Cultivated land extraction method and device based on time sequence spectrum change characteristics

Abstract

The invention relates to the technical field of artificial intelligence, and relates to a cultivated land extraction method and device based on time sequence spectrum change characteristics, and the method comprises the steps: obtaining first information, second information and third information, the first information comprises high-resolution images collected along the railway according to a time sequence, the second information comprises medium-low-resolution images collected along the railway according to the time sequence, and the third information comprises a farmland vector boundary; preprocessing the first information, the second information and the third information to obtain preprocessed sample information; constructing a deep learning data set based on the preprocessed sample information; training a preset cultivated land extraction deep network model by using the deep learning data set to obtain a trained cultivated land extraction deep network model; the cultivated land of the railway scheme research area is extracted based on the trained cultivated land extraction deep network model, and an extraction result is obtained.According to the method, multi-temporal information is introduced, the vegetation time sequence spectrum change rule is obtained, and easily-confused ground objects are accurately judged.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method and device for extracting cultivated land based on time-series spectral variation characteristics. Background Art

[0002] For the extraction of cultivated land in the railway plan study area, remote sensing technology is usually used for identification. Remote sensing technology can quickly identify the current distribution of cultivated land by acquiring high-resolution ground image data. However, the current single-phase image is used for cultivated land extraction, which is prone to cause confusion of ground objects (such as grassland, woodland, etc.), resulting in low recognition accuracy. Summary of the invention

[0003] The object of the present invention is to provide a method and device for extracting cultivated land based on time-series spectral variation characteristics.

[0004] In order to achieve the above objectives, the present application provides the following technical solutions:

[0005] On the one hand, an embodiment of the present application provides a method for extracting cultivated land based on time series spectral variation characteristics, the method comprising:

[0006] Acquire first information, second information and third information, wherein the first information includes high-resolution images collected in time sequence along the railway, the second information includes medium- and low-resolution images collected in time sequence along the railway, and the third information includes cultivated land vector boundaries;

[0007] Preprocessing the first information, the second information and the third information to obtain preprocessed sample information, wherein the preprocessing is used to correct pixel errors between images;

[0008] Constructing a deep learning dataset based on the preprocessed sample information;

[0009] Using the deep learning data set to train a preset deep network model for farmland extraction, to obtain a trained deep network model for farmland extraction;

[0010] Based on the trained deep network model for farmland extraction, farmland in the railway plan study area is extracted to obtain an extraction result, which includes at least one piece of farmland in a local area.

[0011] In a second aspect, an embodiment of the present application provides a device for extracting cultivated land based on time-series spectral variation characteristics, the device comprising:

[0012] An acquisition module, used to acquire first information, second information and third information, wherein the first information includes high-resolution images collected in time sequence along the railway, the second information includes medium- and low-resolution images collected in time sequence along the railway, and the third information includes cultivated land vector boundaries;

[0013] A first processing module, used for preprocessing the first information, the second information and the third information to obtain preprocessed sample information, wherein the preprocessing is used for correcting pixel errors between images;

[0014] A second processing module, used to construct a deep learning data set based on the preprocessed sample information;

[0015] A third processing module is used to train a preset deep network model for farmland extraction using the deep learning data set to obtain a trained deep network model for farmland extraction;

[0016] The fourth processing module is used to extract the cultivated land in the railway plan study area based on the trained cultivated land extraction deep network model to obtain an extraction result, and the extraction result includes at least one piece of cultivated land in a local area.

[0017] In a third aspect, the present application provides a device for extracting cultivated land based on time-series spectral variation characteristics, the device comprising a memory and a processor. The memory is used to store a computer program; the processor is used to implement the steps of the above-mentioned method for extracting cultivated land based on time-series spectral variation characteristics when executing the computer program.

[0018] In a fourth aspect, an embodiment of the present application provides a readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the above-mentioned method for extracting cultivated land based on time-series spectral change characteristics are implemented.

[0019] The beneficial effects of the present invention are:

[0020] The present invention constructs a deep learning dataset through the first information, the second information and the third information, and then uses the deep learning dataset to train a deep network model for cultivated land extraction. The temporal characteristics of medium and low resolution remote sensing images are introduced into the deep network model, so that the deep network can stably locate and respond to cultivated land areas even in complex backgrounds, and can accurately distinguish easily confused objects such as cultivated land, grassland, woodland, and garden. The cultivated land boundary segmentation is fine and smooth, and the cultivated land semantic information is complete and continuous, and it has good robustness and reliability.

