Method and system for agricultural machinery trajectory recognition

Abstract

A method and system for agricultural machinery trajectory recognition, relating to the technical field of agricultural application. The method comprises: on the basis of data sent by a GPS device, determining trajectory point-related data of an agricultural machine, and obtaining a corresponding satellite image; on the basis of a neural network model having a special architecture, processing and analyzing the above data, and determining image features of each trajectory point corresponding to the satellite image and classification information corresponding to each trajectory point; and marking, on the satellite image, a movement trajectory formed by connecting the trajectory points and the classification information corresponding to the trajectory points, and outputting a marked map. On the basis of the solution, the movement trajectory of an agricultural machine can be obtained and identified without manual on-site surveying, efficiently and accurately marking the movement trajectory of the agricultural machine on a satellite map, and visually presenting distribution information of farmland.

Description

Method and system for agricultural machinery trajectory recognition

[0001] This application claims priority to the Chinese patent application filed with the China Patent Office on September 1, 2023, with application number 202311127954.4 and invention name “Method and system for agricultural machinery trajectory identification”, the entire contents of which are incorporated by reference into this application. Technical Field

[0002] The present application relates to the field of agricultural application technology, and more specifically, to a method and system for agricultural machinery trajectory recognition. Background Art

[0003] In agricultural work, obtaining farmland distribution information is crucial. Typically, farmers obtain this information through on-site surveys and then mark it on a designated map for display. This method is inefficient and inaccurate.

[0004] In view of this, how to efficiently and accurately mark the distribution information of farmland on the map without manual on-site survey is a technical problem that needs to be solved urgently.

[0005] Summary of the Invention

[0006] The present application provides a method and system for identifying agricultural machinery trajectories, which can match the movement trajectory of agricultural machinery with satellite maps, analyze the farmland and road parts corresponding to the movement trajectory, and mark them on the satellite map, so as to achieve efficient and accurate distribution information of the marked farmland on the map without human intervention.

[0007] In a first aspect, a method for identifying an agricultural machinery trajectory is provided. The method includes: determining a first trajectory map based on first trajectory point information sent from a global positioning system (GPS) device carried by the agricultural machinery, the first trajectory map including a first movement trajectory formed by connecting a plurality of trajectory points, the first movement trajectory being used to represent the movement trajectory of the agricultural machinery within a first area, the first trajectory map being used to represent a relationship between the first movement trajectory and first state information, the first state information including at least one of the following: speed information, time information, and mileage information; determining an initial point sequence based on the first trajectory point information, the initial point sequence including latitude and longitude coordinates of a location of the agricultural machinery corresponding to each trajectory point, the instantaneous speed of the agricultural machinery, the mileage of the agricultural machinery, and a marking time; determining a first satellite map, the first satellite map being used to represent the first area, the first satellite map and the first trajectory map having the same size parameters and being based on the same geodetic coordinate axis; inputting the first trajectory map, the first satellite map, and the initial point sequence into a first neural network model, outputting a first labeled map, the first labeled map including an identification result for the first movement trajectory, the first neural network model including an encoder and a processor, each of the encoder and the processor including n stages, and the first neural network model being configured as follows:

[0008] The first trajectory map and the first satellite map are superimposed to determine the first input image; in the first stage of the encoder, the first input image is subjected to feature extraction at the first downsampling scale to determine the first feature map, and in the first stage of the processor, the initial point sequence is subjected to a full connection operation to determine the first point sequence, wherein the number of channels used to represent point information in the first point sequence is the same as the number of channels used to represent point information in the first feature map; and based on multi-headed self-attention (multi-headed self-attention), the first input image is subjected to feature extraction at the first downsampling scale to determine the first feature map. The self-attention (MSA) mechanism is used to perform a self-attention operation on the first feature map and the first point sequence to determine the first fusion point sequence; in the mth stage of the encoder, feature extraction is performed on the m-1th feature map at the mth downsampling scale to determine the mth feature map, and in the mth stage of the processor, a full connection operation is performed on the m-1th fusion point sequence to determine the mth point sequence, and the number of channels used to represent point information in the mth point sequence is the same as the number of channels used to represent point information in the mth feature map; and based on the MSA mechanism, a self-attention operation is performed on the mth feature map and the mth point sequence to determine the mth fusion point sequence, where m∈[2,n] is an integer; in When m is equal to n, a full connection operation is performed on the nth fused point sequence to determine a target point sequence, where the target point sequence includes the longitude and latitude coordinates of each trajectory point constituting the first movement trajectory and classification information for candidates, where the classification information for the candidates includes at least farmland and road. Using a normalized exponential function, the classification information to which each trajectory point in the target point sequence constituting the first movement trajectory belongs is determined from the classification information for the candidates. Based on the longitude and latitude coordinates of each trajectory point and the classification information to which each trajectory point belongs, the first movement trajectory and the classification information to which each trajectory point belongs are mapped to a first satellite map, and a first labeled map is output.

[0009] For example, the first trajectory image may be an image based on STD. The first trajectory image is drawn based on the marking time corresponding to each trajectory point, specifically a grayscale image. Each trajectory point constituting the first movement trajectory corresponds to a different grayscale value, and the grayscale value may be associated with the first state information.

[0010] For example, since the size parameters of the first satellite map and the first trajectory map need to be the same, and both are used to represent the same area, and need to be in the same geodetic coordinate system, the map levels and size parameters of the two maps need to be determined in advance.

[0011] For example, the above n stages are also called n stages of the encoder, which can be specifically 4 stages. In each stage, the input image needs to pass through multiple transformer processing blocks to extract image features.

[0012] Based on this technical solution, without manual on-site surveys, the movement trajectories of agricultural machinery can be efficiently and accurately identified using a neural network model based on track point information collected by GPS devices and corresponding satellite maps, and these trajectories can be marked on the satellite map. Furthermore, the movement trajectories can indicate the type of area to which each portion belongs, such as farmland or road. Therefore, based on the movement trajectories belonging to farmland, the distribution of farmland can be intuitively determined.

[0013] In combination with the first aspect, in certain implementations of the first aspect, before the first trajectory map and the first satellite map are input into the first neural network model, the first length and the first width in the size parameters of the first trajectory map and the first satellite map are expanded to the second length and the second width, respectively.

