A method for identifying idle farmland based on LSTM autoencoder

Through the LSTM autoencoder method, the identification of idle farmland is solved, and the traditional remote sensing image recognition method in image quality and seasonal changes are achieved, and more efficient and accurate identification of idle farmland landland is achieved.

CN118628925BActive Publication Date: 2024-11-22AGRI INFORMATION INST OF CHINESE ACAD OF AGRI SCI
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
CN202410705819.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-06-03
Publication Date
2024-11-22
Estimated Expiration
2044-06-03

AI Technical Summary

Technical Problem

Traditional remote sensing image recognition methods are affected by factors such as image quality, seasonal changes and surface coverage types when identifying idle farmland, which makes it difficult to improve the recognition accuracy and efficiency.

Method used

Using the LSTM autoencoder method, the multi-spectral remote sensing image and remote sensing image data are preprocessed, monthly aggregated, superpixel segmentation and generation and adversarial network feature extraction are constructed to construct arable land-idle farmland, idle farmland-idle farmland, idle farmland, idle farmland The LSTM autoencoder model of three models of arable land-cooked land is optimized to improve identification accuracy.

Benefits of technology

It improves the accuracy and efficiency of identifying idle farmland, realizes automatic identification and timing analysis of idle farmland, adapts to the identification needs of different standards, and has a certain universality.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118628925B_ABST
    Figure CN118628925B_ABST
Patent Text Reader

Abstract

The present invention discloses a method for identifying idle cultivated land based on LSTM autoencoder, comprising collecting time series data of a preset area, preprocessing the time series data, aggregating the time series multispectral remote sensing images month by month and taking the median to obtain a time period aggregated time series HLS image, aggregating the time series remote sensing images month by month and taking the low value to obtain a time period aggregated time series SAR image, dividing the land cover into three modes according to the cultivated land features of the time period aggregated time series HLS image, constructing LSTM autoencoder models according to the three modes, optimizing the LSTM autoencoder model, inputting the vegetation index of the idle cultivated land into the LSTM autoencoder model to determine the residual distribution threshold, inputting the data to be identified into the LSTM autoencoder model, and outputting the identification result. The method can not only improve the accuracy of idle cultivated land identification based on LSTM autoencoder, but also has good interpretability, and can be directly applied to the idle cultivated land identification system.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of image recognition, and in particular to a method for identifying idle farmland based on an LSTM autoencoder. Background Art

[0002] Cultivated land is an important basic resource for agricultural production, while idle cultivated land will cause great waste to agricultural production and land resource utilization. With the continuous development of remote sensing technology, the use of remote sensing images to identify idle cultivated land has become an efficient and fast method. However, traditional remote sensing image recognition methods are often affected by multiple factors such as image quality, seasonal changes, and diversity of land cover types, making it difficult to improve recognition accuracy and efficiency. Therefore, a method that can more accurately and efficiently identify idle cultivated land is needed.

[0003] In recent years, deep learning technology has made significant progress in the field of image processing. Among them, long short-term memory network, as a special recurrent neural network, has shown strong ability in processing time series data. Autoencoder is an unsupervised deep learning model that can learn the inherent laws and feature representations of data from input data. Therefore, LSTM and autoencoder are combined to construct an unsupervised learning model that can process time series data for idle farmland identification. Summary of the invention

[0004] The purpose of the present invention is to provide a method for identifying idle farmland based on LSTM autoencoder.

[0005] To achieve the above object, the present invention is implemented according to the following technical solutions:

[0006] The present invention comprises the following steps:

[0007] Collecting time series data of a preset area and preprocessing the time series data; the time series data includes multispectral remote sensing images and remote sensing image data;

[0008] Aggregating the time series multispectral remote sensing images on a monthly basis and taking the median to obtain a time period aggregated time series HLS image, and aggregating the time series remote sensing images on a monthly basis and taking the low value to obtain a time period aggregated time series SAR image; including:

[0009] The superpixel segmentation algorithm is used to segment the time series multispectral remote sensing image to obtain regional superpixel data;

[0010] Generative adversarial networks are used to extract regional image features of regional superpixel data and calculate the cultivated land information degree of regional image features:

[0011]