[0021] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by implementing the embodiments of the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0023] Figure 1 It is a schematic flow chart of the cultivated land extraction method based on time series spectral variation characteristics described in an embodiment of the present invention.

[0024] Figure 2 It is a schematic diagram of the structure of the cultivated land extraction device based on the time series spectrum change characteristics described in an embodiment of the present invention.

[0025] Figure 3 It is a schematic diagram of the structure of the cultivated land extraction equipment based on the time series spectrum change characteristics described in an embodiment of the present invention.

[0026] Labels in the figure: 800, cultivated land extraction equipment based on time series spectral change characteristics; 801, processor; 802, memory; 803, multimedia component; 804, I / O interface; 805, communication component; 901, acquisition module; 902, first processing module; 903, second processing module; 904, third processing module; 905, fourth processing module. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0028] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings. At the same time, in the description of the present invention, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0029] Embodiment 1:

[0030] This embodiment provides a method for extracting cultivated land based on time-series spectral variation characteristics. It can be understood that a scene can be laid out in this embodiment, for example: a scene of extracting and identifying cultivated land near a newly built railway or an existing railway.

[0031] See also Figure 1 , the figure shows that the method includes step S1, step S2, step S3, step S4 and step S5.

[0032] Step S1, obtaining first information, second information and third information, wherein the first information includes high-resolution images collected in time sequence along the railway, the second information includes medium- and low-resolution images collected in time sequence along the railway, and the third information includes cultivated land vector boundaries;

[0033] In this step, the preset resolution threshold is 0.3m×0.3m, and images that are better than the resolution threshold are used as high-resolution images, and images that are worse than the resolution threshold are used as medium- and low-resolution images. The first information can provide the model with deep and refined ground feature information, and the second information is the normalized vegetation index and the synthetic aperture radar image time series, which provides the model with the changing trend of vegetation elements along the railway over time. By collecting the first information, the second information and the third information, the growth laws of easily confused land types such as crops, grasslands, and woodlands are comprehensively considered.

[0034] Step S2, preprocessing the first information, the second information and the third information to obtain preprocessed sample information, wherein the preprocessing is used to correct pixel errors between images;

[0035] The step S2 also includes step S21 and step S22, which are specifically:

[0036] Step S21, projecting the first information, the second information and the third information into the same earth coordinate system to obtain projected sample information;

[0037] Step S22: Correct the three types of sample data included in the projected sample information using a feature point matching method to obtain preprocessed sample information.

[0038] In this embodiment, since there are significant differences between the three heterogeneous data of the first information, the second information and the third information, in order to ensure the effective operation of the model, it is necessary to preprocess the first information, the second information and the third information to eliminate pixel errors so that the position error is less than 1 pixel.

[0039] Step S3: constructing a deep learning data set based on the preprocessed sample information;

[0040] The step S3 also includes step S31, step S32, step S33 and step S34, which are specifically:

[0041] Step S31, using a preset sliding window to segment the high-resolution image in the pre-processed sample information to obtain segmented first image information;

[0042] In this step, due to the high degree of fragmentation of cultivated land along the railway, the preset sliding window size is preset to 256×256 grid, which can map 5 square kilometers of cultivated land, thereby effectively extracting the scope of cultivated land. It should be noted that when using the sliding window to segment the high-resolution image in the preprocessed sample information, a 50% slice overlap is set to ensure parameter sharing and connectivity of the convolutional neural network.

[0043] Step S32, determining geographic boundary information according to the segmented first image information;

[0044] Step S33, using the geographic boundary information to perform block processing on the medium and low resolution images in the pre-processed sample information to obtain segmented second image information;

[0045] In this step, the boundary information can be identified in the high-resolution image, but not in the medium- and low-resolution images. Therefore, the medium- and low-resolution images need to be divided into blocks according to the boundary information identified in the high-resolution image.

[0046] Step S34: import the cultivated land vector data into the segmented first image information and the segmented second image information to obtain a deep learning data set.