[0014] For example, the first length and the first width may be specifically expressed based on the number of pixels, and may be specifically expressed as 1024×1024; similarly, the second length and the second width may be specifically expressed as 1280×1280.

[0015] Based on the above technical solution, it can be ensured that the model can obtain sufficiently rich image edge information.

[0016] In combination with the first aspect, in certain implementations of the first aspect, before the first trajectory map and the first satellite map are superimposed to determine the first input image, a downsampling operation is performed on the first trajectory map and the first satellite map, reducing the second length and the second width in the size parameters of the first trajectory map and the first satellite map to a third length and a third width, and correspondingly increasing the number of image channels in the size parameters of the first trajectory map and the first satellite map.

[0017] For example, two convolution kernels with a stride of 2 may be used to perform downsampling operations on the first trajectory image and the first satellite map, respectively, so as to change the size parameters of the images.

[0018] Based on the above technical solution, the input image size can be reduced while the number of image features is enriched, which helps the subsequent neural network model to accurately extract input image features.

[0019] In combination with the first aspect, in certain implementations of the first aspect, the trajectory points in the first moving trajectory whose classification information is farmland are connected by lines of a first color, and the trajectory points in the first moving trajectory whose classification information is roads are connected by lines of a second color.

[0020] Based on the above technical solution, the classification of the movement trajectory can be intuitively distinguished on the satellite map, which facilitates the direct acquisition of the distribution information of the farmland, so as to facilitate the subsequent statistical analysis of the distribution information of the farmland and determine the corresponding farming plan.

[0021] In combination with the first aspect, in certain implementations of the first aspect, Q groups of data sets including a first trajectory map, a first satellite map, and an initial point sequence are sequentially input into a first neural network model, and the first neural network model correspondingly outputs Q first labeled maps, where Q is an integer greater than 1; the Q first labeled maps are spliced ​​together to determine a second labeled map.

[0022] Based on the above technical solution, while ensuring the processing efficiency of the neural network model, multiple first labeled maps are spliced ​​together to obtain a second labeled map that can fully present large-scale farmland, thereby further increasing the intuitiveness of presenting farmland distribution information.

[0023] In a second aspect, a system for identifying agricultural machinery trajectories is provided, the system comprising: a GPS server, a satellite map server, and an image processing server;

[0024] Among them, the GPS server is used to determine a first trajectory map based on the first trajectory point information sent by the GPS device carried by the agricultural machinery. The first trajectory map includes a first moving trajectory connected by multiple trajectory points. The first moving trajectory is used to represent the moving trajectory of the agricultural machinery in the first area. The first trajectory map is used to represent the relationship between the first moving trajectory and the first status information. The first status information includes at least one of the following information: speed information, time information and mileage information; and determine the initial point sequence based on the first trajectory point information. The initial point sequence includes that each trajectory point corresponds to the longitude and latitude coordinates of the location of the agricultural machinery, the instantaneous speed of the agricultural machinery, the mileage of the agricultural machinery and the marking time.

[0025] The satellite map server is used to determine a first satellite map, where the first satellite map is used to represent a first area. The first satellite map and the first trajectory map have the same size parameters and are under the same geodetic coordinate axis.

[0026] The image processing server is configured to input the first trajectory map, the first satellite map, and the initial point sequence into a first neural network model and output a first labeled map. The first neural network model includes an encoder and a processor, each of which includes n stages. The first neural network model is configured as follows:

[0027] The first trajectory map and the first satellite map are superimposed to determine the first input image; in the first stage of the encoder, the first input image is subjected to feature extraction at the first downsampling scale to determine the first feature map, and in the first stage of the processor, the initial point sequence is subjected to a full connection operation to determine the first point sequence, and the number of channels used to represent point information in the first point sequence is the same as the number of channels used to represent point information in the first feature map; and based on the MSA mechanism, a self-attention operation is performed on the first feature map and the first point sequence to determine the first fusion point sequence; in the mth stage of the encoder, the m-1th feature map is subjected to feature extraction at the mth downsampling scale to determine the mth feature map, and in the mth stage of the processor, the m-1th fusion point sequence is subjected to a full connection operation to determine the mth point sequence, and the number of channels used to represent point information in the mth point sequence is the same as the number of channels used to represent point information in the mth feature map The number of channels of the point information is the same; and based on the MSA mechanism, a self-attention operation is performed on the mth feature map and the mth point sequence to determine the mth fusion point sequence, where m∈[2,n] is an integer; when m is equal to n, a full connection operation is performed on the nth fusion point sequence to determine the target point sequence, which includes the longitude and latitude coordinates of each trajectory point constituting the first movement trajectory and classification information for candidates, where the classification information for candidates includes at least farmland and road; through a normalized exponential function, the classification information of each trajectory point constituting the first movement trajectory in the target point sequence is determined from the classification information for candidates, and according to the longitude and latitude coordinates of each trajectory point and the classification information of each trajectory point, the first movement trajectory and the classification information of each trajectory point constituting the first movement trajectory are mapped to the first satellite map, and a first labeled map is output.

[0028] In combination with the second aspect, in certain implementations of the second aspect, before the image processing server inputs the first trajectory map and the first satellite map into the first neural network model, the image processing server is further used to: expand the first length and the first width in the size parameters of the first trajectory map and the first satellite map to the second length and the second width, respectively.

[0029] In combination with the second aspect, in certain implementations of the second aspect, before the image processing server overlays the first trajectory map and the first satellite map, the image processing server is further configured to: perform a downsampling operation on the first trajectory map and the first satellite map, reduce the second length and second width in the size parameters of the first trajectory map and the first satellite map to a third length and third width, and correspondingly expand the number of image channels in the size parameters of the first trajectory map and the first satellite map.

[0030] In combination with the second aspect, in certain implementations of the second aspect, the above-mentioned image processing server is further used to: sequentially input Q groups of data sets including a first trajectory map, a first satellite map, and a first point sequence into the first neural network model, and the first neural network model correspondingly outputs Q first marked maps, where Q is an integer greater than 1; and splice the Q first marked maps to determine the second marked map.