[0012] The cultivated land information of the i-th region image feature is The image feature of the i-th region is β i , the i-th cultivated land feature is ζ i , regional image feature β i The probability of occurrence is (β i ), cultivated land characteristicsζ i The probability of occurrence is g(ζ i ), regional image feature β i and cultivated land characteristics i The probability of simultaneous occurrence is g(β i ,ζ i ), regional image feature β i The correlation with cultivated land is γ1, the coefficient of variation is σ, and the number of regional image features is m;

[0013] The regional cultivated land information degree is summed up, and the regional superpixel data with cultivated land information degree greater than 0.417 is regarded as cultivated land and marked as valid data, otherwise it is non-cultivated land and marked as invalid data;

[0014] Sort the valid data according to the time sequence, extract the pixel data of the time sequence valid data, sort the pixel data and extract the pixel median;

[0015] Aggregate the effective time series data of pixel median values ​​by quarter, and output the aggregation results as time period aggregated time series HLS images;

[0016] The cultivated land characteristics are calculated according to the aggregated time series HLS images in the time period, and the land cover is divided into three modes, and LSTM autoencoder models are respectively constructed according to the three modes; the three models are cultivated land-idle cultivated land, idle cultivated land-idle cultivated land, and idle cultivated land-cultivated land; the cultivated land characteristics include vegetation index, enhanced vegetation index, soil index, LSWI, VH backscatter coefficient, and VV backscatter coefficient;

[0017] Optimizing the LSTM autoencoder model, inputting the vegetation index of the idle cultivated land into the LSTM autoencoder model to determine the residual distribution threshold;

[0018] The data to be identified is input into the LSTM autoencoder model to obtain a reconstructed residual, and the identification result is output according to the reconstructed residual and the residual distribution threshold.

[0019] Furthermore, the method of aggregating the time series remote sensing images month by month and taking the lowest value to obtain the time period aggregated time series SAR images includes:

[0020] The pixel data is sorted and the lowest pixel value is extracted, and the pixel data of the effective time series data is the lowest pixel value output time period aggregated time series SAR image.

[0021] Furthermore, the method of constructing LSTM autoencoder models according to the three modes respectively includes:

[0022] The LSTM autoencoder model is composed of a decoder and an encoder. The decoder contains two LSTM layers, and the encoder contains two LSTM layers and one Dense layer.

[0023] The three modes of cultivated land-idle cultivated land, idle cultivated land-idle cultivated land, and idle cultivated land-cultivated land are respectively input into the LSTM autoencoder model, and the converter model is used to extract the change features of the three modes respectively to obtain the first feature group, the second feature group, and the third feature group;

[0024] Encode the input data, the expression is:

[0025]

[0026] The transposition is T, and the encoding sequence of the u-th input data is The u-th query feature vector is k u , the u-th change eigenvector is The vector of the u-th eigenvalue is δ u , the number of input data is q, and the change value of the change feature is The u-th deviation vector is S u , query feature vector k u The kernel function is d(k u ), the changing feature vector The kernel function is

[0027] Calculate the similarity of the changed features:

[0028]

[0029] The length of the encoding sequence is M, the L2 norm function is ‖v‖2, and the similarity of the u-th change feature is The query feature vector at the sth moment is k u,s , initial time s o The query feature vector is

[0030] Construct the loss function of the LSTM autoencoder model, the expression is:

[0031]

[0032] The loss function is The coefficient of the balance function is α, and the characteristic vector of the s-th moment change is Initial time s o The changing eigenvector of The first loss weight is The second loss weight is

[0033] Furthermore, the method for optimizing the LSTM autoencoder model includes:

[0034] Compute the average gradient:

[0035]

[0036] The average gradient of the tth iteration is The model parameters for the t-1th iteration are The number of random subsamples is The ath subsample is v a , model parameters The advance gradient is The loss function is ε(v);

[0037] Compute the L2 norm of the average gradient:

[0038]

[0039] The L2 norm of the average gradient is The sth element of the average gradient is The sth regulatory factor is C s , calculate the historical gradient norm:

[0040]

[0041] The historical gradient norm of the tth iteration is The weight parameter is Calculate the correction gradient:

[0042]

[0043] The correction gradient of the tth iteration is When the historical gradient norm is greater than the L2 norm of the average gradient, update the correction gradient:

