Crop identification method and system, storage medium and electronic equipment

Abstract

The invention relates to a crop identification method and system, a storage medium and electronic equipment. The method comprises the following steps: collecting a time sequence multispectral remote sensing image and geographic position coordinate information of a target area; processing the image to obtain a time sequence multispectral remote sensing image after initial processing; tokenizing the pre-processed image to obtain a time sequence multispectral remote sensing image after tokenizing; performing time and space feature extraction on the time sequence multispectral remote sensing image after the tokenization to obtain a time feature and a space feature of the time sequence multispectral remote sensing image; fusing the time features and the space features to obtain global time-space features; and analyzing the global time-space characteristics to obtain a crop identification result of the target area. According to the method, spatial-temporal data decoupling is realized through tokenization, the efficient feature extraction model is constructed by utilizing the spatial state model, the model has a global receptive field and linear calculation complexity due to the characteristics of the spatial state model, and the model can be placed in a spatial-temporal decoupling architecture to efficiently extract the features of the sequence.

Description

Technical Field

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

[0002] Crop fine segmentation technology aims to determine the crop category in each smallest unit of geographic space, and can be used to obtain distribution range data of different types of crops, such as rice and rapeseed. Distribution range data can help agricultural practitioners improve the utilization efficiency of agricultural resources, thereby indirectly increasing production. It can also be used to warn of the risks of natural disasters to crops, helping farmers take timely measures to deal with disasters and reduce losses. Therefore, crop type identification technology is of great significance to achieving sustainable development goals and ensuring food security.

[0003] At present, existing crop species recognition methods use deep learning to extract temporal and spatial features. For example, random forest, support vector machine, TempCNN and LSTM methods treat each single pixel in the time series remote sensing data as a sample, learn the temporal features in a single pixel, and ignore the two-dimensional spatial features. In UNet-3D, it is explored to directly feed the time series remote sensing image into the model and use the three-dimensional convolution kernel to extract features. This method uses both temporal and spatial information, but this modeling method does not decouple the temporal and spatial information and has low accuracy. Other methods combine the spatial model with the temporal model, such as combining FPN with ConvLSTM and UNet respectively. In other studies, two ViTs are stacked as the network backbone to extract the temporal and spatial features of the time series remote sensing image in turn, and good results are achieved. However, the design of extracting temporal and spatial features in turn easily degrades the features of the first dimension. In addition, although the CNN-based model is efficient, it is difficult to have a global receptive field. The ViT-based model has a global receptive field but has high computational complexity, can process small data size, and faces the problem of spatial information loss. Summary of the invention

[0004] In response to the above problems, the present application provides a crop identification method, system, storage medium and electronic device, which solves the technical problem that the related technology cannot have the characteristics of spatiotemporal data decoupling, low computational complexity and global receptive field, which makes it easy for the network to extract crop species characteristics due to insufficient spatiotemporal feature extraction, feature loss and other problems, resulting in poor effect of the method.

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

[0006] Collect time-series multispectral remote sensing images and geographic location coordinate information of the target area;

[0007] Performing image processing according to the multispectral time-series remote sensing image and the geographic location coordinate information to obtain a time-series multispectral remote sensing image after initial processing;

[0008] Performing tokenization preprocessing on the initially processed time-series multispectral remote sensing image to obtain a tokenized time-series multispectral remote sensing image;

[0009] Using a pre-established time feature extraction model to extract time features from the tokenized time series multispectral remote sensing image to obtain time features of the time series multispectral remote sensing image;

[0010] Using a pre-established spatial feature extraction model to extract spatial features from the tokenized time series multispectral remote sensing image to obtain spatial features of the time series multispectral remote sensing image;

[0011] The temporal feature and the spatial feature are integrated to obtain a global temporal and spatial feature;

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

[0013] In some embodiments, performing image processing according to the multi-spectral time-series remote sensing image and the geographic location coordinate information to obtain the time-series multi-spectral remote sensing image after initial processing includes:

[0014] By formula:

[0015]

[0016] Performing interpolation processing on the multispectral time-series remote sensing image;

[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 through the formula:

[0019]

[0020] Performing filtering processing on the multi-spectral time series remote sensing image after interpolation processing to obtain the time series multi-spectral remote sensing image after initial processing;

[0021] Among them, y′ i is the multispectral time series remote sensing image after filtering, y i+j is the value of the corresponding position in the multispectral time series remote sensing image after interpolation processing, c j are the pre-computed filter coefficients.

