A crop recognition method, system, storage medium and electronic device

Abstract

This application relates to a crop identification method, system, storage medium, and electronic device, comprising: acquiring a temporal multispectral remote sensing image and geographic coordinate information of a target area; processing the image to obtain a pre-processed temporal multispectral remote sensing image; tokenizing the pre-processed image to obtain a tokenized temporal multispectral remote sensing image; extracting temporal and spatial features from the tokenized temporal multispectral remote sensing image to obtain temporal and spatial features of the temporal multispectral remote sensing image; fusing the temporal and spatial features to obtain global temporal and spatial features; and parsing the global temporal and spatial features to obtain the crop identification result of the target area. This application achieves spatiotemporal data decoupling through tokenization and constructs an efficient feature extraction model using a spatial state model. The characteristics of the spatial state model give it a global receptive field and linear computational complexity, allowing it to be placed within a spatiotemporal decoupling architecture for efficient feature extraction of sequences.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a crop identification method, system, storage medium, and electronic device. Background Technology

[0002] Crop fine segmentation technology aims to identify crop categories within each smallest unit of geographic space. It can be used to obtain distribution data for different crop types, such as rice and rapeseed. This distribution data can help agricultural practitioners improve the efficiency of agricultural resource utilization, thereby indirectly increasing yields. It can also be used to provide early warnings of the risks posed to crops by natural disasters, helping farmers take timely measures to cope with disasters and reduce losses. Therefore, crop identification technology is of great significance for achieving sustainable development goals and ensuring food security.

[0003] Current crop species identification methods use deep learning to extract temporal and spatial features. Methods such as Random Forest, Support Vector Machine, TempCNN, and LSTM treat each individual pixel in time-series remote sensing data as a sample, learning the temporal features of a single pixel while ignoring two-dimensional spatial features. UNet-3D explored directly feeding time-series remote sensing images into the model and using 3D convolutional kernels to extract features. This method utilizes both temporal and spatial information, but it doesn't decouple spatiotemporal information, resulting in lower accuracy. Other methods combine spatial and temporal models, such as combining FPN with ConvLSTM and UNet respectively. Some studies stack two ViTs as network backbones, extracting temporal and spatial features from time-series remote sensing images sequentially, achieving good results. However, the sequential extraction of temporal and spatial features can easily lead to feature degradation in the first dimension. Furthermore, while CNN-based models are efficient, they struggle to possess a global receptive field, while ViT-based models have a global receptive field but suffer from high computational complexity, limited data processing capacity, and the problem of spatial information loss. Summary of the Invention

[0004] To address the aforementioned issues, this application provides a crop identification method, system, storage medium, and electronic device. It solves the technical problem that related technologies cannot simultaneously possess the characteristics of spatiotemporal data decoupling, low computational complexity, and a global receptive field, which easily leads to insufficient spatiotemporal feature extraction and feature loss when the network extracts crop type features, resulting in poor method performance.

[0005] In a first aspect, this application provides a method for identifying crops, the method comprising:

[0006] Acquire temporal multispectral remote sensing images and geographic coordinate information of the target area;

[0007] Image processing is performed based on the multispectral time-series remote sensing image and the geographic location coordinate information to obtain a pre-processed time-series multispectral remote sensing image.

[0008] The preprocessed temporal multispectral remote sensing image is subjected to tokenization preprocessing to obtain a tokenized temporal multispectral remote sensing image.

[0009] A pre-established temporal feature extraction model is used to extract temporal features from the tokenized temporal multispectral remote sensing image to obtain the temporal features of the temporal multispectral remote sensing image.

[0010] A pre-established spatial feature extraction model is used to extract spatial features from the tokenized temporal multispectral remote sensing image to obtain the spatial features of the temporal multispectral remote sensing image.

[0011] The temporal features and spatial features are fused to obtain global temporal-spatial features;

[0012] The global temporal and spatial features are analyzed to obtain the crop identification results for the target area.

[0013] In some embodiments, the step of performing image processing based on the multispectral time-series remote sensing image and the geographic location coordinate information to obtain a pre-processed time-series multispectral remote sensing image includes:

[0014] Through the formula:

[0015]

[0016] The multispectral time-series remote sensing image is subjected to interpolation processing;

[0017] Where (x1, y1) and (x2, y2) are known points, x is the independent variable to be interpolated, and y is the estimated value of x;

[0018] Then, using the formula:

[0019]

[0020] The interpolated multispectral time-series remote sensing image is filtered to obtain the initial processed time-series multispectral remote sensing image.