[0047] In this step, the cultivated land vector data is rasterized, the cultivated land pixel value is set to 1, the non-cultivated land pixel value is set to 2, and the geographic boundary information of the first image information and the second image information is imported for segmentation to obtain cultivated land sample data corresponding to the first image information and the second image information, thereby constructing a deep learning data set. Each set of data in the data set includes the first image information, the second image information and the corresponding cultivated land sample information.

[0048] Step S4, using the deep learning data set to train a preset farmland extraction deep network model to obtain a trained farmland extraction deep network model;

[0049] In this step, a specific implementation method is to divide the deep learning dataset into a training set, a validation set, and a test set in a ratio of 8:1:1 to support the training, validation, and testing of the deep learning model.

[0050] Step S5: extract the cultivated land in the railway plan study area based on the trained cultivated land extraction deep network model to obtain an extraction result, wherein the extraction result includes at least one piece of cultivated land in a local area.

[0051] In this step, the results of multiple local areas are spatially merged to obtain the cultivated land distribution information of the final target area.

[0052] The step S5 also includes step S51, step S52, step S53 and step S54, which are specifically:

[0053] Step S51, downsampling the first information of the cultivated land to be extracted to obtain first feature information;

[0054] In this step, downsampling includes four stages, each stage is connected through a maximum pooling layer, and four downsampling operations are performed on the first information.

[0055] Step S52, upsampling the second information of the cultivated land to be extracted to obtain second feature information;

[0056] In this step, the second information of the cultivated land to be extracted is upsampled so that its size is consistent with the first information of the cultivated land to be extracted after four downsamplings.

[0057] Step S53: stacking the first characteristic information and the second characteristic information to obtain stacked characteristic information;

[0058] The step S53 also includes step S531, step S532, step S533, step S534 and step S535, which are specifically:

[0059] Step S531, obtaining four dilated convolutional layers, wherein the expansion rates of the four dilated convolutional layers are 1, 2, 4, and 8 respectively;

[0060] Step S532: Use the four dilated convolution layers to perform dilated convolution on the first feature information respectively to obtain four third feature information;

[0061] Step S533, superimposing the four third characteristic information to obtain fourth characteristic information;

[0062] Step S534, convolve the fourth feature information to obtain first feature information;

[0063] Step S535: stack the first feature information and the second feature information to obtain stacked feature information.

[0064] In this embodiment, atrous convolutions with expansion rates of 1, 2, 4, and 8 are used in parallel to balance computational efficiency and model effectiveness, increase the receptive field, and capture multi-scale information. Feature maps of different scales are then fused through 1×1 convolution, and multi-temporal features are introduced through medium and low resolution time series image slices and superimposed with the multi-temporal features to generate stacked feature information. The present invention introduces vegetation temporal change characteristics to effectively solve the problem of accurately distinguishing easily confused land types in remote sensing images (such as grassland and cultivated land in the growth period).

[0065] Step S54: up-sample the stacked feature information to obtain an extraction result.

[0066] The step S54 also includes step S541, step S542 and step S543, which are specifically:

[0067] Step S541, obtaining four upsampling modules, wherein the upsampling modules include an attention gating unit and a convolutional layer;

[0068] Step S542: sending the stacked feature information to the four upsampling modules for repeated upsampling to obtain a prediction probability map;

[0069] Step S543: Determine the extraction result according to the predicted probability map.

[0070] In this step, the predicted probability map is binarized, with 0.5 as the threshold, pixels greater than 0.5 are cultivated land, and pixels less than or equal to 0.5 are non-cultivated land, so as to obtain the final predicted raster map, i.e. the extraction result.

[0071] In this embodiment, upsampling also includes four stages, and the stacked feature information is upsampled four times. Finally, the number of feature map channels is adjusted to 1 through a layer of convolution, and a logical activation function is applied to generate a prediction probability map. Between the corresponding encoder and decoder stages, a jump connection is used to retain multi-level feature information, and an attention gate unit is introduced to filter important feature information to enhance the expressiveness of the decoding process.

