semantic feature map of the surface state of the goods; Using a transformation matrix and a bias matrix, a linear transformation is performed on the shallow-deep concatenated hidden semantic feature map of the surface state of the goods to obtain a linearly transformed shallow-deep concatenated hidden semantic feature map of the surface state of the goods; use The activation function performs activation processing on the linear transformation cargo surface state shallow-deep splicing hidden semantic feature map to obtain an activated linear transformation cargo surface state shallow-deep splicing hidden semantic feature map; Taking the feature values ​​of each position in the shallow-deep concatenated hidden semantic feature map of the activated linear transformation cargo surface state as the threshold weight, the position-weighted sum of the shallow feature map of the cargo surface state and the enhanced deep feature map of the cargo surface state is calculated to obtain the multi-scale fusion feature map of the cargo surface state.

3. The aviation logistics management method for cross-border e-commerce according to claim 2 is characterized in that: The shallow-deep splicing hidden semantic feature map of the surface state of the goods is linearly transformed using a transformation matrix and a bias matrix to obtain a linearly transformed shallow-deep splicing hidden semantic feature map of the surface state of the goods, including: calculating the matrix multiplication of each feature matrix along the channel dimension in the shallow-deep splicing hidden semantic feature map of the surface state of the goods and the transformation matrix, and then adding each feature matrix along the channel dimension in the obtained feature map to the bias matrix by position to obtain the linearly transformed shallow-deep splicing hidden semantic feature map of the surface state of the goods.

4. The aviation logistics management method for cross-border e-commerce according to claim 3 is characterized in that: Taking each position feature value in the activated linear transformation cargo surface state shallow-deep splicing hidden semantic feature map as a threshold value weight, calculating the position weighted sum of the cargo surface state shallow feature map and the enhanced cargo surface state deep feature map to obtain the cargo surface state multi-scale fusion feature map, including: Calculate the product of the characteristic value of each position in the shallow characteristic map of the surface state of the goods and the threshold value weight to obtain a weighted shallow characteristic map of the surface state of the goods; Calculate the product of the feature value of each position in the enhanced cargo surface state deep feature map and one minus the threshold value weight to obtain a weighted enhanced cargo surface state deep feature map; The weighted item cargo surface state shallow feature map and the weighted item enhanced cargo surface state deep feature map are added according to the position points to obtain the cargo surface state multi-scale fusion feature map.

5. The aviation logistics management method for cross-border e-commerce according to claim 4 is characterized in that: Based on the multi-scale fusion feature map of the surface state of the goods, determining the type label of the goods to be sorted includes: The multi-scale fusion feature map of the surface state of the goods is input into an automatic sorting module based on a classifier to obtain a sorting result, and the sorting result is used to represent the type label of the goods to be sorted.

6. The aviation logistics management method for cross-border e-commerce according to claim 5, characterized in that: Based on the type labels of the goods to be sorted, the goods to be sorted are divided into corresponding capacity pools, including: based on the sorting results, the goods to be sorted are divided into corresponding capacity pools by an AGV vehicle.

7. The aviation logistics management method for cross-border e-commerce according to claim 6 is characterized in that: The method further includes a training step for training the cargo surface state feature extractor based on the bidirectional pyramid network, the feature multi-scale perception enhancement module, the semantic mask association fusion module and the classifier-based automatic sorting module; Wherein, the training step includes: Acquire training data, wherein the training data includes a training surface state image of the goods to be sorted acquired by a camera and a real result representing a type label of the goods to be sorted; Inputting the training surface state image of the goods to be sorted into the goods surface state feature extractor based on the bidirectional pyramid network to obtain a shallow feature map of the training goods surface state and a deep feature map of the training goods surface state; Inputting the training cargo surface state deep feature map into the feature multi-scale perception enhancement module to obtain a training enhanced cargo surface state deep feature map; Inputting the shallow feature map of the surface state of the training cargo and the deep feature map of the surface state of the training enhanced cargo into the semantic mask association fusion module to obtain a multi-scale fusion feature map of the surface state of the training cargo; Inputting the multi-scale fusion feature map of the surface state of the training goods into the classifier-based automatic sorting module to obtain a training sorting result; Calculating a cross entropy loss function value between the training sorting result and the true result as a classification loss function value; The cargo surface state feature extractor based on the bidirectional pyramid network, the feature multi-scale perception enhancement module, the semantic mask association fusion module and the classifier-based automatic sorting module are trained based on the classification loss function value.

Images