[0021] Where, y′ i It is a multispectral time-series remote sensing image after filtering. i+j It is the value at the corresponding position in the interpolated multispectral time-series remote sensing image, c j These are the pre-calculated filter coefficients.

[0022] In some embodiments, the step of performing tokenization preprocessing on the initially processed temporal multispectral remote sensing image to obtain a tokenized temporal multispectral remote sensing image includes:

[0023] The pre-processed temporal multispectral remote sensing image is subjected to three-dimensional convolution in both time and space dimensions to extract remote sensing image information in a single location and single time format.

[0024] The remote sensing image information in the format of a single location and a single time is projected onto a unified dimension through a linear layer to obtain remote sensing image data tokens of equal size and non-overlapping, that is, tokenized temporal multispectral remote sensing images.

[0025] In some embodiments, the step of extracting temporal features from the tokenized temporal multispectral remote sensing image using a pre-established temporal feature extraction model to obtain the temporal features of the temporal multispectral remote sensing image includes:

[0026] A token sequence is constructed based on the tokenized temporal multispectral remote sensing image. The token sequence embeds the absolute time location information corresponding to each remote sensing image data token and K learning class tokens, where K is the number of crop types.

[0027] The constructed token sequence is used as input to a pre-established temporal feature extraction model to obtain the temporal features of the temporal multispectral remote sensing image.

[0028] In some embodiments, the step of extracting spatial features from the tokenized temporal multispectral remote sensing image using a pre-established spatial feature extraction model to obtain the spatial features of the temporal multispectral remote sensing image includes:

[0029] A token sequence is constructed based on the tokenized temporal multispectral remote sensing image. The token sequence embeds the absolute spatial location information corresponding to each remote sensing image data token and K learning class tokens, where K is the number of crop types.

[0030] The constructed token sequence is used as input to a pre-established spatial feature extraction model to obtain the spatial features of the temporal multispectral remote sensing image.

[0031] In some embodiments, fusing the temporal features and the spatial features to obtain global temporal-spatial features includes:

[0032] The temporal and spatial features are fused using a pre-designed spatiotemporal residual structure to obtain global spatiotemporal features.

[0033] In some embodiments, parsing the global spatiotemporal features to obtain the crop identification result of the target region includes:

[0034] The global temporal and spatial features are reorganized to obtain a three-dimensional vector with dimensions H×W×K;

[0035] Perform a max operation on the three-dimensional vector, and take the crop corresponding to the type K with the highest probability as the crop identification result of the target area.

[0036] Secondly, a crop identification system, the system comprising:

[0037] The acquisition unit is used to acquire temporal multispectral remote sensing images and geographic coordinate information of the target area;

[0038] The first processing unit is used to perform image processing based on the multispectral time-series remote sensing image and the geographic location coordinate information to obtain a pre-processed time-series multispectral remote sensing image.

[0039] The second processing unit is used to perform tokenization preprocessing on the initially processed temporal multispectral remote sensing image to obtain a tokenized temporal multispectral remote sensing image.

[0040] The first extraction unit is used to extract time features from the tokenized time-series multispectral remote sensing image using a pre-established time feature extraction model, so as to obtain the time features of the time-series multispectral remote sensing image.

[0041] The second extraction unit is used to extract spatial features from the tokenized temporal multispectral remote sensing image using a pre-established spatial feature extraction model, so as to obtain the spatial features of the temporal multispectral remote sensing image.

[0042] A fusion unit is used to fuse the temporal features and the spatial features to obtain global temporal-spatial features;

[0043] The parsing unit is used to parse the global temporal and spatial features to obtain the crop identification results of the target area.

[0044] Thirdly, a storage medium storing a computer program that can be executed by one or more processors to implement the crop identification method as described in the first aspect above.

[0045] Fourthly, an electronic device includes a memory and a processor, wherein a computer program is stored on the memory, and the memory and the processor are communicatively connected to each other, wherein when the computer program is executed by the processor, it performs the crop identification method as described in the first aspect above.