[0072] Embodiment 2:

[0073] like Figure 2As shown, this embodiment provides a cultivated land extraction device based on time series spectrum change characteristics, the device includes an acquisition module 901, a first processing module 902, a second processing module 903, a third processing module 904 and a fourth processing module 905, wherein specifically:

[0074] The acquisition module 901 is used to acquire first information, second information and third information, wherein the first information includes high-resolution images collected in time sequence along the railway, the second information includes medium- and low-resolution images collected in time sequence along the railway, and the third information includes cultivated land vector boundaries;

[0075] A first processing module 902 is used to preprocess the first information, the second information and the third information to obtain preprocessed sample information, wherein the preprocessing is used to correct pixel errors between images;

[0076] A second processing module 903, configured to construct a deep learning data set based on the preprocessed sample information;

[0077] The third processing module 904 is used to train the preset farmland extraction deep network model using the deep learning data set to obtain a trained farmland extraction deep network model;

[0078] The fourth processing module 905 is used to extract the cultivated land in the railway plan study area based on the trained cultivated land extraction deep network model to obtain an extraction result, and the extraction result includes at least one piece of cultivated land in a local area.

[0079] In a specific embodiment of the present disclosure, the first processing module further includes a first processing unit and a second processing unit, specifically:

[0080] A first processing unit, configured to project the first information, the second information, and the third information into the same earth coordinate system to obtain projected sample information;

[0081] The second processing unit is used to correct the three types of sample data included in the projected sample information by using a feature point matching method to obtain pre-processed sample information.

[0082] In a specific embodiment of the present disclosure, the second processing module further includes a third processing unit, a fourth processing unit, a fifth processing unit and a sixth processing unit, wherein specifically:

[0083] A third processing unit is used to segment the high-resolution image in the pre-processed sample information using a preset sliding window to obtain segmented first image information;

[0084] a fourth processing unit, configured to determine geographic boundary information according to the segmented first image information;

[0085] A fifth processing unit, configured to perform block processing on the medium and low resolution images in the pre-processed sample information by using the geographic boundary information to obtain segmented second image information;

[0086] The sixth processing unit is used to import the cultivated land vector data into the segmented first image information and the segmented second image information to obtain a deep learning data set.

[0087] In a specific implementation of the present disclosure, the fourth processing module further includes a seventh processing unit, an eighth processing unit, a ninth processing unit and a tenth processing unit, wherein specifically:

[0088] A seventh processing unit, configured to downsample the first information of the cultivated land to be extracted to obtain first feature information;

[0089] An eighth processing unit, configured to upsample the second information of the cultivated land to be extracted to obtain second feature information;

[0090] a ninth processing unit, configured to stack the first feature information and the second feature information to obtain stacked feature information;

[0091] The tenth processing unit is used to upsample the stacked feature information to obtain an extraction result.

[0092] In a specific implementation of the present disclosure, the ninth processing unit further includes a first acquisition unit, an eleventh processing unit, a twelfth processing unit, a thirteenth processing unit, and a fourteenth processing unit, which are specifically:

[0093] A first acquisition unit is used to acquire four dilated convolutional layers, where the expansion rates of the four dilated convolutional layers are 1, 2, 4, and 8, respectively;

[0094] an eleventh processing unit, configured to use the four dilated convolution layers to perform dilated convolutions on the first feature information respectively to obtain four third feature information;

[0095] a twelfth processing unit, configured to superimpose the four third feature information to obtain fourth feature information;

[0096] a thirteenth processing unit, configured to convolve the fourth feature information to obtain first feature information;

[0097] A fourteenth processing unit is used to stack the first feature information and the second feature information to obtain stacked feature information.

[0098] In a specific implementation of the present disclosure, the tenth processing unit further includes a second acquisition unit, a fifteenth processing unit, and a sixteenth processing unit, wherein specifically:

[0099] A second acquisition unit, used to acquire four upsampling modules, wherein the upsampling modules include an attention gating unit and a convolutional layer;

[0100] A fifteenth processing unit, configured to send the stacked feature information to the four upsampling modules for repeated upsampling to obtain a prediction probability map;

[0101] A sixteenth processing unit is used to determine an extraction result according to the predicted probability map.

[0102] It should be noted that, regarding the device in the above embodiment, the specific manner in which each module performs the operation has been described in detail in the embodiment of the method, and will not be elaborated here.

[0103] Embodiment 3:

[0104] Corresponding to the above method embodiment, this embodiment also provides a cultivated land extraction device based on time series spectral change characteristics. The cultivated land extraction device based on time series spectral change characteristics described below and the cultivated land extraction method based on time series spectral change characteristics described above can be referenced to each other.