[0044]

[0045] The updated t-th iteration correction gradient is The hereditary coefficient is γ, the deviation constant is p, and the first-order momentum and second-order momentum are calculated:

[0046]

[0047] The first-order momentum of the tth iteration is y t , the first-order momentum y at the t-1th iteration t-1, the first exponential decay rate is η1, and the second-order momentum of the tth iteration is h t , the second-order momentum of the t-1th iteration is h t-1 , the second exponential decay rate is η2, correcting the first-order momentum and the second-order momentum:

[0048]

[0049] The first exponential decay rate of the tth iteration is The second exponential decay rate of the tth iteration is The corrected first-order momentum is The corrected second-order momentum is

[0050] Update model parameters:

[0051]

[0052] The model parameters for the tth iteration are The learning rate is ρ, and the constant of change is

[0053] Furthermore, the method of inputting the vegetation index of the idle cultivated land into the LSTM autoencoder model to determine the residual distribution threshold comprises:

[0054] By obtaining the predicted residual distribution of the LSTM autoencoder model, the residual distribution of the 90% confidence space is output as the residual distribution threshold.

[0055] Furthermore, the method of outputting the recognition result according to the reconstruction residual and the residual distribution threshold comprises:

[0056] If the reconstruction residual is less than the residual distribution threshold, it is idle farmland; if the reconstruction residual is greater than T, it is non-idle farmland;

[0057] When judged as idle cultivated land: when the reconstruction residual of the first LSTM autoencoder model is less than the first residual distribution threshold, it is cultivated land-idle cultivated land; when the reconstruction residual of the second LSTM autoencoder model is less than the second residual distribution threshold, it is cultivated land-idle cultivated land; when the reconstruction residual of the third LSTM autoencoder model is less than the third residual distribution threshold, it is cultivated land-idle cultivated land.

[0058] In a second aspect, an embodiment of the present application further provides an electronic device, including:

[0059] A processor; and a memory arranged to store computer executable instructions, which when executed cause the processor to perform the method steps described in the first aspect.

[0060] In a third aspect, an embodiment of the present application further provides a computer-readable storage medium, which stores one or more programs. When the one or more programs are executed by an electronic device including multiple applications, the electronic device executes the method steps described in the first aspect.

[0061] The beneficial effects of the present invention are:

[0062] The present invention is a method for identifying idle farmland based on LSTM autoencoder. Compared with the prior art, the present invention has the following technical effects:

[0063] The present invention can improve the accuracy of multiple idle cultivated land identification through preprocessing, obtaining time period aggregated time series HLS images, obtaining time period aggregated time series SAR images, calculating cultivated land characteristics, building a model, obtaining predicted residual distribution, determining residual distribution thresholds and obtaining identification results, thereby improving the precision of idle cultivated land identification, optimizing idle cultivated land identification, greatly saving resources and improving work efficiency, and can realize automatic identification of idle cultivated land of LSTM autoencoder, and perform time series analysis on idle cultivated land identification of LSTM autoencoder in real time, which is of great significance to idle cultivated land identification, can adapt to idle cultivated land identification of different standards and different idle cultivated land identification needs, and has certain universality. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 A flowchart of a method for identifying idle farmland based on an LSTM autoencoder according to the present invention;

[0065] Figure 2 It is a schematic diagram of the structure of an electronic device in an embodiment of this specification. DETAILED DESCRIPTION

[0066] The present invention is further described below by means of specific embodiments. The illustrative embodiments and descriptions of the present invention are used to explain the present invention but are not intended to limit the present invention.

[0067] The present invention provides a method for identifying idle farmland based on LSTM autoencoder, comprising the following steps:

[0068] like Figure 1 As shown, in this embodiment, the following steps are included:

[0069] Collecting time series data of a preset area and preprocessing the time series data; the time series data includes multispectral remote sensing images and remote sensing image data; the preprocessing includes correction, denoising, filtering, edge enhancement, registration, and sharpening;

[0070] In the actual assessment, the basic sampling process is to conduct sampling through visual interpretation of high-resolution remote sensing images for two consecutive years. The sampling categories are cultivated land-idle cultivated land, idle cultivated land-idle cultivated land, and idle cultivated land-cultivated land, and each type of samples is not less than 5,000;