[0046] This application provides a crop identification method, system, storage medium, and electronic device, comprising: acquiring a time-series multispectral remote sensing image and geographic location coordinate information of a target area; performing image processing based on the multispectral time-series remote sensing image and the geographic location coordinate information to obtain a pre-processed time-series multispectral remote sensing image; performing tokenization preprocessing on the pre-processed time-series multispectral remote sensing image to obtain a tokenized time-series multispectral remote sensing image; extracting time features from the tokenized time-series multispectral remote sensing image using a pre-established time feature extraction model to obtain time features of the time-series multispectral remote sensing image; extracting spatial features from the tokenized time-series multispectral remote sensing image using a pre-established spatial feature extraction model to obtain spatial features of the time-series multispectral remote sensing image; fusing the time features and the spatial features to obtain global time-space features; and parsing the global time-space features to obtain the crop identification result of the target area. This application achieves spatiotemporal data decoupling through tokenization and constructs an efficient feature extraction model using a spatial state model. The characteristics of the spatial state model give it a global receptive field and linear computational complexity, which can be placed in a spatiotemporal decoupling architecture to efficiently extract features from sequences. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0048] Figure 1 A flowchart illustrating a crop identification method provided in this application embodiment;

[0049] Figure 2 This is a schematic diagram of the structure of the feature extraction model provided in the embodiments of this application;

[0050] Figure 3 This is a general architecture diagram of the crop identification method provided in the embodiments of this application;

[0051] Figure 4 This is a schematic diagram of the structure of a crop identification system provided in an embodiment of this application;

[0052] Figure 5 This is a connection block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0053] The following detailed description of the embodiments of this application, in conjunction with the accompanying drawings, will provide a thorough understanding of how this application uses technical means to solve technical problems and achieve corresponding technical effects, enabling its implementation. The embodiments of this application and the various features within them can be combined with each other without conflict, and all resulting technical solutions are within the protection scope of this application.

[0054] Example 1

[0055] Figure 1 This is a flowchart illustrating a crop identification method provided in an embodiment of this application, as shown below. Figure 1 As shown, this method includes:

[0056] S101. Acquire temporal multispectral remote sensing images and geographic coordinate information of the target area;

[0057] S102. Perform image processing based on the multispectral time-series remote sensing image and the geographic location coordinate information to obtain a pre-processed time-series multispectral remote sensing image.

[0058] S103. Perform tokenization preprocessing on the preprocessed temporal multispectral remote sensing image to obtain a tokenized temporal multispectral remote sensing image.

[0059] S104. The time features of the tokenized time-series multispectral remote sensing image are extracted using a pre-established time feature extraction model to obtain the time features of the time-series multispectral remote sensing image.

[0060] S105. Using a pre-established spatial feature extraction model, spatial features are extracted from the tokenized temporal multispectral remote sensing image to obtain the spatial features of the temporal multispectral remote sensing image.

[0061] S106. The temporal features and spatial features are fused to obtain global temporal-spatial features;

[0062] S107. Analyze the global temporal and spatial features to obtain the crop identification results of the target area.

[0063] It should be noted that, in order to efficiently utilize the temporal and spatial information in time-series remote sensing images, this invention proposes an efficient crop species identification method based on spatiotemporal decoupling, which can quickly and accurately obtain the distribution range data of different types of crops, such as rice and rapeseed. This invention includes a spatiotemporal decoupling architecture and an efficient feature extraction model. In the spatiotemporal decoupling architecture, data tokenization, temporal feature extraction, spatial feature extraction, and spatiotemporal feature fusion are performed sequentially to finally obtain a prediction. Secondly, an efficient feature extraction model is constructed using a spatial state model. The characteristics of the spatial state model give it a global receptive field and linear computational complexity, allowing it to be placed within the spatiotemporal decoupling architecture to efficiently extract features from sequences. Furthermore, to obtain readily usable data, a remote sensing intelligent device is also invented, including a remote sensing image acquisition module, an image processing module, and a communication module. The acquisition module acquires time-series multispectral remote sensing images and geographic coordinate information of the target area. The image processing module performs interpolation and filtering on the images. The communication module transmits the processed image data to a host computer via a local area network or Bluetooth for subsequent data preprocessing and feature extraction. Specifically, remote sensing intelligent devices can be hardware devices such as drones.

[0064] In some embodiments, the step of performing image processing based on the multispectral time-series remote sensing image and the geographic location coordinate information to obtain a pre-processed time-series multispectral remote sensing image includes:

[0065] Through the formula:

[0066]

[0067] The multispectral time-series remote sensing image is subjected to interpolation processing;

[0068] Where (x1, y1) and (x2, y2) are known points, x is the independent variable to be interpolated, and y is the estimated value of x;

[0069] Then, using the formula:

[0070]

[0071] The interpolated multispectral time-series remote sensing image is filtered to obtain the initial processed time-series multispectral remote sensing image.