[0105] Figure 3 FIG. 8 is a block diagram of a farmland extraction device 800 based on time series spectrum variation characteristics according to an exemplary embodiment. Figure 3 As shown, the cultivated land extraction device 800 based on time series spectrum change characteristics may include: a processor 801, a memory 802. The cultivated land extraction device 800 based on time series spectrum change characteristics may also include one or more of a multimedia component 803, an I / O interface 804, and a communication component 805.

[0106] The processor 801 is used to control the overall operation of the cultivated land extraction device 800 based on time series spectral change characteristics to complete all or part of the steps in the cultivated land extraction method based on time series spectral change characteristics. The memory 802 is used to store various types of data to support the operation of the cultivated land extraction device 800 based on time series spectral change characteristics. These data may include, for example, instructions for any application or method operating on the cultivated land extraction device 800 based on time series spectral change characteristics, and application-related data, such as contact data, sent and received messages, pictures, audio, video, etc. The memory 802 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 memory, flash memory, disk or optical disk. The multimedia component 803 may include a screen and an audio component. The screen may be, for example, 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 for receiving external audio signals. The received audio signal may be further stored in the memory 802 or sent via the communication component 805. The audio component also includes at least one speaker for outputting audio signals. The I / O interface 804 provides an interface between the processor 801 and other interface modules, and the above-mentioned other interface modules can be keyboards, mice, buttons, etc. These buttons can be virtual buttons or physical buttons. The communication component 805 is used for wired or wireless communication between the cultivated land extraction device 800 based on the time series spectrum change characteristics 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 805 can include: Wi-Fi module, Bluetooth module, NFC module.

[0107] In an exemplary embodiment, the cultivated land extraction device 800 based on time series spectral change characteristics can be implemented by one or more application specific integrated circuits (Application Specific Integrated Circuit, referred to as ASIC), digital signal processors (Digital Signal Processor, referred to as DSP), digital signal processing devices (Digital Signal Processing Device, referred to as DSPD), programmable logic devices (Programmable Logic Device, referred to as PLD), field programmable gate arrays (Field Programmable Gate Array, referred to as FPGA), controllers, microcontrollers, microprocessors or other electronic components to execute the above-mentioned cultivated land extraction method based on time series spectral change characteristics.

[0108] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, and when the program instructions are executed by a processor, the steps of the above-mentioned method for extracting cultivated land based on time-series spectral change characteristics are implemented. For example, the computer-readable storage medium can be the above-mentioned memory 802 including program instructions, and the above-mentioned program instructions can be executed by the processor 801 of the device 800 for extracting cultivated land based on time-series spectral change characteristics to complete the above-mentioned method for extracting cultivated land based on time-series spectral change characteristics.

[0109] Embodiment 4:

[0110] Corresponding to the above method embodiment, a readable storage medium is also provided in this embodiment. The readable storage medium described below and the cultivated land extraction method based on time series spectral change characteristics described above can refer to each other.

[0111] A readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method for extracting cultivated land based on time-series spectral variation characteristics of the above method embodiment.

[0112] The readable storage medium may specifically be a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, or other readable storage medium that can store program codes.

[0113] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0114] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for extracting cultivated land based on time series spectral variation characteristics, characterized in that: include: Acquire first information, second information and third information, wherein the first information includes high-resolution images collected in time sequence along the railway, the second information includes medium- and low-resolution images collected in time sequence along the railway, and the third information includes cultivated land vector boundaries; Preprocessing the first information, the second information and the third information to obtain preprocessed sample information, wherein the preprocessing is used to correct pixel errors between images; Constructing a deep learning dataset based on the preprocessed sample information; Using the deep learning data set to train a preset deep network model for farmland extraction, to obtain a trained deep network model for farmland extraction; Based on the trained deep network model for farmland extraction, farmland in the railway plan study area is extracted to obtain an extraction result, which includes at least one piece of farmland in a local area.

2. The cultivated land extraction method based on time series spectrum variation characteristics according to claim 1 is characterized in that: Preprocessing the first information, the second information, and the third information to obtain preprocessed sample information includes: Projecting the first information, the second information, and the third information into the same earth coordinate system to obtain projected sample information; The three types of sample data included in the projected sample information are corrected using a feature point matching method to obtain preprocessed sample information.