[0071] Aggregating the time series multispectral remote sensing images on a monthly basis and taking the median to obtain a time period aggregated time series HLS image, and aggregating the time series remote sensing images on a monthly basis and taking the low value to obtain a time period aggregated time series SAR image; including:

[0072] The superpixel segmentation algorithm is used to segment the time series multispectral remote sensing image to obtain regional superpixel data;

[0073] Generative adversarial networks are used to extract regional image features of regional superpixel data and calculate the cultivated land information degree of regional image features:

[0074]

[0075] The cultivated land information of the i-th region image feature is The image feature of the i-th region is β i , the i-th cultivated land feature is ζ i , regional image feature β i The probability of occurrence is (β i ), cultivated land characteristicsζ i The probability of occurrence is g(ζ i ), regional image feature β i and cultivated land characteristics i The probability of simultaneous occurrence is g(β i ,ζ i ), regional image feature β i The correlation with cultivated land is γ1, the coefficient of variation is σ, and the number of regional image features is m;

[0076] The regional cultivated land information degree is summed up, and the regional superpixel data with cultivated land information degree greater than 0.417 is regarded as cultivated land and marked as valid data, otherwise it is non-cultivated land and marked as invalid data;

[0077] Sort the valid data according to the time sequence, extract the pixel data of the time sequence valid data, sort the pixel data and extract the pixel median;

[0078] Aggregate the effective time series data of pixel median values ​​by quarter, and output the aggregation results as time period aggregated time series HLS images;

[0079] In the actual evaluation, cloud-covered low-quality pixels were excluded, and the multispectral remote sensing images used HLS fusion image products, which integrated the images of Sentinel-2 and Landsat-8 / 9. The SAR image data used the GRD data products VH and VV backscatter coefficients of Sentinel-1.

[0080] The cultivated land characteristics are calculated according to the aggregated time series HLS images in the time period, and the land cover is divided into three modes, and LSTM autoencoder models are respectively constructed according to the three modes; the three models are cultivated land-idle cultivated land, idle cultivated land-idle cultivated land, and idle cultivated land-cultivated land; the cultivated land characteristics include vegetation index, enhanced vegetation index, soil index, LSWI, VH backscatter coefficient, and VV backscatter coefficient;

[0081] In the actual evaluation, three LSTM autoencoder models were constructed respectively, using the vegetation index, enhanced vegetation index, soil index, LSWI, σ VV , σ VH Six features, construct feature time series vectors, and train models;

[0082] Optimizing the LSTM autoencoder model, inputting the vegetation index of the idle cultivated land into the LSTM autoencoder model to determine the residual distribution threshold;

[0083] In the actual evaluation, the training data of three types of samples are used and respectively input into the trained models M1, M2, and M3 for reconstruction to obtain the prediction residual distribution of the three types of sample models. The residual distribution thresholds T1, T2, and T3 are taken in the 90% confidence space to calculate the residual:

[0084]

[0085] The actual observation value of the jth time is f j , the j-th prediction value is The jth residual is χ j ;

[0086] The absolute values ​​of the residuals of the six features are averaged to obtain the mean absolute residual;

[0087] Get the 90% mode interval of the mean absolute residual of the training samples as the threshold;

[0088] The data to be identified is input into the LSTM autoencoder model to obtain a reconstructed residual, and the identification result is output according to the reconstructed residual and the residual distribution threshold.

[0089] In this embodiment, the method of aggregating the time series remote sensing images month by month and taking the low value to obtain the time period aggregated time series SAR images includes:

[0090] The pixel data is sorted and the lowest pixel value is extracted, and the pixel data of the effective time series data is the lowest pixel value output time period aggregated time series SAR image.

[0091] In this embodiment, the method of constructing the LSTM autoencoder model according to the three modes respectively includes:

[0092] The LSTM autoencoder model is composed of a decoder and an encoder. The decoder contains two LSTM layers, and the encoder contains two LSTM layers and one Dense layer.