[0072] Where, y′ i It is a multispectral time-series remote sensing image after filtering. i+j It is the value at the corresponding position in the interpolated multispectral time-series remote sensing image, c j These are the pre-calculated filter coefficients.

[0073] It should be noted that the filtering uses the Savitzky-Golay filtering algorithm, where the filter coefficients are usually calculated using the least squares method.

[0074] In some embodiments, the step of performing tokenization preprocessing on the initially processed temporal multispectral remote sensing image to obtain a tokenized temporal multispectral remote sensing image includes:

[0075] The pre-processed temporal multispectral remote sensing image is subjected to three-dimensional convolution in both time and space dimensions to extract remote sensing image information in a single location and single time format.

[0076] The remote sensing image information in the format of a single location and a single time is projected onto a unified dimension through a linear layer to obtain remote sensing image data tokens of equal size and non-overlapping, that is, tokenized temporal multispectral remote sensing images.

[0077] It should be noted that this invention preprocesses temporal multispectral remote sensing images using tokenization to achieve preliminary spatiotemporal decoupling of the data. The temporal multispectral data consists of T remote sensing images of size H×W with C bands, where T is the time series length, forming a four-dimensional data cube. First, the original data is convolved in three dimensions with a stride (h, w, t) in the spatial dimension. After convolution, the original data becomes N... H ×N W ×N T N non-overlapping blocks of dimension C, where N H =H / h, N W =W / w, N T =T / t After these blocks are linearly projected onto D, a series of non-overlapping data tokens of the same size are obtained, which are then used... This can be represented as a vector, where each data token is a 1-dimensional vector of dimension D, containing information about a single time and a single location.

[0078] In some embodiments, the step of extracting temporal features from the tokenized temporal multispectral remote sensing image using a pre-established temporal feature extraction model to obtain the temporal features of the temporal multispectral remote sensing image includes:

[0079] A token sequence is constructed based on the tokenized temporal multispectral remote sensing image. The token sequence embeds the absolute time location information corresponding to each remote sensing image data token and K learning class tokens, where K is the number of crop types.

[0080] The constructed token sequence is used as input to a pre-established temporal feature extraction model to obtain the temporal features of the temporal multispectral remote sensing image.

[0081] It should be noted that, specifically, combining the tokenization steps in the previous section, Z is first reshaped into... N = N H ×N W Z T For N data token sequences The combination of Z. Each Z t It is time-series data from a single spatial location. Subsequently, for each Z... t Embedding is performed, adding time location information and learning tokens. A time location encoding sequence is constructed using the absolute number of days in a year (DOY) corresponding to each data token. The specific operation is to first process the DOY sequence. Transformed into one-hot encoding Then, by mapping the dimension to D dimensions through a linear layer, we obtain the dimension and Z. t Completely consistent time position coding Finally, Z t and P t By embedding time information through addition, we obtain Learning tokens Insert into Z t At the very center, K represents the number of crop categories in the task, thus yielding:

[0082]

[0083] Each embedded token sequence has been completed. Z passes through L consecutive feature extraction models in parallel. cls This includes time features across K categories. Finally, we retain only Z from each token sequence. cls Discard the remaining data tokens and obtain Overall, it was obtained That is, the temporal characteristics of all categories at each spatial location.

[0084] The structure of the feature extraction model in this invention is as follows: Figure 2 As shown, the temporal feature extraction model has the same structure as the spatial feature extraction model. In practical applications, multiple feature extraction models are usually stacked. In the l-th feature extraction model, the input is assumed to be Z. l-1 , will Z l-1 The data are fed into two linear layers and projected to the same dimension to obtain x and z. x is then processed by one-dimensional convolution in two directions, SiLU activation function, and spatial state model to obtain y. forward and y backward z is activated by an activation function to obtain z′, y forward and y backwardAfter multiplying by z′ respectively, and then restoring the dimensions through a linear layer, the final output Z is obtained. l The above process can be expressed by the following formula:

[0085] x = Linear x (Norm(Z l-1 ))

[0086] z = Linear z (Norm(Z l-1 ))

[0087] y backward =SSM(Conv1d(SiLU(x)))

[0088] y forward =SSM(Conv1d(SiLU(flip(x))))

[0089] z′=SiLU(z)

[0090] Z l =Linear Z (y backward ⊙z′+y forward ⊙z′)+Z l-1

[0091] Where Norm is the normalization operation, Linear is the linear layer, SSM is the spatial state model, Conv1d is the one-dimensional convolution operation, and flip is the flip operation.