[0031] In a third aspect, a device for identifying agricultural machinery trajectories is provided, comprising a processor and a memory, wherein the processor and the memory are connected, wherein the memory is used to store program code, and the processor is used to call the program code to execute a method in any possible implementation of the method design of the first aspect above.

[0032] In a fourth aspect, a chip system is provided, which is applied to an electronic device; the chip system includes one or more interface circuits and one or more processors; the interface circuit and the processor are interconnected through lines; the interface circuit is used to echo a signal from the memory of the electronic device and send a signal to the processor, the signal including a computer instruction stored in the memory; when the processor executes the computer instruction, the electronic device executes a method in any possible implementation of the method design of the first aspect above.

[0033] In a fifth aspect, a computer-readable storage medium is provided, storing computer instructions, which are executed by a processor to implement the method in any possible implementation manner in the method design of the first aspect.

[0034] In a sixth aspect, a computer program product is provided. When the computer program code or instructions are executed on a computer, the computer executes a method in any possible implementation of the method design of the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] FIG1 is a schematic block diagram of a system 100 for agricultural machinery trajectory recognition according to an embodiment of the present application;

[0036] FIG2 is a schematic block diagram of a method 200 for identifying agricultural machinery trajectories according to an embodiment of the present application;

[0037] FIG3 is a schematic diagram of a trajectory diagram proposed in an embodiment of the present application;

[0038] FIG4 is a schematic block diagram of a neural network model architecture 400 proposed in an embodiment of the present application;

[0039] FIG5 is a schematic block diagram of another agricultural machinery trajectory recognition system 500 provided in an embodiment of the present application. DETAILED DESCRIPTION

[0040] In the description of the embodiments of the present application, unless otherwise specified, " / " means or, for example, A / B can mean A or B; "and / or" in this article is a kind of association relationship that describes associated objects, indicating that there can be three relationships, for example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In this application, "at least one" refers to one or more, and "more than one" refers to two or more. "At least one of the following items" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b, or c can mean: a, b, c, ab, ac, bc, or abc, where a, b, c can be single or multiple.

[0041] In the embodiments of this application, prefixes such as "first" and "second" are used only to distinguish different description objects and have no limiting effect on the position, order, priority, quantity, or content of the described objects. The use of prefixes such as ordinal numbers in the embodiments of this application to distinguish description objects does not constitute a limitation on the described objects. For a statement of the described objects, please refer to the description in the context of the claims or embodiments, and the use of such prefixes should not constitute an unnecessary limitation.

[0042] The technical solutions in the embodiments of the present application will be described below with reference to the accompanying drawings.

[0043] Obtaining farmland distribution information is a crucial step in farming, influencing subsequent farming strategies. Typically, farmers obtain this information through on-site surveys, then mark it on a designated map for display. This method is inefficient and inaccurate.

[0044] As agricultural automation, informatization, and intelligent technologies advance, farming methods that require human intervention will gradually be replaced. Marking farmland distribution information on maps without the need for manual on-site surveys is also a key step in the development of agricultural technology.

[0045] In view of this, the embodiments of the present application propose a method, device and system for agricultural machinery trajectory identification. Through a transformer model configured with special functions, satellite maps and information collected by global positioning system (GPS) equipment are combined to perform feature extraction and data fusion operations, accurately and efficiently identify farmland areas and road areas in satellite maps, and mark the farmland areas and road areas on the satellite maps.

[0046] FIG1 is a schematic block diagram of a system 100 for identifying agricultural machinery trajectories according to an embodiment of the present application.

[0047] The system 100 includes: an agricultural machine 110, a GPS device 120, a GPS server 130, a satellite map server 140, an image processing server 150, and a control center 160. These devices can be connected via a wireless network to enable data exchange between the devices.

[0048] Among them, agricultural machinery 110 is machinery used in agricultural production, such as a seed drill, a drone for spreading fertilizer, or a combine harvester. These agricultural machinery have a characteristic, that is, the movement speed and movement trajectory of the agricultural machinery in the working area are different from the movement speed and movement trajectory of the agricultural machinery in the non-working area.

[0049] GPS device 120, mounted on agricultural machine 110, collects GPS data about agricultural machine 110 as it moves. This GPS data can be a sequence of points, each of which is referred to as a track point of agricultural machine 110. Each point can correspond to the following information: the longitude and latitude coordinates of the agricultural machine's location, the machine's instantaneous speed, the machine's mileage, and the time the track point was recorded. This GPS data is then uploaded to GPS server 130.

[0050] In some possible embodiments, the reporting mechanism of the GPS device 120 is as follows:

[0051] Each time a point data is collected, it is reported to the GPS server 130; or,

[0052] Based on the timer mechanism, the corresponding GPS data is reported to the GPS server 130 every time a preset time period passes; or,

[0053] Based on the motion state of the agricultural machine 110, whenever it is detected that the agricultural machine 110 has stopped moving, that is, when the longitude and latitude coordinates and mileage corresponding to the collected multiple track points are the same, or when the instantaneous speed corresponding to the multiple track points is 0, the corresponding GPS data is reported to the GPS server 130; or

[0054] Based on the mileage mechanism, each time a preset mileage is passed, the corresponding GPS data is reported to the GPS server 130; or,

[0055] Based on the data stack mechanism, each time point data is collected, the point data is stored in a data stack with a preset length. Whenever the data stack is full, the corresponding GPS data is reported to the GPS server 130.

[0056] The GPS server 130, which can be located remotely from the farming area, generates a corresponding trajectory map based on GPS data. This trajectory map includes the movement trajectory of the agricultural machine 110. The trajectory map is then sent to the satellite map server 140 and the image processing server 150. Furthermore, a trajectory point sequence corresponding to the trajectory map is generated based on the GPS data and sent to the image processing server 150.

[0057] The satellite map server 140 can also be a server located at a remote end of the farming area, used to collect a satellite map that is compatible with the above-mentioned trajectory map. The satellite map has the same size parameters as the above-mentioned trajectory map and corresponds to the same map area, and then sends the satellite map to the image processing server 150.

[0058] In some possible embodiments, the size parameters of the track map and the satellite map can be uniformly set to the same parameters by the server. In addition, the satellite map corresponding to the track map can be determined by the following method:

[0059] According to the above size parameters, the satellite map is divided into multiple tiles, and different tiles correspond to different areas. Each track point in the above track map can carry its corresponding longitude and latitude coordinates. Therefore, according to the longitude and latitude coordinates of each track point, the corresponding satellite map can be locked.