[0022] In some embodiments, the tokenization preprocessing of the initially processed time series multispectral remote sensing image to obtain the tokenized time series multispectral remote sensing image includes:

[0023] Using three-dimensional convolution on the time and space dimensions of the initially processed time-series multispectral remote sensing image to extract remote sensing image information in a format of a single location and a single time;

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

[0025] In some embodiments, the extracting time features of the tokenized time series multispectral remote sensing images using a pre-established time feature extraction model to obtain the time features of the time series multispectral remote sensing images includes:

[0026] Constructing a token sequence according to the tokenized time-series multispectral remote sensing image, wherein the token sequence embeds the absolute time position 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 taken as input and input into the pre-established temporal feature extraction model to obtain the temporal features of the time series multispectral remote sensing image.

[0028] In some embodiments, the use of a pre-established spatial feature extraction model to extract spatial features from the tokenized time series multispectral remote sensing image to obtain spatial features of the time series multispectral remote sensing image includes:

[0029] Constructing a token sequence according to the tokenized time-series multispectral remote sensing image, wherein the token sequence embeds the absolute spatial position 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 taken as input and input into the pre-established spatial feature extraction model to obtain the spatial features of the time series multispectral remote sensing image.

[0031] In some embodiments, the fusing of the temporal feature and the spatial feature to obtain a global temporal and spatial feature includes:

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

[0033] In some embodiments, the analyzing the global temporal and spatial features to obtain the crop identification result of the target area includes:

[0034] Reorganize the dimensions of the global time-space features to obtain a three-dimensional vector with dimensions of H×W×K;

[0035] A max operation is performed on the three-dimensional vector, and the crop corresponding to the type K with the highest probability is taken as the crop recognition result of the target area.

[0036] In a second aspect, a crop identification system is provided, the system comprising:

[0037] An acquisition unit, used to acquire time-series multispectral remote sensing images and geographic location coordinate information of the target area;

[0038] A first processing unit is used to perform image processing according to the multi-spectral time-series remote sensing image and the geographic location coordinate information to obtain a time-series multi-spectral remote sensing image after initial processing;

[0039] A second processing unit is used to perform tokenization preprocessing on the initially processed time-series multispectral remote sensing image to obtain a tokenized time-series multispectral remote sensing image;

[0040] A 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 time features of the time series multispectral remote sensing image;

[0041] A second extraction unit is used to extract 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;

[0042] A fusion unit, used for fusing the temporal feature and the spatial feature to obtain a global temporal and spatial feature;

[0043] The parsing unit is used to parse the global time-space features to obtain crop recognition results in the target area.

[0044] In a third aspect, a storage medium stores a computer program that can be executed by one or more processors and can be used to implement the crop identification method as described in the first aspect.

[0045] In a fourth aspect, an electronic device includes a memory and a processor, wherein the memory stores a computer program, the memory and the processor are communicatively connected to each other, and when the computer program is executed by the processor, the crop identification method as described in the first aspect is executed.

[0046] The present application provides a crop identification method, system, storage medium and electronic device, including: collecting time-series multispectral remote sensing images and geographic location coordinate information of a target area; performing image processing according to the multispectral time-series remote sensing images and the geographic location coordinate information to obtain an initially processed time-series multispectral remote sensing image; performing tokenization preprocessing on the initially processed time-series multispectral remote sensing image to obtain a tokenized time-series multispectral remote sensing image; using a pre-established time feature extraction model to extract time features of the tokenized time-series multispectral remote sensing image to obtain the time features of the time-series multispectral remote sensing image; using a pre-established spatial feature extraction model to extract spatial features of the tokenized time-series multispectral remote sensing image to obtain the spatial features of the time-series multispectral remote sensing image; fusing the time features and the spatial features to obtain global time-space features; parsing the global time-space features to obtain a 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 the model a global receptive field and linear computational complexity, and it can be placed in a spatiotemporal decoupling architecture to efficiently extract sequence features. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0048] Figure 1 A schematic diagram of a process flow of a crop identification method provided in an embodiment of the present application;