3. The cultivated land extraction method based on time series spectrum variation characteristics according to claim 1 is characterized in that: Constructing a deep learning data set based on the preprocessed sample information includes: Using a preset sliding window to segment the high-resolution image in the preprocessed sample information to obtain segmented first image information; Determining geographic boundary information according to the segmented first image information; Using the geographic boundary information, the medium and low resolution images in the preprocessed sample information are processed into blocks to obtain segmented second image information; The cultivated land vector data is imported into the segmented first image information and the segmented second image information to obtain a deep learning data set.

4. The method for extracting cultivated land based on time series spectrum variation characteristics according to claim 1, characterized in that: The cultivated land in the railway plan study area is extracted based on the trained cultivated land extraction deep network model, including: Downsampling the first information of the cultivated land to be extracted to obtain first feature information; Upsampling the second information of the cultivated land to be extracted to obtain second feature information; stacking the first feature information and the second feature information to obtain stacked feature information; The stacked feature information is upsampled to obtain an extraction result.

5. The method for extracting cultivated land based on time series spectrum variation characteristics according to claim 4, characterized in that: The stacked feature information is upsampled to obtain an extraction result, including: Obtain four upsampling modules, each of which includes an attention gating unit and a convolutional layer; Sending the stacked feature information to the four upsampling modules for repeated upsampling to obtain a prediction probability map; An extraction result is determined according to the predicted probability map.

6. A device for extracting cultivated land based on time-series spectral variation characteristics, characterized in that: include: An acquisition module, used to acquire first information, second information and third information, wherein the first information includes high-resolution images collected in time sequence along the railway, the second information includes medium- and low-resolution images collected in time sequence along the railway, and the third information includes cultivated land vector boundaries; A first processing module, used for preprocessing the first information, the second information and the third information to obtain preprocessed sample information, wherein the preprocessing is used for correcting pixel errors between images; A second processing module, used to construct a deep learning data set based on the preprocessed sample information; A third processing module is used to train a preset deep network model for farmland extraction using the deep learning data set to obtain a trained deep network model for farmland extraction; The fourth processing module is used to extract the cultivated land in the railway plan study area based on the trained cultivated land extraction deep network model to obtain an extraction result, and the extraction result includes at least one piece of cultivated land in a local area.

7. The cultivated land extraction device based on time series spectrum variation characteristics according to claim 6 is characterized in that: The first processing module comprises: A first processing unit, configured to project the first information, the second information, and the third information into the same earth coordinate system to obtain projected sample information; The second processing unit is used to correct the three types of sample data included in the projected sample information by using a feature point matching method to obtain pre-processed sample information.

8. The cultivated land extraction device based on time series spectrum variation characteristics according to claim 6 is characterized in that: The second processing module comprises: A third processing unit is used to segment the high-resolution image in the pre-processed sample information using a preset sliding window to obtain segmented first image information; a fourth processing unit, configured to determine geographic boundary information according to the segmented first image information; A fifth processing unit, configured to perform block processing on the medium and low resolution images in the pre-processed sample information by using the geographic boundary information to obtain segmented second image information; The sixth processing unit is used to import the cultivated land vector data into the segmented first image information and the segmented second image information to obtain a deep learning data set.

9. The cultivated land extraction device based on time series spectrum variation characteristics according to claim 6, characterized in that: The fourth processing module comprises: A seventh processing unit, configured to downsample the first information of the cultivated land to be extracted to obtain first feature information; An eighth processing unit, configured to upsample the second information of the cultivated land to be extracted to obtain second feature information; a ninth processing unit, configured to stack the first feature information and the second feature information to obtain stacked feature information; The tenth processing unit is used to upsample the stacked feature information to obtain an extraction result.

10. The method for extracting cultivated land based on time series spectrum variation characteristics according to claim 9, characterized in that: The tenth processing unit comprises: A second acquisition unit, used to acquire four upsampling modules, wherein the upsampling modules include an attention gating unit and a convolutional layer; A fifteenth processing unit, configured to send the stacked feature information to the four upsampling modules for repeated upsampling to obtain a prediction probability map; A sixteenth processing unit is used to determine an extraction result according to the predicted probability map.

Images