[0060] Image processing server 150, which can also be located remotely from the farming area, is configured to receive the track map and track point sequence sent by GPS server 130, as well as the satellite map sent by satellite map server 140. The server then performs corresponding digital image processing operations using a transformer model, matching the track point sequence information with the processed image features to obtain a satellite map labeled with farmland and road areas. The labeled satellite map is then transmitted to control center 160.

[0061] In some possible embodiments, the GPS server 130, satellite map server 140 and image processing server 150 may be independent and separately deployed servers, or they may be deployed together or even be a large server with integrated functions.

[0062] The control center 160 is used to perform statistical analysis of farming data based on the above-mentioned marked satellite map, and to carry out corresponding farming planning based on the statistical analysis results, for example, counting the farming operation area of ​​agricultural machinery; or determining the spatial distribution of crops; or planning the farming operation route of the agricultural machinery 110, etc.

[0063] Based on the above system 100, an embodiment of the present application proposes a method for agricultural machinery trajectory recognition.

[0064] FIG2 is a schematic block diagram of a method 200 for identifying agricultural machinery trajectories according to an embodiment of the present application.

[0065] S210: Determine a first track map according to first track point information sent by the GPS device carried by the agricultural machine.

[0066] Among them, the first trajectory map includes a first moving trajectory formed by connecting multiple trajectory points, and the first moving trajectory is used to represent the moving trajectory of the agricultural machinery in the first area. The first trajectory map is used to represent the relationship between the first moving trajectory and the first status information, and the first status information includes at least one of the following information: speed information, time information and mileage information.

[0067] In some possible embodiments, the first trajectory graph may be an image based on the standard deviation (STD) of an image. The first trajectory graph is drawn based on the time at which each trajectory point is marked, and is specifically a grayscale graph. Each trajectory point constituting the first movement trajectory corresponds to a different grayscale value, and the grayscale value may be associated with the first state information.

[0068] FIG3 is a schematic diagram of a trajectory diagram proposed in an embodiment of the present application.

[0069] Figure 3 (a) is a speed graph, in which the grayscale value of the trajectory point is associated with the speed information. The smaller the grayscale value, the faster the speed.

[0070] Figure 3 (b) is a cumulative time graph, in which the grayscale value of the trajectory point is associated with the time information. The smaller the grayscale value, the longer the cumulative time.

[0071] (c) in FIG3 is a mileage diagram, in which the grayscale value of the track point is associated with the mileage information. The smaller the grayscale value, the longer the mileage.

[0072] S220: Determine an initial point sequence based on the first trajectory point information, where each trajectory point corresponds to the longitude and latitude coordinates of the location of the agricultural machine, the instantaneous speed of the agricultural machine, the mileage of the agricultural machine, and the marking time.

[0073] It should be understood that, based on S220, since the initial point sequence is determined based on the first trajectory point information, the first trajectory point information can at least represent the information included in the initial point sequence. However, the first trajectory point information and the initial point sequence have different data formats. The former can be ordinary data information without a special data format, while the latter represents the first trajectory point information in the form of a sequence. This data format is more suitable for the neural network model required in subsequent processes, thereby facilitating the subsequent matching of trajectory points with image features.

[0074] S230: Determine a first satellite map, where the first satellite map is used to represent the first area. The first satellite map and the first trajectory map have the same size parameters and are in the same geodetic coordinate system.

[0075] Based on S230, it can be seen that the size parameters of the first satellite map and the first trajectory map need to be the same, and both are used to represent the same area. They also need to be in the same geodetic coordinate system. Therefore, the map levels and size parameters of the two maps need to be determined in advance to accurately establish the corresponding relationship between the two maps.

[0076] In some possible embodiments, an appropriate map level needs to be selected for each first trajectory map. If the level is too low, the trajectory may be too thin, losing much detail. If the level is too high, the image will be localized, failing to capture more global information. Therefore, before implementing this solution, the level of the first trajectory map can be determined through preliminary experiments. The level of the first trajectory map can be 18.

[0077] In some possible embodiments, since the first movement trajectory may be long and the size of the first trajectory image is limited, the first movement trajectory often requires multiple first trajectory images to be composed, and may even reach thousands. If an image that is too large is directly input into the neural network model, the model will run very slowly and will contain a lot of invalid information, such as a pure black background without a trajectory, which will result in a waste of computing resources. Therefore, before executing this solution, the size parameters of the first trajectory image can be determined through preliminary experiments. The size parameters can be a first length and a first width. The first length and the first width can be specifically expressed based on the number of pixels, for example, it can be 1024×1024. In addition, for each first trajectory image, pixel information can be added to its edges, and the first length and the first width in the size parameters of the first trajectory image can be expanded to the second length and the second width respectively, for example, it can be 1280×1280, and then input into the model for calculation, thereby ensuring that the model can obtain sufficiently rich image edge information.

[0078] Similar to the first trajectory map, before implementing this solution, you can determine the size parameters of the first satellite map through preliminary experiments. These size parameters are also the first length and the first width. Then, expand the first length and the first width in the size parameters of the first satellite map to the second length and the second width respectively. The specific operations are as follows:

[0079] Taking the first trajectory map with a size parameter of 1280×1280 as an example, we first set up a tile server on a satellite map of the same level as the first trajectory map. Then, we locked the corresponding tile based on the longitude and latitude coordinates of each trajectory point in the first movement trajectory. We then combined 4×4 256×256 tiles into a large 1024×1024 tile. Finally, we added pixel information to the edge of the satellite map corresponding to this large tile and expanded its size parameter to 1280×1280, ensuring that the model can obtain sufficient image edge information.

[0080] S240: Input the first trajectory map, the first satellite map, and the initial point sequence into a first neural network model, and output a first marked map.

[0081] The first neural network model includes an encoder and a processor, and the encoder and the processor each include n stages. The first neural network model is configured as follows:

[0082] S241: Superimpose the first trajectory map and the first satellite map to determine a first input image.

[0083] It should be understood that since both are in the same geodetic coordinate system, the image orientations of the first trajectory map and the first satellite map can be kept consistent.