[0049] Figure 2 A schematic diagram of the structure of a feature extraction model provided in an embodiment of the present application;

[0050] Figure 3 The overall architecture diagram of the crop identification method provided in the embodiment of the present application;

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

[0052] Figure 5 A connection block diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0053] The following will describe the implementation methods of the present application in detail with reference to the accompanying drawings and embodiments, so that the implementation process of how the present application applies technical means to solve technical problems and achieve corresponding technical effects can be fully understood and implemented accordingly. The embodiments of the present application and the various features in the embodiments can be combined with each other without conflict, and the technical solutions formed are all within the scope of protection of the present application.

[0054] Embodiment 1

[0055] Figure 1 A schematic diagram of a method for identifying crops provided in an embodiment of the present application is shown in FIG. Figure 1 As shown, the method includes:

[0056] S101, collecting time-series multispectral remote sensing images and geographic location coordinate information of the target area;

[0057] S102, performing image processing according to the multispectral time-series remote sensing image and the geographical location coordinate information to obtain a time-series multispectral remote sensing image after initial processing;

[0058] S103, performing tokenization preprocessing on the initially processed time series multispectral remote sensing image to obtain a tokenized time series multispectral remote sensing image;

[0059] S104, 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;

[0060] S105, using a pre-established spatial feature extraction model to extract spatial features from the tokenized time series multispectral remote sensing image to obtain spatial features of the time series multispectral remote sensing image;

[0061] S106, fusing the temporal feature and the spatial feature to obtain a global temporal and spatial feature;

[0062] S107: Analyze the global temporal and spatial features to obtain crop recognition results in the target area.

[0063] It should be noted that, in order to efficiently utilize the temporal and spatial information in the time-series remote sensing images, the present invention proposes an efficient crop species identification method based on time-space decoupling, which can quickly and accurately obtain the distribution range data of different types of crops, such as rice, rapeseed, etc. The present invention includes a time-space decoupling architecture and an efficient feature extraction model. In the time-space decoupling architecture, data tokenization, time feature extraction, spatial feature extraction, and time-space feature fusion are successively performed 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 enable the model to have a global receptive field and linear computational complexity, and can be placed in the time-space decoupling architecture to efficiently extract sequence features. In addition, in order to obtain ready-to-use data, a remote sensing intelligent device is also invented, including a remote sensing image acquisition module, an image processing module, and a communication module. Among them, the acquisition module acquires the time-series multispectral remote sensing images and geographic location coordinate information of the target area, the image processing module performs interpolation and filtering on the image, and the communication module transmits the processed image data to the host computer via a local area network or Bluetooth for subsequent data preprocessing and feature extraction. Specifically, the remote sensing intelligent device may be a hardware device such as a drone.

[0064] In some embodiments, performing image processing according to the multi-spectral time-series remote sensing image and the geographic location coordinate information to obtain the time-series multi-spectral remote sensing image after initial processing includes:

[0065] By formula:

[0066]

[0067] Performing interpolation processing on the multispectral time-series remote sensing image;

[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 through the formula:

[0070]

[0071] Performing filtering processing on the multi-spectral time series remote sensing image after interpolation processing to obtain the time series multi-spectral remote sensing image after initial processing;

[0072] Among them, y′ i is the multispectral time series remote sensing image after filtering, y i+j is the value of the corresponding position in the multispectral time series remote sensing image after interpolation processing, c j are the pre-computed filter coefficients.

[0073] It should be noted that the filtering adopts the Savitzky-Golay filtering algorithm, wherein the filtering coefficients are usually calculated by the least square method.