[0093] The three modes of cultivated land-idle cultivated land, idle cultivated land-idle cultivated land, and idle cultivated land-cultivated land are respectively input into the LSTM autoencoder model, and the converter model is used to extract the change features of the three modes respectively to obtain the first feature group, the second feature group, and the third feature group;

[0094] Encode the input data, the expression is:

[0095]

[0096] The transposition is T, and the encoding sequence of the u-th input data is The u-th query feature vector is k u , the u-th change eigenvector is The vector of the u-th eigenvalue is δ u , the number of input data is q, and the change value of the change feature is The u-th deviation vector is S u , query feature vector k u The kernel function is d(k u ), the changing feature vector The kernel function is

[0097] Calculate the similarity of the changed features:

[0098]

[0099] The length of the encoding sequence is M, the L2 norm function is ‖·‖2, and the similarity of the u-th change feature is The query feature vector at the sth moment is k u,s , initial time s o The query feature vector is

[0100] Construct the loss function of the LSTM autoencoder model, the expression is:

[0101]

[0102] The loss function is The coefficient of the balance function is α, and the characteristic vector of the s-th moment change is Initial time s o The changing eigenvector of The first loss weight is The second loss weight is

[0103] In this embodiment, the method for optimizing the LSTM autoencoder model includes:

[0104] Compute the average gradient:

[0105]

[0106] The average gradient of the tth iteration is The model parameters for the t-1th iteration are The number of random subsamples is m, and the ath subsample is v a , model parameters The advance gradient is The loss function is ε(·);

[0107] Compute the L2 norm of the average gradient:

[0108]

[0109] The L2 norm of the average gradient is The sth element of the average gradient is The sth regulatory factor is C s , calculate the historical gradient norm:

[0110]

[0111] The historical gradient norm of the tth iteration is The weight parameter is Calculate the correction gradient:

[0112]

[0113] The correction gradient of the tth iteration is When the historical gradient norm is greater than the L2 norm of the average gradient, update the correction gradient:

[0114]

[0115] The updated t-th iteration correction gradient is The hereditary coefficient is γ, the deviation constant is p, and the first-order momentum and second-order momentum are calculated:

[0116]

[0117] The first-order momentum of the tth iteration is y t , the first-order momentum y at the t-1th iteration t-1 , the first exponential decay rate is η1, and the second-order momentum of the tth iteration is h t , the second-order momentum of the t-1th iteration is h t-1 , the second exponential decay rate is η2, correcting the first-order momentum and the second-order momentum:

[0118]

[0119] The first exponential decay rate of the tth iteration is The second exponential decay rate of the tth iteration is The corrected first-order momentum is The corrected second-order momentum is

[0120] Update model parameters:

[0121]

[0122] The model parameters for the tth iteration are The learning rate is ρ, and the constant of change is

[0123] In this embodiment, the method of inputting the vegetation index of the idle cultivated land into the LSTM autoencoder model to determine the residual distribution threshold includes:

[0124] By obtaining the predicted residual distribution of the LSTM autoencoder model, the residual distribution of the 90% confidence space is output as the residual distribution threshold.

[0125] In this embodiment, the method of outputting the recognition result according to the reconstruction residual and the residual distribution threshold includes:

[0126] If the reconstruction residual is less than the residual distribution threshold, it is idle farmland; if the reconstruction residual is greater than T, it is non-idle farmland;

[0127] When judged as idle cultivated land: when the reconstruction residual of the first LSTM autoencoder model is less than the first residual distribution threshold, it is cultivated land-idle cultivated land; when the reconstruction residual of the second LSTM autoencoder model is less than the second residual distribution threshold, it is cultivated land-idle cultivated land; when the reconstruction residual of the third LSTM autoencoder model is less than the third residual distribution threshold, it is cultivated land-idle cultivated land.

[0128] Figure 2 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present application. Figure 2At the hardware level, the electronic device includes a processor, and optionally also includes an internal bus, a network interface, and a memory. The memory may include a memory, such as a high-speed random access memory (RAM), and may also include a non-volatile memory (non-volatile memory), such as at least one disk storage. Of course, the electronic device may also include hardware required for other services.

[0129] The processor, network interface and memory can be interconnected through an internal bus, which can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 2 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0130] The memory is used to store the program. Specifically, the program may include a program code, and the program code includes a computer operation instruction. The memory may include a memory and a non-volatile memory, and provides instructions and data to the processor.