[0084] After obtaining the first input image, the first input image is transmitted to an encoder having n stages, and the following n stages of feature extraction are performed in sequence.

[0085] S242: At each stage of the encoder, feature extraction is performed on the image at different downsampling scales to determine the mth feature map. Correspondingly, at each corresponding stage of the processor, a full connection operation is performed on the point sequence to determine the mth fusion point sequence. Based on the MSA mechanism, a self-attention operation is performed on the mth feature map and the mth point sequence to determine the mth fusion point sequence. The specific operations of S242 are as follows:

[0086] In the first stage of the encoder, feature extraction is performed on the first input image at the first downsampling scale to determine the first feature map, and in the first stage of the processor, a full connection operation is performed on the initial point sequence to determine the first point sequence, and the number of channels used to represent point information in the first point sequence is the same as the number of channels used to represent point information in the first feature map; and based on the mechanism, a self-attention operation is performed on the first feature map and the first point sequence to determine the first fused point sequence;

[0087] In the mth stage of the encoder, feature extraction is performed on the m-1th feature map at the mth downsampling scale to determine the mth feature map, and in the mth stage of the processor, a full connection operation is performed on the m-1th fusion point sequence to determine the mth point sequence, and the number of channels used to represent point information in the mth point sequence is the same as the number of channels used to represent point information in the mth feature map; and based on the MSA mechanism, a self-attention operation is performed on the mth feature map and the mth point sequence to determine the mth fusion point sequence, where m∈[2,n] is an integer;

[0088] It should be understood that the above S242 is a traversal process, in which the first input image undergoes n stages of feature extraction operations in sequence.

[0089] In some possible embodiments, the n stages are also referred to as n stages of the encoder, specifically four stages. In each stage, the input image needs to pass through multiple transformer processing blocks to extract image features. The specific downsampling operation in each stage is as follows:

[0090] Take the above first input image as an example, which is a 24-channel image.

[0091] In stage 1, two blocks are used to downsample the input image to 1 / 4, with 24 input image channels and 32 output image channels, to determine the first processing result image.

[0092] In stage 2, two blocks are used to perform a downsampling operation on the first processing result image with a scale of 1 / 8, with an input image channel of 32 and an output image channel of 64, to determine the second processing result image;

[0093] In stage 3, 6 blocks are used to downsample the second processing result image by 1 / 16, with an input image channel of 64 and an output image channel of 128, to determine the third processing result image.

[0094] In stage 4, two blocks are used to perform a downsampling operation on the third processing result image with a scale of 1 / 32, with an input image channel of 128 and an output image channel of 256, to determine the fourth processing result image, which is the first feature map mentioned above.

[0095] When the number of image channels of the first input image is another number, the number of blocks, input image channels, and output image channels corresponding to each stage may be adaptively adjusted.

[0096] It should be understood that as the stage goes deeper, the size of the output processing result image will decrease, but the number of image channels will increase to store richer information.

[0097] S243: When m is equal to n, perform a full connection operation on the nth fusion point sequence to determine the target point sequence.

[0098] The target point sequence includes the longitude and latitude coordinates of each trajectory point constituting the first moving trajectory and classification information for candidates, and the classification information for candidates includes at least farmland and road.

[0099] In some possible embodiments, when the agricultural machinery is a land-based agricultural machinery, the candidate classification information may include farmland and roads. When the agricultural machinery is an agricultural drone that spreads fertilizer in the air, the candidate classification information may include farmland and air routes. Furthermore, since trajectory points also correspond to elevation information, the above process may also take this elevation information into account, so the candidate classification information may also include mountainous areas.

[0100] S244: Determine the classification information to which each trajectory point in the target point sequence that constitutes the first movement trajectory belongs from the candidate classification information using a normalized exponential function (softmax), and map the first movement trajectory and the classification information to which each trajectory point that constitutes the first movement trajectory belongs to the first satellite map based on the longitude and latitude coordinates of each trajectory point and the classification information to which each trajectory point belongs, thereby outputting a first labeled map.

[0101] It should be understood that the candidate classification information should include at least three categories: background, road, and farmland. Since each of the above trajectory points is strongly associated with the first movement trajectory and lies on the same plane as the first movement trajectory, the background category is not discriminative and can be ignored. Therefore, the candidate classification information only needs to include at least the road and farmland categories.

[0102] It should be understood that the size parameters of the first marker map correspond to the first satellite map and the first trajectory map.

[0103] It should be understood that the above-mentioned classification information for candidates should include at least three categories: background, road and farmland. Since each of the above-mentioned trajectory points is strongly associated with the first mobile trajectory and is on the same plane as the first mobile trajectory, the background classification is not distinguishable, so this classification can be ignored. Therefore, the classification information for candidates only needs to include at least the two categories of road and farmland. Based on this, it can be seen that the recognition result for the first mobile trajectory includes the corresponding position of each trajectory point of the first mobile trajectory in the first satellite map, and the classification information to which each trajectory point belongs, so that it can be determined whether the area covered by the first mobile trajectory is farmland or road.

[0104] Based on this technical solution, without manual on-site surveys, the movement trajectories of agricultural machinery can be efficiently and accurately identified using a neural network model based on track point information collected by GPS devices and corresponding satellite maps, and these trajectories can be marked on the satellite map. Furthermore, the movement trajectories can indicate the type of area to which each portion belongs, such as farmland or road. Therefore, based on the movement trajectories belonging to farmland, the distribution of farmland can be intuitively determined.

[0105] In some possible embodiments, based on the above S241 to S244 and corresponding embodiments, the embodiment of the present application also proposes a neural network model architecture 400.

[0106] FIG4 is a schematic block diagram of a neural network model architecture 400 proposed in an embodiment of the present application.

[0107] Referring to FIG4 , it can be seen that, from a vertical perspective, the neural network model architecture 400 can include two branches: one for sequence processing and one for image processing. In each stage, the point sequence information full connection and image feature extraction functions of these two branches cooperate with each other. After the processing of n stages is completed, the two branches converge into one. From a horizontal perspective, the neural network model architecture 400 can include an input layer 410, a feature extraction layer 420 with four stages, a feature point conversion layer 430, a feature fusion layer 440, a linear conversion layer 450, and an output layer 460.