[0074] In some embodiments, the tokenization preprocessing of the initially processed time series multispectral remote sensing image to obtain the tokenized time series multispectral remote sensing image includes:

[0075] Using three-dimensional convolution on the time and space dimensions of the initially processed time-series multispectral remote sensing image to extract remote sensing image information in a format of a single location and a single time;

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

[0077] It should be noted that the present invention preprocesses the time series multispectral remote sensing images in a tokenized manner to achieve preliminary data time-space decoupling. The time series multispectral data consists of T remote sensing images with a size of H×W and a band number of C, where T is the length of the time series and is a four-dimensional data cube. First, the original data is convolved in three dimensions with a step size of (h, w, t) in the spatial dimension. After the convolution, the original data becomes N H ×N W ×N T non-overlapping blocks of dimension C, where N H =H / h, N W =W / w,N T = T / t. These blocks are then projected onto D by a linear layer to obtain a series of data tokens of the same size and without overlap, using To represent, each data token is a 1-dimensional vector with dimension D, which contains information of a single time and a single location.

[0078] In some embodiments, the extracting time features of the tokenized time series multispectral remote sensing images using a pre-established time feature extraction model to obtain the time features of the time series multispectral remote sensing images includes:

[0079] Constructing a token sequence according to the tokenized time-series multispectral remote sensing image, wherein the token sequence embeds the absolute time position 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 taken as input and input into the pre-established temporal feature extraction model to obtain the temporal features of the time series multispectral remote sensing image.

[0081] It should be noted that, specifically, combined with the tokenization steps in the previous part, Z is first reshaped into N=N H ×N W , Z T is a sequence of N data tokens Each Z t is the time series data of a single spatial location. Then, for each Z t Embed, add time position information and learn class tokens. Construct a time position encoding sequence with the absolute day of the year (DOY) corresponding to each data token The specific operation is to first convert the DOY sequence Convert to one-hot encoding Then map the dimension to D through the linear layer to get the dimension and Z t Completely consistent temporal position encoding Finally, Z t and P t The time information is embedded in the addition method, and we get Learning token Insert to Z t In the middle, K is the number of crop categories in the task, that is,

[0082]

[0083] Each token sequence embedded in Through L consecutive feature extraction models in parallel, Z cls It contains the time features of K categories. Finally, we only keep Z in each token sequence cls , discarding the rest of the data tokens, we get Overall, we have obtained That is, the temporal characteristics of all categories at each spatial position.

[0084] The structure of the feature extraction model in the present invention is as follows Figure 2 As shown in the figure, 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 lth feature extraction model, assuming that the input is Z l-1 , Z l-1 Send them to two linear layers respectively and project them to the same dimension to get x and z. After x passes through one-dimensional convolution in two directions, SiLU activation function and spatial state model, y is obtained. forward and backward , z is obtained through the activation function z′, y forward and backwardAfter multiplying with z′ respectively and restoring the dimension through the linear layer, the final output Z is obtained l The above process can be expressed as:

[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] Among them, 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 flipping operation.

[0092] In some embodiments, the use of a pre-established spatial feature extraction model to extract spatial features from the tokenized time series multispectral remote sensing image to obtain spatial features of the time series multispectral remote sensing image includes:

[0093] Constructing a token sequence according to the tokenized time-series multispectral remote sensing image, wherein the token sequence embeds the absolute spatial position 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 taken as input and input into the pre-established spatial feature extraction model to obtain the spatial features of the time series multispectral remote sensing image.

[0095] It should be noted that, combined with the time feature extraction content in the previous part, Swapping the first two dimensions of That is, K Currently only Z is changed s The dimension of Z is the temporal feature of a category at all spatial locations. sWhen embedding spatial position information, the relative position code P s The sine and cosine functions are periodic and can express position information relatively well. In relative position coding, the sine and cosine functions are used to realize the relative expression of position, and then converted into the absolute position that can be directly calculated in the model. The specific expression is as follows:

[0096]

[0097] Among them, 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. The absolute position is expressed as follows:

[0098]

[0099] Z s and P s The spatial location information is embedded in the addition method, and the Then all the embedded learning token sequences are L consecutive feature extraction models to obtain K Each Contains temporal features of a single category enhanced by spatial features.

[0100] In some embodiments, the fusing of the temporal feature and the spatial feature to obtain a global temporal and spatial feature includes:

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

[0102] It should be noted that, specifically, the present invention utilizes the spatiotemporal residual structure to fuse the spatiotemporal features. In the spatiotemporal residual structure, Z S Skip the middle structure and Directly add to get the final global time and space characteristics

[0103] In some embodiments, the analyzing the global temporal and spatial features to obtain the crop identification result of the target area includes:

[0104] Reorganize the dimensions of the global time-space features to obtain a three-dimensional vector with dimensions of H×W×K;

[0105] A max operation is performed on the three-dimensional vector, and the crop corresponding to the type K with the highest probability is taken as the crop recognition result of the target area.

