Remote sensing image automatic classification system and method based on deep learning

Abstract

The invention discloses a remote sensing image automatic classification system and method based on deep learning, which relates to the field of image processing technology, and comprises an infrared remote sensing module, an image classification module, and a landform determination module. The infrared remote sensing module is used to scan the ground of a grazing area by using infrared remote sensing technology. The image classification module is used to distinguish the types of infrared scanning images and to establish a far-infrared scanning map of the grazing area. The landform determination module is used to name the landform according to the analysis result of heat distribution and to predict the changing rules of various landforms with the seasons. The infrared remote sensing module comprises a near-infrared band sensor, a mid-infrared band sensor, a far-infrared band sensor, a data acquisition module, a sensor control module, a GPS positioning module, a scanning area definition module, and a drone. The invention has the characteristic of automatically determining the infrared image signal corresponding to the landform.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a remote sensing image automatic classification system and method based on deep learning. Background Art

[0002] Infrared remote sensing technology is increasingly being used to identify grasslands and water sources in grazing. Remote sensing technology can use drones to monitor the coverage and growth of large areas of grassland, quickly assess the quality of grassland in grazing areas, and help herders manage grazing scientifically. Remote sensing technology can also quickly locate and assess the quantity, distribution, and changes of water sources, analyze the spectral characteristics of water bodies, assess water quality, and determine the turbidity and pollution of water bodies, providing a reference for pasture water use.

[0003] In infrared remote sensing systems, near-infrared and mid-infrared are usually used for detection and imaging. The near-infrared band is more sensitive to the reflection of plants and water bodies, and the mid-infrared band is very effective for detecting surface soil. In the prior art, since it is not known in advance what kind of landforms exist in different locations, two-band detectors are used to scan the grazing area, and then the image of which band is used to scan different areas in the remote sensing map is manually selected. This method requires manual judgment and is time-consuming and labor-intensive. If a multi-band sensor that integrates the ability to detect multiple bands at the same time is used, the cost is too high for herders. Therefore, it is necessary to design an automatic classification system and method for remote sensing images based on deep learning that can automatically determine. Summary of the invention

[0004] The purpose of the present invention is to provide a remote sensing image automatic classification system and method based on deep learning to solve the problems raised in the above background technology.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: a remote sensing image automatic classification system and method based on deep learning, including an infrared remote sensing module, an image classification module, and a landform determination module. The infrared remote sensing module is used to scan the ground of a grazing area using infrared remote sensing technology. The image classification module is used to distinguish the types of infrared scanning images and establish a far-infrared scanning map of the grazing area. The landform determination module is used to name the landform according to the analysis results of the thermal distribution and predict the changing patterns of various landforms with the seasons.

[0006] According to the above technical solution, the infrared remote sensing module includes a near-infrared band sensor, a mid-infrared band sensor, a far-infrared band sensor, a data acquisition module, a sensor control module, a GPS positioning module, a scanning area definition module, and a drone. The infrared band sensor, the mid-infrared band sensor, and the far-infrared band sensor are all electrically connected to the data acquisition module and the sensor control module. The GPS positioning module is electrically connected to the scanning area definition module. The infrared band sensor is used to receive light in the near-infrared band, the mid-infrared band sensor is used to receive light in the mid-infrared band, and the far-infrared band sensor is used to receive light in the far-infrared band. The data acquisition module is used to collect image signals of the three sensors, and the sensor control module is used to control the work of the three sensors. The GPS positioning module is installed on the drone to locate the current scanning area, and the scanning area definition module is used to determine the specific position of the current scanning area in the grazing area. The drone is used to carry the three sensors for aerial overhead cruise scanning;

[0007] The image classification module includes a database, a map building module, and an image stitching module. The database is electrically connected to the data acquisition module, the map building module is electrically connected to the image stitching module and the scanning area definition module, the database is used to store image signals of three sensors, the map building module is used to build a far-infrared scanning map of the grazing area, and the image stitching module is used to select scanning images of different sensors according to different landforms, and stitch the images as a whole into the far-infrared scanning map;

[0008] The landform determination module includes a thermal distribution analysis module, a landform definition module, a timing module, a change law analysis module, and a regional prediction module. The thermal distribution analysis module is electrically connected to the scanning area definition module, the landform definition module is electrically connected to the thermal distribution analysis module, the timing module is electrically connected to the landform definition module, the change law analysis module is electrically connected to the timing module, and the regional prediction module is electrically connected to the change law analysis module; the thermal distribution analysis module is used to analyze the changes in the thermal distribution of the far-infrared scanning map at different times of the day based on the image signal collected by the far-infrared band sensor, the landform definition module is used to define the landform at different locations of the far-infrared scanning map based on the thermal distribution analysis results, the timing module is used to count the time of image signal collection, the change law analysis module uses a deep learning algorithm to analyze the change laws of various landform areas over time, and the regional prediction module is used to predict the changes in future landform areas over time.