[0108] The input layer 410 is used to execute the above-mentioned S241 to input the determined input image into the feature extraction layer 420 .

[0109] The feature extraction layer 420 is used to perform the full connection operation on the point sequence in the above S242 and perform feature extraction at a specific downsampling scale on the input image.

[0110] The feature point conversion layer 430 is used to perform a full connection operation on the point sequence so that the number of channels used to represent point information in the point sequence is the same as the number of channels used to represent point information in the feature map of the same stage.

[0111] The feature fusion layer 440 is used to perform the self-attention operation on the point sequence and feature map in the above S242.

[0112] The linear transformation layer 450 is used to execute the above S243, perform a full connection operation on the fusion point sequence output after being processed by n stages, and determine a feature point set.

[0113] The output layer 460 is used to execute the above S244 to determine the classification information corresponding to each trajectory point, mark the first moving trajectory with the classification information of each trajectory point on the first satellite map, and output the first marked map.

[0114] The specific operations of each of the above layers are detailed in the corresponding embodiments above and will not be repeated here.

[0115] In some possible embodiments, the first neural network model further includes a preprocessing layer, located between the input layer 410 and the feature extraction layer 420. In this preprocessing layer, a downsampling operation is performed on the first trajectory image and the first satellite map, reducing the second length and second width in the size parameters of the first trajectory image and the first satellite map to a third length and third width, and correspondingly expanding the number of image channels in the size parameters of the first trajectory image and the first satellite map. For example, if the first trajectory image and the first satellite map are three-channel images, a convolution kernel can be used in the preprocessing layer to downsample the first trajectory image and the first satellite map, respectively, reducing the length and width of the two images and expanding the number of channels of the two images to 24.

[0116] Based on the above technical solution, the input image size can be reduced while the number of image features is enriched, which helps the subsequent neural network model to accurately extract input image features.

[0117] In some possible embodiments, the first marked map includes a first satellite map and a first movement trajectory, wherein the first movement trajectory is marked at corresponding locations on the first satellite map, and trajectory points in the first movement trajectory classified as farmland are connected by lines of a first color, and trajectory points in the first movement trajectory classified as roads are connected by lines of a second color. For example, the portion of the first movement trajectory belonging to farmland is represented by a green line, and the portion of the first movement trajectory belonging to roads is represented by a red line.

[0118] Based on the above technical solution, the classification of the movement trajectory can be intuitively distinguished on the satellite map, which facilitates the direct acquisition of the distribution information of the farmland, so as to facilitate the subsequent statistical analysis of the distribution information of the farmland and determine the corresponding farming plan.

[0119] In some possible embodiments, since the scale of farmland is usually large, the span of the movement trajectory of agricultural machinery is also correspondingly wide. Based on the above embodiments, it can be seen that in order to ensure the processing efficiency of the neural network model, the size parameters of the input image are limited. And a complete large-scale farmland is fully presented only through a trajectory map and satellite map of limited size. In view of this, a whole piece of farmland can be presented respectively through multiple trajectory maps and multiple satellite maps. Correspondingly, the above-mentioned first trajectory map, a first satellite map and a first point sequence can be called a set of data sets. In view of the above situation, it is necessary to input Q groups of data sets into the first neural network model in sequence, then the first neural network model correspondingly outputs Q first labeled maps, where Q is an integer greater than 1; and then the Q first labeled maps are spliced ​​to determine the second labeled map. The second labeled map can include a complete large-scale farmland.

[0120] Based on the above technical solution, while ensuring the processing efficiency of the neural network model, multiple first labeled maps are spliced ​​together to obtain a second labeled map that can fully present large-scale farmland, thereby further increasing the intuitiveness of presenting farmland distribution information.

[0121] In some possible embodiments, after receiving the dataset, the first neural network model may further determine the validity of the dataset by first obtaining coordinate points corresponding to the four corners of the first satellite map and determining the coordinate region corresponding to these four coordinate points, which may be represented by longitude and latitude coordinates. The model then determines whether the coordinates of each trajectory point in the first point sequence corresponding to the first trajectory map are within the aforementioned coordinate region. If all trajectory points are within the aforementioned coordinate region, the dataset is considered valid and subsequent data processing can proceed. If any trajectory points are not within the aforementioned coordinate region, the dataset is considered abnormal, and feedback information may be sent to the upstream device, instructing the upstream device to resend the dataset. Alternatively, if the number of trajectory points not within the aforementioned coordinate region is less than a tolerable number (e.g., 1 or 5), these trajectory points may be deleted from the first point sequence, and data processing may then be performed based on the dataset after the deletion of these trajectory points. If the number of trajectory points not within the aforementioned coordinate region is greater than or equal to the tolerable number, feedback information may be sent to the upstream device, instructing the upstream device to resend the dataset.

[0122] Based on the above technical solution, the first neural network model can be prevented from performing subsequent data processing procedures with relatively high computational overhead for unreasonable data sets, thereby avoiding unnecessary large computational overhead.

[0123] In some possible embodiments, after obtaining the first labeled map or the second labeled map, a corresponding agricultural task is performed based on the first labeled map or the second labeled map and a corresponding pre-trained model, which may be a neural network model. The agricultural task includes at least one of the following:

[0124] Based on the first model, count the farming area of ​​agricultural machinery; or,

[0125] determining the spatial distribution of crops in the first region based on the second model; or

[0126] Based on the third model, the farming operation route of the agricultural machinery is planned.

[0127] Based on the above technical solution, it is possible to further reduce manual participation in the above-mentioned agricultural work and realize fully automatic closed-loop management of the collection, analysis and planning of farmland distribution information.

[0128] In some possible embodiments, the performance of the first neural network model can be evaluated based on a focal loss function, and the parameters of the first neural network model can be adjusted based on the output result of the focal loss, wherein the focal loss includes a first weight and a second weight, the first weight corresponds to the roads in the classification information, and the second weight corresponds to the farmland in the classification information. After multiple experiments, it was determined that the first neural network model is easier to converge when the first weight is set to 2.5 to 3.0.

[0129] In addition, embodiments of the present application also provide a system for implementing any of the above methods. FIG5 is a schematic block diagram of another system 500 for agricultural machinery trajectory recognition provided by an embodiment of the present application.