[0106] It should be noted that for the global spatiotemporal feature Z TS, split the second dimension N = N H ×N W , to restore it to four dimensions, we get Next, restore the spatial dimension to the original H×W and convert Z′ TS After normalization, the dimension D is linearly changed to h×w. Finally, its dimension is reshaped, and the dimension h×w is combined with N H ×N W Combine to get That is, the probability of each category at each pixel position. Finally, perform a max operation on Y and take the category with the highest probability at each position as the final recognition result at that position.

[0107] In summary, if Figure 3 As shown, it is the overall architecture diagram of the crop identification method provided by the embodiment of the present application. Specifically, a spatiotemporal feature extraction and fusion framework for crop species prediction is adopted. First, the input time series multispectral remote sensing image is processed, and it is converted into a feature sequence using linear projection, and a token representation in a unified format is generated through dimensional reshaping to prepare for subsequent feature extraction. In the feature extraction process, the model extracts information from two dimensions, time and space, respectively. The time feature extraction is achieved by embedding the time position information and inputting the multi-layer transformer encoder, and the spatial feature extraction is achieved by embedding the spatial position information on the basis of the time feature and completing it through another set of transformers. The entire feature extraction process is carried out in parallel, which greatly improves the processing efficiency. Subsequently, the time and space features are combined through the feature fusion module to generate the spatiotemporal joint features, and finally the recognition result of the crop species is generated through the decoding operation.

[0108] In order to more intuitively demonstrate the effect of the crop identification method of the present application, Table 1 is a comparison table of the effects of the existing methods and the method of the present application. It can be seen from Table 1 that the method of the present 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 is a comparison table of the recognition and reasoning speed and video memory usage of the existing methods and the method of the present application. It can be seen from Table 2 that the method of the present application (i.e., the TSMamba column in Table 2) is faster and occupies less video memory than other existing methods, which proves the advancement and efficiency of the present invention.

[0112]

[0113] Table 2

[0114] In summary, an embodiment of the present application provides a crop identification method, including: collecting time-series multispectral remote sensing images and geographic location coordinate information of a target area; performing image processing according to the multispectral time-series remote sensing images and the geographic location coordinate information to obtain an initially processed time-series multispectral remote sensing image; performing tokenization preprocessing on the initially processed time-series multispectral remote sensing image to obtain a tokenized time-series multispectral remote sensing image; using a pre-established time feature extraction model to extract time features from the tokenized time-series multispectral remote sensing image to obtain the time features of the time-series multispectral remote sensing image; using a pre-established spatial feature extraction model to extract spatial features from the tokenized time-series multispectral remote sensing image to obtain the spatial features of the time-series multispectral remote sensing image; fusing the time features and the spatial features to obtain global time-space features; parsing the global time-space features to obtain a crop identification result for 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 the model a global receptive field and linear computational complexity, and it can be placed in a spatiotemporal decoupling architecture to efficiently extract sequence features.

[0115] Embodiment 2

[0116] Based on the crop identification method disclosed in the above embodiment of the present invention, Figure 4 Specifically disclosed is a crop identification system applying the crop identification method.

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

[0118] The acquisition unit 401 is used to acquire time-series 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 according to the multi-spectral time-series remote sensing image and the geographic location coordinate information to obtain a time-series multi-spectral remote sensing image after initial processing;

[0120] The second processing unit 403 is used to perform tokenization preprocessing on the time series multispectral remote sensing image after the initial processing to obtain a tokenized time series multispectral remote sensing image;

[0121] A 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 to obtain 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 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;

[0123] A fusion unit 406, configured to fuse the temporal feature and the spatial feature to obtain a global temporal and spatial feature;

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

[0125] The specific working process of the collection 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 analysis unit 407 in the crop identification system disclosed in the above embodiment of the present invention can be referred to the corresponding content in the crop identification method disclosed in the above embodiment of the present invention, and will not be repeated here.