[0131] The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it, forming a device for identifying idle farmland based on the LSTM autoencoder at the logical level. The processor executes the program stored in the memory and is specifically used to execute any of the aforementioned idle farmland identification methods based on the LSTM autoencoder.

[0132] The above application Figure 1The illustrated embodiment discloses a method for identifying idle farmland based on LSTM autoencoder, which can be applied to a processor or implemented by a processor. The processor may be an integrated circuit chip with signal processing capability. In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in the processor or an instruction in the form of software. The above processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the method disclosed in the embodiments of the present application can be directly embodied as being executed by a hardware decoding processor, or executed by a combination of hardware and software modules in a decoding processor. The software module can be located in a storage medium mature in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0133] The electronic device may also perform Figure 1 A method for identifying idle farmland based on LSTM autoencoder is proposed and implemented Figure 1 The functions of the illustrated embodiment will not be described in detail in the embodiments of the present application.

[0134] An embodiment of the present application also proposes a computer-readable storage medium, which stores one or more programs, and the one or more programs include instructions. When the instructions are executed by an electronic device including multiple applications, any one of the aforementioned idle farmland identification methods based on LSTM autoencoders is executed.

[0135] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0136] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0137] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0138] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0139] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0140] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0141] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0142] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0143] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0144] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for identifying idle farmland based on LSTM autoencoder, characterized in that: The following steps are involved: Collecting time series data of a preset area and preprocessing the time series data; the time series data includes multispectral remote sensing images and remote sensing image data; Aggregating the multispectral remote sensing images on a monthly basis and taking the median to obtain a time period aggregated time series HLS image, and aggregating the remote sensing image data on a monthly basis and taking the low value to obtain a time period aggregated time series SAR image; including: The superpixel segmentation algorithm is used to segment the multispectral remote sensing image to obtain regional superpixel data; Generative adversarial networks are used to extract regional image features of regional superpixel data and calculate the cultivated land information degree of regional image features: The cultivated land information degree of the image feature of the i-th region is , the image feature of the i-th region is , the i-th cultivated land feature is , regional image features The probability of occurrence is , characteristics of cultivated land The probability of occurrence is , regional image features and land characteristics The probability of simultaneous occurrence is , regional image features The correlation with cultivated land is , the coefficient of variation is , the number of regional image features is m; The regional cultivated land information degree is summed up, and the regional superpixel data with cultivated land information degree greater than 0.417 is regarded as cultivated land and marked as valid data, otherwise it is non-cultivated land and marked as invalid data; Sort the valid data according to the time sequence, extract the pixel data of the time sequence valid data, sort the pixel data and extract the pixel median; Aggregate the effective time series data of pixel median values ​​by quarter, and output the aggregation results as time period aggregated time series HLS images; The cultivated land characteristics are calculated according to the aggregated time series HLS images in the time period, and the land cover is divided into three modes, and LSTM autoencoder models are respectively constructed according to the three modes; the three models are cultivated land-idle cultivated land, idle cultivated land-idle cultivated land, and idle cultivated land-cultivated land; the cultivated land characteristics include vegetation index, enhanced vegetation index, soil index, LSWI, VH backscatter coefficient, and VV backscatter coefficient; Optimizing the LSTM autoencoder model, inputting the vegetation index of the idle cultivated land into the LSTM autoencoder model to determine the residual distribution threshold; The data to be identified is input into the LSTM autoencoder model to obtain a reconstructed residual, and the identification result is output according to the reconstructed residual and the residual distribution threshold.

2. According to claim 1, a method for identifying idle farmland based on LSTM autoencoder, characterized in that: The method of aggregating the remote sensing image data on a monthly basis and taking the lowest value to obtain a time-period aggregated time series SAR image comprises: The pixel data is sorted and the lowest pixel value is extracted. The pixel data of the time series valid data is used as the pixel lowest value output time period to aggregate the time series SAR image.