[0130] The system 500 can be used as a subsystem of the above-mentioned system 100 . The system 500 includes: a GPS server 510 , a satellite map server 520 and an image processing server 530 .

[0131] Among them, the GPS server 510 is used to determine a first trajectory map based on the first trajectory point information sent by the GPS device carried by the agricultural machinery. The first trajectory map includes a first moving trajectory connected by multiple trajectory points. The first moving trajectory is used to represent the moving trajectory of the agricultural machinery in the first area. The first trajectory map is used to represent the relationship between the first moving trajectory and the first status information. The first status information includes at least one of the following information: speed information, time information and mileage information; and determine the initial point sequence based on the first trajectory point information. The initial point sequence includes that each trajectory point corresponds to the longitude and latitude coordinates of the location of the agricultural machinery, the instantaneous speed of the agricultural machinery, the mileage of the agricultural machinery and the marking time.

[0132] The satellite map server 520 is used to determine a first satellite map, where the first satellite map is used to represent the first area. The first satellite map and the first trajectory map have the same size parameters and are under the same geodetic coordinate axis.

[0133] The image processing server 530 is configured to input the first trajectory map, the first satellite map, and the initial point sequence into a first neural network model and output a first labeled map. The first neural network model includes an encoder and a processor, each of which includes n stages. The first neural network model is configured as follows:

[0134] Overlaying the first trajectory map and the first satellite map to determine a first input image;

[0135] In the first stage of the encoder, feature extraction is performed on the first input image at the first downsampling scale to determine the first feature map, and in the first stage of the processor, a full connection operation is performed on the initial point sequence to determine the first point sequence, and the number of channels used to represent point information in the first point sequence is the same as the number of channels used to represent point information in the first feature map; and based on the MSA mechanism, a self-attention operation is performed on the first feature map and the first point sequence to determine the first fused point sequence;

[0136] In the mth stage of the encoder, feature extraction is performed on the m-1th feature map at the mth downsampling scale to determine the mth feature map, and in the mth stage of the processor, a full connection operation is performed on the m-1th fusion point sequence to determine the mth point sequence, and the number of channels used to represent point information in the mth point sequence is the same as the number of channels used to represent point information in the mth feature map; and based on the MSA mechanism, a self-attention operation is performed on the mth feature map and the mth point sequence to determine the mth fusion point sequence, where m∈[2,n] is an integer;

[0137] When m is equal to n, a full connection operation is performed on the nth fusion point sequence to determine a target point sequence, where the target point sequence includes the longitude and latitude coordinates of each trajectory point constituting the first moving trajectory and classification information for candidates, where the classification information for candidates includes at least farmland and road;

[0138] The classification information of each trajectory point constituting the first movement trajectory in the target point sequence is determined from the candidate classification information using a normalized exponential function. The first movement trajectory and the classification information of each trajectory point constituting the first movement trajectory are mapped to a first satellite map based on the longitude and latitude coordinates of each trajectory point and the classification information to which each trajectory point belongs, thereby outputting a first labeled map.

[0139] In some possible embodiments, before the image processing server 530 inputs the first trajectory map and the first satellite map into the first neural network model, the image processing server 530 is further configured to: expand the first length and the first width in the size parameters of the first trajectory map and the first satellite map to the second length and the second width, respectively.

[0140] In some possible embodiments, before the image processing server 530 overlays the first trajectory image and the first satellite map, the image processing server 530 is further configured to: perform a downsampling operation on the first trajectory image and the first satellite map, reduce the second length and the second width in the size parameters of the first trajectory image and the first satellite map to a third length and a third width, and correspondingly expand the number of image channels in the size parameters of the first trajectory image and the first satellite map.

[0141] In some possible embodiments, the image processing server 530 is further used to: sequentially input Q groups of data sets including a first trajectory map, a first satellite map, and a first point sequence into the first neural network model, and the first neural network model correspondingly outputs Q first marked maps, where Q is an integer greater than 1; and splice the Q first marked maps to determine a second marked map.

[0142] In some possible embodiments, the embodiments of the present application also propose a device for agricultural machinery trajectory identification, which includes a processor and a memory, wherein the processor and the memory are connected, wherein the memory is used to store program code, and the processor is used to call the program code to execute the method in any possible implementation method of the above embodiments.

[0143] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0144] In the multiple embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0145] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0146] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0147] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory, a magnetic disk, or an optical disk.

[0148] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A method for identifying agricultural machinery trajectories, characterized in that: The method comprises: Determine a first trajectory map according to first trajectory point information sent by a global positioning system (GPS) device carried by the agricultural machine, wherein the first trajectory map includes a first moving trajectory formed by connecting a plurality of trajectory points, the first moving trajectory is used to represent the moving trajectory of the agricultural machine in a first area, and the first trajectory map is used to represent the relationship between the first moving trajectory and first state information, wherein the first state information includes at least one of the following information: speed information, time information, and mileage information; Determine an initial point sequence according to the first track point information, wherein each track point corresponds to the longitude and latitude coordinates of the location of the agricultural machine, the instantaneous speed of the agricultural machine, the mileage of the agricultural machine and the marking time; Determine a first satellite map, where the first satellite map is used to represent the first area, and the first satellite map and the first trajectory map have the same size parameters and are under the same geodetic coordinate axis; The first trajectory map, the first satellite map and the initial point sequence are input into a first neural network model, and a first marked map is output, wherein the first marked map includes a recognition result for the first moving trajectory, the first neural network model includes an encoder and a processor, the encoder and the processor each include n stages, and the first neural network model is configured as follows: Overlaying the first trajectory map and the first satellite map to determine a first input image; In the first stage of the encoder, feature extraction is performed on the first input image at a first downsampling scale to determine a first feature map, and in the first stage of the processor, a full connection operation is performed on the initial point sequence to determine a first point sequence, wherein the number of channels used to represent point information in the first point sequence is the same as the number of channels used to represent point information in the first feature map; and based on a multi-head self-attention MSA mechanism, a self-attention operation is performed on the first feature map and the first point sequence to determine a first fused point sequence; At the mth stage of the encoder, feature extraction is performed on the m-1th feature map at the mth downsampling scale to determine the mth feature map, and at the mth stage of the processor, a full connection operation is performed on the m-1th fusion point sequence to determine the mth point sequence, where the number of channels used to represent point information in the mth point sequence is the same as the number of channels used to represent point information in the mth feature map; and based on the MSA mechanism, a self-attention operation is performed on the mth feature map and the mth point sequence to determine the mth fusion point sequence, where m∈[2,n] is an integer; In the case where m is equal to n, a full connection operation is performed on the nth fusion point sequence to determine a target point sequence, wherein the target point sequence includes the longitude and latitude coordinates of each trajectory point constituting the first moving trajectory and classification information for candidates, wherein the classification information for candidates includes at least farmland and road; The classification information to which each trajectory point constituting the first moving trajectory in the target point sequence belongs is determined from the classification information for candidates by a normalized exponential function, and the first moving trajectory and the classification information to which each trajectory point constituting the first moving trajectory belongs are mapped to the first satellite map according to the longitude and latitude coordinates of each trajectory point and the classification information to which each trajectory point belongs, and the first marked map is output.