[0126] In summary, an embodiment of the present application provides a crop identification system, including: collecting time-series multispectral remote sensing images and geographic location coordinate information of a target area; performing image processing according to the multispectral time-series remote sensing images and the geographic location coordinate information to obtain an initially processed time-series multispectral remote sensing image; performing tokenization preprocessing on the initially processed time-series multispectral remote sensing image to obtain a tokenized time-series multispectral remote sensing image; using a pre-established time feature extraction model to extract time features of the tokenized time-series multispectral remote sensing image to obtain the time features of the time-series multispectral remote sensing image; using a pre-established spatial feature extraction model to extract spatial features of the tokenized time-series multispectral remote sensing image to obtain the spatial features of the time-series multispectral remote sensing image; fusing the time features and the spatial features to obtain global time-space features; parsing the global time-space features to obtain a 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 the model a global receptive field and linear computational complexity, and it can be placed in a spatiotemporal decoupling architecture to efficiently extract sequence features.

[0127] Embodiment 3

[0128] This embodiment also provides a computer-readable storage medium, such as a flash memory, a hard disk, a multimedia card, a card-type memory (for example, an SD or DX memory, etc.), a random access memory (RAM), a static random access memory (SRAM), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a magnetic memory, a disk, an optical disk, a server, an App application store, etc., on which a computer program is stored. When the computer program is executed by a processor, the method steps of the first embodiment can be implemented, and this embodiment will not be repeated here.

[0129] Embodiment 4

[0130] Figure 5 A connection block diagram of an electronic device 500 provided in an embodiment of the present application, such as 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 in Embodiment 1. The memory 502 is used to store various types of data, which may include instructions of any application or method in the electronic device, and data related to the application.

[0132] The processor 501 can be 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 the above-mentioned embodiment 1.

[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, disk or optical disk.

[0134] The multimedia component 503 may include a screen and an audio component, the screen may be a touch screen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone, and the microphone is used to receive external audio signals. The received audio signal may be further stored in a memory or sent via a communication component. The audio component also includes at least one speaker for outputting audio signals.

[0135] The I / O interface 504 provides an interface between the processor 501 and other interface models, which may be a keyboard, a mouse, buttons, etc. These buttons may be virtual buttons or physical buttons.

[0136] The communication component 505 is used for wired or wireless communication between the electronic device 500 and other devices. Wireless communication, such as Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G or 4G, or a combination of one or more of them, so the corresponding communication component 505 can include: Wi-Fi model, Bluetooth model, NFC model.

[0137] In summary, the present application provides a crop identification method, system, storage medium and electronic device, which method includes: collecting time-series multispectral remote sensing images and geographic location coordinate information of the target area; performing image processing according to the multispectral time-series remote sensing images and the geographic location coordinate information to obtain the time-series multispectral remote sensing images after initial processing; performing tokenization preprocessing on the time-series multispectral remote sensing images after initial processing to obtain the tokenized time-series multispectral remote sensing images; using a pre-established time feature extraction model to extract time features of the tokenized time-series multispectral remote sensing images to obtain the time features of the time-series multispectral remote sensing images; using a pre-established spatial feature extraction model to extract spatial features of the tokenized time-series multispectral remote sensing images to obtain the spatial features of the time-series multispectral remote sensing images; fusing the time features and the spatial features to obtain global time-space features; parsing the global time-space features to obtain the crop identification results 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 the model a global receptive field and linear computational complexity, and it can be placed in a spatiotemporal decoupling architecture to efficiently extract sequence features.

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

[0139] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device including the element.

[0140] Although the implementation methods disclosed in this application are as above, the above contents are only the implementation methods adopted for facilitating the understanding of this application, and are not intended to limit this application. Any technician in the technical field to which this application belongs can make any modifications and changes in the form and details of the implementation without departing from the spirit and scope disclosed in this application, but the scope of patent protection of this application shall still be based on the scope defined in the attached claims.