[0092] In some embodiments, the step of extracting spatial features from the tokenized temporal multispectral remote sensing image using a pre-established spatial feature extraction model to obtain the spatial features of the temporal multispectral remote sensing image includes:

[0093] A token sequence is constructed based on the tokenized temporal multispectral remote sensing image. The token sequence embeds the absolute spatial location information corresponding to each remote sensing image data token and K learning class tokens, where K is the number of crop types.

[0094] The constructed token sequence is used as input to a pre-established spatial feature extraction model to obtain the spatial features of the temporal multispectral remote sensing image.

[0095] It should be noted that, combined with the time feature extraction content in the previous section, The first two dimensions were swapped to obtain That is, K Currently, only Z has been changed. s The dimension of Z is also the content of a category's temporal features across all spatial locations. Subsequently, embedding is performed, for each Z... sWhen embedding spatial location information, the relative position code P is used. s Sine and cosine functions are periodic and can effectively represent positional information. In relative position encoding, sine and cosine functions are used to express the relative nature of position, which is then converted into an absolute position that can be directly calculated in the model. The specific formula is as follows:

[0096]

[0097] Where pos represents the position of an element in the sequence, k is the relative position to pos, i represents the index of the dimension, and D is the total number of dimensions. Using absolute positional form, it can be represented as follows:

[0098]

[0099] With Z s and P s Spatial location information is embedded by addition to obtain Then, extract the features from all L consecutive embedded learned token sequences to obtain K features. Each It includes temporal features that are spatially enhanced for a single category.

[0100] In some embodiments, fusing the temporal features and the spatial features to obtain global temporal-spatial features includes:

[0101] The temporal and spatial features are fused using a pre-designed spatiotemporal residual structure to obtain global spatiotemporal features.

[0102] It should be noted that, specifically, this invention utilizes a spatiotemporal residual structure to fuse spatiotemporal features. In the spatiotemporal residual structure, Z... S Skip the intermediate structure and By directly adding them together, we obtain the final global time-space features.

[0103] In some embodiments, parsing the global spatiotemporal features to obtain the crop identification result of the target region includes:

[0104] The global temporal and spatial features are reorganized to obtain a three-dimensional vector with dimensions H×W×K;

[0105] Perform a max operation on the three-dimensional vector, and take the crop corresponding to the type K with the highest probability as the crop identification result of the target area.

[0106] It should be noted that for the global spatiotemporal feature Z TSSplit the second dimension into N=N H ×N W To restore it to four dimensions, we get Next, restore the spatial dimensions to the original H×W, and then set Z′... TS After normalization, the dimension D is linearly transformed into h×w. Finally, its dimension is reshaped, and the dimension h×w is compared with N. H ×N W Combined, we obtain This refers to the probability of each category at each pixel location. Finally, a max operation is performed on Y, and the category with the highest probability at each location is taken as the final recognition result for that location.

[0107] In summary, such as Figure 3 The diagram shows the overall architecture of the crop identification method provided in this application embodiment. Specifically, it employs a spatiotemporal feature extraction and fusion framework for crop type prediction. First, the input temporal multispectral remote sensing image is processed, transformed into a feature sequence using linear projection, and then reshaped into a unified token representation to prepare for subsequent feature extraction. During feature extraction, the model extracts information from both temporal and spatial dimensions. Temporal feature extraction is achieved by embedding temporal location information and inputting it into a multi-layer transformer encoder, while spatial feature extraction embeds spatial location information based on temporal features and is completed using another set of transformers. The entire feature extraction process is performed in parallel, significantly improving processing efficiency. Subsequently, the temporal and spatial features are combined through a feature fusion module to generate spatiotemporal joint features, and finally, a decoding operation is performed to generate the crop type identification result.

[0108] To more intuitively demonstrate the effectiveness of the crop identification method of this application, Table 1 is a comparison table of the effectiveness of existing methods and the method of this application. As can be seen from Table 1, the method of this application (i.e., the Ours column in Table 1) has more accurate identification results than other existing methods.

[0109]

[0110] Table 1

[0111] Table 2 compares the recognition and reasoning speed and memory usage of existing methods and the method of this application. As can be seen from Table 2, the method of this application (i.e., the TSMamba column in Table 2) is faster and uses less memory than other existing methods, which proves the advancement and efficiency of this invention.