2. The method according to claim 1, characterized in that Before inputting the first trajectory map and the first satellite map into the first neural network model, the method further includes: The first length and the first width in the size parameters of the first trajectory map and the first satellite map are respectively extended to the second length and the second width.

3. The method according to claim 2, characterized in that Before superimposing the first trajectory map and the first satellite map to determine the first input image, the method further includes: A downsampling operation is performed on the first trajectory map and the first satellite map, and the second length and the second width in the size parameters of the first trajectory map and the first satellite map are reduced to a third length and a third width, and the number of image channels in the size parameters of the first trajectory map and the first satellite map is correspondingly expanded.

4. The method according to claim 1, characterized in that The track points whose classification information in the first moving track is farmland are connected by lines of a first color, and the track points whose classification information in the first moving track is road are connected by lines of a second color.

5. The method according to any one of claims 1 to 4, characterized in that The method further comprises: Inputting Q groups of data sets including the first trajectory map, the first satellite map and the initial point sequence into the first neural network model in sequence, the first neural network model correspondingly outputs Q first marked maps, where Q is an integer greater than 1; The Q first marked maps are concatenated to determine a second marked map.

6. A system for identifying agricultural machinery trajectories, characterized in that: The device comprises: a global positioning system GPS server, a satellite map server and an image processing server; The GPS server is used to determine a first trajectory map according to first trajectory point information sent by a GPS device carried by the agricultural machinery, the first trajectory map including a first moving trajectory connected by a plurality of trajectory points, the first moving trajectory being used to represent the moving trajectory of the agricultural machinery in a first area, the first trajectory map being used to represent the relationship between the first moving trajectory and first state information, the first state information including at least one of the following information: speed information, time information and mileage information; and determine an initial point sequence according to the first trajectory point information, the initial point sequence including the longitude and latitude coordinates of the location of the agricultural machinery corresponding to each trajectory point, the instantaneous speed of the agricultural machinery, the mileage of the agricultural machinery and the marking time; The satellite map server is used to determine a first satellite map, where the first satellite map is used to represent the first area, and the first satellite map and the first trajectory map have the same size parameters and are under the same geodetic coordinate axis; The image processing server is used to input the first trajectory map, the first satellite map and the initial point sequence into a first neural network model, and output a first marked map, wherein the first marked map includes a recognition result for the first moving trajectory, the first neural network model includes an encoder and a processor, the encoder and the processor each include n stages, and the first neural network model is configured as follows: Overlaying the first trajectory map and the first satellite map to determine a first input image; In the first stage of the encoder, feature extraction is performed on the first input image at a first downsampling scale to determine a first feature map, and in the first stage of the processor, a full connection operation is performed on the initial point sequence to determine a first point sequence, wherein the number of channels used to represent point information in the first point sequence is the same as the number of channels used to represent point information in the first feature map; and based on the MSA mechanism, a self-attention operation is performed on the first feature map and the first point sequence to determine a first fused point sequence; At the mth stage of the encoder, feature extraction is performed on the m-1th feature map at the mth downsampling scale to determine the mth feature map, and at the mth stage of the processor, a full connection operation is performed on the m-1th fusion point sequence to determine the mth point sequence, where the number of channels used to represent point information in the mth point sequence is the same as the number of channels used to represent point information in the mth feature map; and based on the MSA mechanism, a self-attention operation is performed on the mth feature map and the mth point sequence to determine the mth fusion point sequence, where m∈[2,n] is an integer; In the case where m is equal to n, a full connection operation is performed on the nth fusion point sequence to determine a target point sequence, wherein the target point sequence includes the longitude and latitude coordinates of each trajectory point constituting the first moving trajectory and classification information for candidates, wherein the classification information for candidates includes at least farmland and road; The classification information to which each trajectory point constituting the first moving trajectory in the target point sequence belongs is determined from the classification information for candidates by a normalized exponential function, and the first moving trajectory and the classification information to which each trajectory point constituting the first moving trajectory belongs are mapped to the first satellite map according to the longitude and latitude coordinates of each trajectory point and the classification information to which each trajectory point belongs, and the first marked map is output.

7. The system according to claim 6, characterized in that Before the image processing server inputs the first trajectory map and the first satellite map into the first neural network model, the image processing server is further used for: The first length and the first width in the size parameters of the first trajectory map and the first satellite map are respectively extended to the second length and the second width.

8. The system according to claim 7, characterized in that Before the image processing server superimposes the first trajectory map and the first satellite map, the image processing server is further used for: A downsampling operation is performed on the first trajectory map and the first satellite map, and the second length and the second width in the size parameters of the first trajectory map and the first satellite map are reduced to a third length and a third width, and the number of image channels in the size parameters of the first trajectory map and the first satellite map is correspondingly expanded.

9. The system according to claim 8, characterized in that The image processing server is also used for: Inputting Q groups of data sets including the first trajectory map, the first satellite map and the initial point sequence into the first neural network model in sequence, the first neural network model correspondingly outputs Q first marked maps, where Q is an integer greater than 1; The Q first marked maps are concatenated to determine a second marked map.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, which are executed by a processor to implement the method according to any one of claims 1 to 5.

Images