Claims

1. A method for identifying crops, characterized in that: The method comprises: Collect time-series multispectral remote sensing images and geographic location coordinate information of the target area; Performing image processing according to the multispectral time-series remote sensing image and the geographic location coordinate information to obtain a time-series multispectral remote sensing image after initial processing; Performing tokenization preprocessing on the initially processed time-series multispectral remote sensing image to obtain a tokenized time-series multispectral remote sensing image; Using a pre-established time feature extraction model to extract time features from the tokenized time series multispectral remote sensing image to obtain time features of the time series multispectral remote sensing image; Using a pre-established spatial feature extraction model to extract spatial features from the tokenized time series multispectral remote sensing image to obtain spatial features of the time series multispectral remote sensing image; The temporal feature and the spatial feature are integrated to obtain a global temporal and spatial feature; The global temporal and spatial features are analyzed to obtain crop identification results in the target area.

2. The method according to claim 1, characterized in that The performing of image processing according to the multi-spectral time-series remote sensing image and the geographic location coordinate information to obtain the time-series multi-spectral remote sensing image after initial processing includes: By formula: Performing interpolation processing on the multispectral time-series remote sensing image; Among them, (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 through the formula: Performing filtering processing on the multi-spectral time series remote sensing image after interpolation processing to obtain the time series multi-spectral remote sensing image after initial processing; Among them, y′ i is the multispectral time series remote sensing image after filtering, y i+j is the value of the corresponding position in the multispectral time series remote sensing image after interpolation processing, c j are the pre-computed filter coefficients.

3. The method according to claim 1, characterized in that The tokenization preprocessing of the initially processed time series multispectral remote sensing image to obtain a tokenized time series multispectral remote sensing image comprises: Using three-dimensional convolution on the time and space dimensions of the initially processed time-series multispectral remote sensing image to extract remote sensing image information in a format of a single location and a single time; The remote sensing image information in the format of a single position and a single time is projected to a uniform dimension through a linear layer to obtain remote sensing image data tokens of equal size and non-overlapping, that is, tokenized time-series multispectral remote sensing images.

4. The method according to claim 3, characterized in that The method of extracting time features from the tokenized time series multispectral remote sensing images using a pre-established time feature extraction model to obtain time features of the time series multispectral remote sensing images includes: Constructing a token sequence according to the tokenized time-series multispectral remote sensing image, wherein the token sequence embeds the absolute time position information corresponding to each remote sensing image data token and K learning class tokens, where K is the number of crop types; The constructed token sequence is taken as input and input into the pre-established temporal feature extraction model to obtain the temporal features of the time series multispectral remote sensing image.

5. The method according to claim 3, characterized in that: The method of 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 includes: Constructing a token sequence according to the tokenized time-series multispectral remote sensing image, wherein the token sequence embeds the absolute spatial position information corresponding to each remote sensing image data token and K learning class tokens, where K is the number of crop types; The constructed token sequence is taken as input and input into the pre-established spatial feature extraction model to obtain the spatial features of the time series multispectral remote sensing image.

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

7. The method according to claim 1, characterized in that The analyzing the global temporal and spatial features to obtain the crop identification result of the target area includes: Reorganize the dimensions of the global time-space features to obtain a three-dimensional vector with dimensions of H×W×K; A max operation is performed on the three-dimensional vector, and the crop corresponding to the type K with the highest probability is taken as the crop recognition result of the target area.

8. A crop identification system, characterized in that: The system comprises: An acquisition unit, used to acquire time-series multispectral remote sensing images and geographic location coordinate information of the target area; A first processing unit is used to perform image processing according to the multi-spectral time-series remote sensing image and the geographic location coordinate information to obtain a time-series multi-spectral remote sensing image after initial processing; A second processing unit is used to perform tokenization preprocessing on the initially processed time-series multispectral remote sensing image to obtain a tokenized time-series multispectral remote sensing image; A 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 time features of the time series multispectral remote sensing image; A second extraction unit is used to extract 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; A fusion unit, used for fusing the temporal feature and the spatial feature to obtain a global temporal and spatial feature; The parsing unit is used to parse the global time-space features to obtain crop recognition results in the target area.

9. 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 7.

10. An electronic device, characterized in that: The invention comprises a memory and a processor, wherein a computer program is stored in the memory, and the memory and the processor are connected to communicate with each other. When the computer program is executed by the processor, the crop identification method according to any one of claims 1 to 7 is executed.

Images