[0112]

[0113] Table 2

[0114] In summary, this application provides a crop identification method, comprising: acquiring a temporal multispectral remote sensing image and geographic coordinate information of a target area; performing image processing based on the multispectral temporal remote sensing image and the geographic coordinate information to obtain a pre-processed temporal multispectral remote sensing image; performing tokenization preprocessing on the pre-processed temporal multispectral remote sensing image to obtain a tokenized temporal multispectral remote sensing image; extracting temporal features from the tokenized temporal multispectral remote sensing image using a pre-established temporal feature extraction model to obtain temporal features of the temporal multispectral remote sensing image; extracting spatial features from the tokenized temporal multispectral remote sensing image using a pre-established spatial feature extraction model to obtain spatial features of the temporal multispectral remote sensing image; fusing the temporal features and the spatial features to obtain global temporal and spatial features; and parsing the global temporal and spatial features to obtain the crop identification result of the target area. This application achieves spatiotemporal data decoupling through tokenization and constructs an efficient feature extraction model using a spatial state model. The characteristics of the spatial state model give it a global receptive field and linear computational complexity, which can be placed in a spatiotemporal decoupling architecture to efficiently extract features from sequences.

[0115] Example 2

[0116] Based on the crop identification method disclosed in the above embodiments of the present invention, Figure 4 Specifically, a crop identification system that applies this crop identification method is disclosed.

[0117] like Figure 4 As shown in the figure, an embodiment of the present invention discloses a crop identification system, the system comprising:

[0118] Acquisition unit 401 is used to acquire temporal multispectral remote sensing images and geographic location coordinate information of the target area;

[0119] The first processing unit 402 is used to perform image processing based on the multispectral time-series remote sensing image and the geographic location coordinate information to obtain a pre-processed time-series multispectral remote sensing image.

[0120] The second processing unit 403 is used to perform tokenization preprocessing on the initially processed temporal multispectral remote sensing image to obtain a tokenized temporal multispectral remote sensing image.

[0121] The first extraction unit 404 is used to extract time features from the tokenized time-series multispectral remote sensing image using a pre-established time feature extraction model, so as to obtain the time features of the time-series multispectral remote sensing image.

[0122] The second extraction unit 405 is used to extract spatial features from the tokenized temporal multispectral remote sensing image using a pre-established spatial feature extraction model, so as to obtain the spatial features of the temporal multispectral remote sensing image.

[0123] The fusion unit 406 is used to fuse the temporal features and the spatial features to obtain global temporal and spatial features;

[0124] The parsing unit 407 is used to parse the global temporal and spatial features to obtain the crop identification results of the target area.

[0125] The specific working process of the acquisition unit 401, the first processing unit 402, the second processing unit 403, the first extraction unit 404, the second extraction unit 405, the fusion unit 406, and the parsing unit 407 in the crop identification system disclosed in the above embodiments of the present invention can be found in the corresponding content of the crop identification method disclosed in the above embodiments of the present invention, and will not be repeated here.

[0126] In summary, this application provides a crop identification system, comprising: acquiring a temporal multispectral remote sensing image and geographic coordinate information of a target area; performing image processing based on the multispectral temporal remote sensing image and the geographic coordinate information to obtain a pre-processed temporal multispectral remote sensing image; performing tokenization preprocessing on the pre-processed temporal multispectral remote sensing image to obtain a tokenized temporal multispectral remote sensing image; extracting temporal features from the tokenized temporal multispectral remote sensing image using a pre-established temporal feature extraction model to obtain temporal features of the temporal multispectral remote sensing image; extracting spatial features from the tokenized temporal multispectral remote sensing image using a pre-established spatial feature extraction model to obtain spatial features of the temporal multispectral remote sensing image; fusing the temporal features and the spatial features to obtain global temporal and spatial features; and parsing the global temporal and spatial features to obtain the crop identification result of the target area. This application achieves spatiotemporal data decoupling through tokenization and constructs an efficient feature extraction model using a spatial state model. The characteristics of the spatial state model give it a global receptive field and linear computational complexity, which can be placed in a spatiotemporal decoupling architecture to efficiently extract features from sequences.

[0127] Example 3

[0128] This embodiment also provides a computer-readable storage medium, such as flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, server, App application store, etc., which stores a computer program. When the computer program is executed by a processor, it can implement the method steps as described in Embodiment 1. This embodiment will not repeat the description here.