3. According to the method for identifying idle farmland based on LSTM autoencoder according to claim 1, it is characterized in that: The method of constructing the LSTM autoencoder model according to the three modes respectively includes: The LSTM autoencoder model is composed of a decoder and an encoder. The decoder contains two LSTM layers, and the encoder contains two LSTM layers and one Dense layer. The three modes of cultivated land-idle cultivated land, idle cultivated land-idle cultivated land, and idle cultivated land-cultivated land are respectively input into the LSTM autoencoder model, and the converter model is used to extract the change features of the three modes respectively to obtain the first feature group, the second feature group, and the third feature group; Encode the input data, the expression is: The transposition is T, and the encoding sequence of the u-th input data is , the u-th query feature vector is , the u-th change eigenvector is , the vector of the u-th eigenvalue is , the number of input data is q, and the change value of the change feature is , the u-th deviation vector is , query feature vector The kernel function is , the changing feature vector The kernel function is ; Calculate the similarity of the changed features: The length of the encoding sequence is M, and the L2 norm function is , the similarity of the u-th change feature is , the query feature vector at the sth moment is , the initial moment The query feature vector is ; Construct the loss function of the LSTM autoencoder model, the expression is: The loss function is , the coefficient of the balance function is , the s-th moment changing feature vector is , the initial moment The changing characteristic vector of , the first loss weight is , the second loss weight is .

4. According to the method for identifying idle farmland based on LSTM autoencoder according to claim 1, it is characterized in that: The method for optimizing the LSTM autoencoder model comprises: Compute the average gradient: The average gradient of the tth iteration is , the model parameters for the t-1th iteration are , the number of random subsamples is , the ath subsample is , model parameters The advance gradient is , the loss function is ; Compute the L2 norm of the average gradient: The L2 norm of the average gradient is , the sth element of the average gradient is , the sth regulatory factor is , calculate the historical gradient norm: The historical gradient norm of the tth iteration is , the weight parameter is , calculate the correction gradient: The correction gradient of the tth iteration is , when the historical gradient norm is greater than the L2 norm of the average gradient, update the correction gradient: The updated t-th iteration correction gradient is The heritability coefficient is , with a deviation constant of p, calculate the first-order momentum and the second-order momentum: The first-order momentum of the tth iteration is , the first-order momentum at the t-1th iteration The first exponential decay rate is , the second-order momentum of the tth iteration is , the second-order momentum of the t-1th iteration is The second exponential decay rate is , correcting the first-order momentum and the second-order momentum: The first exponential decay rate of the tth iteration is , the second exponential decay rate of the tth iteration is , the corrected first-order momentum is , the corrected second-order momentum is ; Update model parameters: The model parameters for the tth iteration are , the learning rate is , the constant of change is .

5. According to the method for identifying idle farmland based on LSTM autoencoder according to claim 1, it is characterized in that: The method of inputting the vegetation index of the idle cultivated land into the LSTM autoencoder model to determine the residual distribution threshold comprises: By obtaining the predicted residual distribution of the LSTM autoencoder model, the residual distribution of the 90% confidence space is output as the residual distribution threshold.

6. According to the method for identifying idle farmland based on LSTM autoencoder according to claim 1, it is characterized in that: The method for outputting a recognition result according to the reconstruction residual and the residual distribution threshold comprises: If the reconstruction residual is less than the residual distribution threshold, it is idle farmland; if the reconstruction residual is greater than T, it is non-idle farmland; When judged as idle cultivated land: when the reconstruction residual of the first LSTM autoencoder model is less than the first residual distribution threshold, it is cultivated land-idle cultivated land; when the reconstruction residual of the second LSTM autoencoder model is less than the second residual distribution threshold, it is cultivated land-idle cultivated land; when the reconstruction residual of the third LSTM autoencoder model is less than the third residual distribution threshold, it is cultivated land-idle cultivated land.

7. An electronic device comprising: processor; as well as A memory arranged to store computer executable instructions, which, when executed, cause the processor to perform the method of any one of claims 1 to 6.

8. A computer-readable storage medium storing one or more programs, which, when executed by an electronic device including a plurality of application programs, enables the electronic device to execute the method according to any one of claims 1 to 6.

Citation Information

Patent Citations

  • Hyperspectral image classification method based on auto-encoder and 3D deep residual network

    CN112232280A

  • Cultivation area change detection method and device based on invariant information sample screening

    CN113989657A

  • Cultivated land change detection method and device

    CN114972191A

  • Deep learning-based remote sensing image cultivated land non-agricultural change detection method

    CN115690591A

  • Remote sensing image-based cultivated land extraction method, device, equipment and medium

    CN116994140A