[0129] Example 4

[0130] Figure 5 A connection block diagram of an electronic device 500 provided in this application embodiment is shown below. Figure 5 As shown, the electronic device 500 may include: a processor 501, a memory 502, a multimedia component 503, an input / output (I / O) interface 504, and a communication component 505.

[0131] The processor 501 is used to execute all or part of the steps in the crop identification method as described in Embodiment 1. The memory 502 is used to store various types of data, which may include, for example, instructions for any application or method in the electronic device, as well as application-related data.

[0132] The processor 501 may be implemented as an application-specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field-programmable gate array (FPGA), a controller, a microcontroller, a microprocessor, or other electronic components, and is used to execute the crop identification method in Embodiment 1 above.

[0133] The memory 502 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0134] Multimedia component 503 may include a screen, which may be a touchscreen, and an audio component for outputting and / or inputting audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signals may be further stored in memory or transmitted via a communication component. The audio component also includes at least one speaker for outputting audio signals.

[0135] I / O interface 504 provides an interface between processor 501 and other interface models, such as keyboards, mice, buttons, etc. These buttons can be virtual buttons or physical buttons.

[0136] Communication component 505 is used for wired or wireless communication between the electronic device 500 and other devices. Wireless communication includes, for example, Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, or 4G, or one or more combinations thereof. Therefore, the corresponding communication component 505 may include: Wi-Fi model, Bluetooth model, NFC model.

[0137] In summary, this application provides a crop identification method, system, storage medium, and electronic device. The method includes: acquiring a temporal multispectral remote sensing image and geographic location coordinate information of a target area; performing image processing based on the multispectral temporal remote sensing image and the geographic location coordinate information to obtain a pre-processed temporal multispectral remote sensing image; performing tokenization preprocessing on the pre-processed temporal multispectral remote sensing image to obtain a tokenized temporal multispectral remote sensing image; extracting temporal features from the tokenized temporal multispectral remote sensing image using a pre-established temporal feature extraction model to obtain temporal features of the temporal multispectral remote sensing image; extracting spatial features from the tokenized temporal multispectral remote sensing image using a pre-established spatial feature extraction model to obtain spatial features of the temporal multispectral remote sensing image; fusing the temporal features and the spatial features to obtain global temporal and spatial features; and parsing the global temporal and spatial features to obtain the crop identification result of the target area. This application achieves spatiotemporal data decoupling through tokenization and constructs an efficient feature extraction model using a spatial state model. The characteristics of the spatial state model give it a global receptive field and linear computational complexity, which can be placed in a spatiotemporal decoupling architecture to efficiently extract features from sequences.

[0138] In the several embodiments provided in this application, it should be understood that the disclosed methods can also be implemented in other ways. The method embodiments described above are merely illustrative.

[0139] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0140] Although the embodiments disclosed in this application are as described above, the above content is merely for the purpose of facilitating understanding of this application and is not intended to limit this application. Any person skilled in the art to which this application pertains may make any modifications and changes in the form and details of the implementation without departing from the spirit and scope disclosed in this application; however, the scope of patent protection of this application shall still be determined by the scope defined in the appended claims.

Claims

1. A method for identifying crops, characterized in that, The method includes: Acquire temporal multispectral remote sensing images and geographic coordinate information of the target area; Image processing is performed based on the multispectral time-series remote sensing image and the geographic location coordinate information to obtain a pre-processed time-series multispectral remote sensing image. The preprocessed temporal multispectral remote sensing image is preprocessed with tokenization to obtain a tokenized temporal multispectral remote sensing image. Specifically, this includes: applying three-dimensional convolution to the preprocessed temporal multispectral remote sensing image in both time and space dimensions to extract remote sensing image information in a single-location, single-time format; projecting the single-location, single-time remote sensing image information onto a unified dimension through a linear layer to obtain equal-sized and non-overlapping remote sensing image data tokens, i.e., the tokenized temporal multispectral remote sensing image. A pre-established temporal feature extraction model is used to extract temporal features from the tokenized temporal multispectral remote sensing image to obtain the temporal features of the temporal multispectral remote sensing image. Specifically, this includes: constructing a token sequence based on the tokenized temporal multispectral remote sensing image, wherein the token sequence embeds the absolute time location information corresponding to each remote sensing image data token and K learned class tokens, where K is the number of crop types; and inputting the constructed token sequence into the pre-established temporal feature extraction model to obtain the temporal features of the temporal multispectral remote sensing image. A pre-established spatial feature extraction model is used to extract spatial features from the tokenized temporal multispectral remote sensing image to obtain the spatial features of the temporal multispectral remote sensing image. Specifically, this includes: constructing a token sequence based on the tokenized temporal multispectral remote sensing image, wherein the token sequence embeds the absolute spatial location information corresponding to each remote sensing image data token and K learned class tokens, where K is the number of crop types; and using the constructed token sequence as input to the pre-established spatial feature extraction model to obtain the spatial features of the temporal multispectral remote sensing image. The temporal features and spatial features are fused to obtain global temporal-spatial features; The global temporal and spatial features are analyzed to obtain the crop identification results for the target area.

2. The method according to claim 1, characterized in that, The step of image processing based on the multispectral time-series remote sensing image and the geographic location coordinate information to obtain a pre-processed time-series multispectral remote sensing image includes: Through the formula: The multispectral time-series remote sensing image is subjected to interpolation processing; Where (x1,y1) and (x2,y2) are known points, x is the independent variable to be interpolated, and y is the estimated value of x; Then, using the formula: The interpolated multispectral time-series remote sensing image is filtered to obtain the initial processed time-series multispectral remote sensing image. Where, y′ i It is a multispectral time-series remote sensing image after filtering. i+j It is the value at the corresponding position in the interpolated multispectral time-series remote sensing image, c j These are the pre-calculated filter coefficients.

3. The method according to claim 1, characterized in that, The process of fusing the temporal features and the spatial features to obtain global temporal-spatial features includes: The temporal and spatial features are fused using a pre-designed spatiotemporal residual structure to obtain global spatiotemporal features.

4. The method according to claim 1, characterized in that, The step of parsing the global spatiotemporal features to obtain the crop identification result of the target region includes: The global temporal and spatial features are reorganized to obtain a three-dimensional vector with dimensions H×W×K; Perform a max operation on the three-dimensional vector, and take the crop corresponding to the type K with the highest probability as the crop identification result of the target area.

5. A crop identification system, characterized in that, The system includes: The acquisition unit is used to acquire temporal multispectral remote sensing images and geographic coordinate information of the target area; The first processing unit is used to perform image processing based on the multispectral time-series remote sensing image and the geographic location coordinate information to obtain a pre-processed time-series multispectral remote sensing image. The second processing unit is used to perform tokenization preprocessing on the preprocessed temporal multispectral remote sensing image to obtain a tokenized temporal multispectral remote sensing image. Specifically, it includes: performing three-dimensional convolution on the preprocessed temporal multispectral remote sensing image in the time and spatial dimensions to extract remote sensing image information in the format of a single location and a single time; projecting the remote sensing image information in the format of a single location and a single time onto a unified dimension through a linear layer to obtain remote sensing image data tokens of equal size and non-overlapping, i.e., the tokenized temporal multispectral remote sensing image. The first extraction unit is used to extract time features from the tokenized time-series multispectral remote sensing image using a pre-established time feature extraction model to obtain the time features of the time-series multispectral remote sensing image. Specifically, it includes: constructing a token sequence based on the tokenized time-series multispectral remote sensing image, wherein the token sequence embeds the absolute time location information corresponding to each remote sensing image data token and K learning class tokens, where K is the number of crop types; and inputting the constructed token sequence as input to the pre-established time feature extraction model to obtain the time features of the time-series multispectral remote sensing image. The second extraction unit is used to extract spatial features from the tokenized temporal multispectral remote sensing image using a pre-established spatial feature extraction model to obtain the spatial features of the temporal multispectral remote sensing image. Specifically, it includes: constructing a token sequence based on the tokenized temporal multispectral remote sensing image, wherein the token sequence embeds the absolute spatial location information corresponding to each remote sensing image data token and K learning class tokens, where K is the number of crop types; and inputting the constructed token sequence as input to the pre-established spatial feature extraction model to obtain the spatial features of the temporal multispectral remote sensing image. A fusion unit is used to fuse the temporal features and the spatial features to obtain global temporal-spatial features; The parsing unit is used to parse the global temporal and spatial features to obtain the crop identification results of the target area.

6. A storage medium, characterized in that, The computer program stored in the storage medium can be executed by one or more processors and can be used to implement the crop identification method as described in any one of claims 1 to 4.

7. An electronic device, characterized in that, It includes a memory and a processor, wherein a computer program is stored in the memory, and the memory and the processor are communicatively connected to each other. When the computer program is executed by the processor, it performs the crop identification method as described in any one of claims 1 to 4.

Images