[0009] Automatic classification methods for remote sensing images based on deep learning include:

[0010] S1. Install near-infrared band sensors, mid-infrared band sensors and far-infrared band sensors on the drone, record the edge of the grazing area and the GPS coordinates of the drone, and set the drone's cruise route so that the scanning area it passes through just covers the grazing area;

[0011] S2, start the drone, use the infrared remote sensing module to perform aerial overhead cruise scanning of the grazing area, scan x times at fixed intervals within a day, record the image signal collected by the far-infrared band sensor and obtain the change of heat distribution over time, and generate a far-infrared scanning map. According to the analysis results of the heat distribution, define the landforms at different positions of the far-infrared scanning map, and determine the landform areas of water bodies, soil and vegetation;

[0012] S3. Before grazing, the drone is started again to perform aerial overhead cruise scanning, and the image signals collected by the near-infrared band sensor and the mid-infrared band sensor are recorded. The two image signals are loaded into the far-infrared scanning map respectively, and the corresponding image signals are selected according to the types of different landform areas;

[0013] S4. The far-infrared scanning map processed by herders is used to plan grazing activities in the grazing area. The drone is used to conduct multiple cruise scans above the grazing area at fixed intervals within a year. The far-infrared band sensor is used to record the change pattern of the thermal distribution of the far-infrared scanning map within a year, and the specific conditions of water bodies, soil and vegetation are updated. The deep learning algorithm is used to analyze the change patterns of various landform areas over time, and the change of future landform areas over time is predicted.

[0014] According to the above technical solution, in S2, the image signal collected by the infrared band sensor generates a thermal distribution map, the continuous areas with the same temperature in the thermal distribution map are distinguished from each other, and the temperature scanned in each continuous area in one day is collected. The temperature scanned in a certain continuous area in one day is , the average temperature scanned at each time is obtained , calculate the variance of temperature , where the variance ranges of continuous regions representing water bodies, vegetation, and soil are , , ,and The specific value is selected according to the actual situation. When the temperature variance of a continuous area is within a certain range, the continuous area is determined to be the landform type corresponding to this range.

[0015] According to the above technical solution, in S3, the process of loading the two image signals into the far-infrared scanning map is specifically as follows:

[0016] S3-1, extract feature points from the infrared images collected by the near-infrared band sensor and the mid-infrared band sensor respectively, perform registration based on the extracted feature points, ensure that the two infrared images are aligned with the pixels of the far-infrared scanning map, and the three images are superimposed on each other;

[0017] S3-2. According to the types of each continuous area divided in the far-infrared scanning map, the corresponding area in the infrared image of the matching type is intercepted, and the infrared image collected by the near-infrared band sensor is used for the continuous area representing water bodies and vegetation, and the infrared image collected by the mid-infrared band sensor is used for the continuous area representing soil.

[0018] According to the above technical solution, in S4, the specific process of using the far-infrared band sensor to record the changing law of the thermal distribution of the far-infrared scanning map within one year is:

[0019] S4-1. Use a UAV equipped with a far-infrared band sensor to scan once every k days in a year, record the shape and size of each continuous area representing water bodies, vegetation and soil, incorporate the collected images into the far-infrared scanning map, and derive the pattern of landform changes within a year. Perform scans on the same number of days each year for n consecutive years to derive the pattern of landform changes within n years.

[0020] According to the above technical solution, in S4, the specific conditions of water bodies, soil and vegetation are updated in the following manner:

[0021] S4-2. When the drone is performing overhead cruise scanning, the coordinates of the drone's current infrared scanning map are obtained based on GPS positioning, the real-time scanning area is determined, and the latest far-infrared scanning map of the grazing area in the database is read. When there are only continuous areas of water and vegetation in the range corresponding to the real-time scanning area in the far-infrared scanning map, the mid-infrared band sensor is turned off and the near-infrared band sensor is turned on. When there are only continuous areas of soil in the range corresponding to the real-time scanning area in the far-infrared scanning map, the near-infrared band sensor is turned off and the mid-infrared band sensor is turned on.

[0022] S4-3: When the scan area only contains continuous areas of water and vegetation, and the distance between the scan area boundary and the continuous area of ​​soil is less than When the mid-infrared band sensor is turned on, when there is only a continuous area of ​​soil in the scanning area, and the distance between the scanning area boundary and the continuous area of ​​water and vegetation is less than When the near infrared band sensor is turned on, The duration after the latest far-infrared scanning map of this grazing area in the database is obtained. is the coefficient of how fast the landform changes over time.

[0023] According to the above technical solution, in S4, the specific method of using a deep learning algorithm to analyze the change patterns of various landform areas over time and predict the change of future landform areas over time is:

[0024] S4-4. Compare the far-infrared scanning maps collected at the same time in n years horizontally to obtain overlapping continuous areas. These areas are periodic change areas that occur with the seasons. When the continuous areas scanned from two consecutive years have overlapping parts but the shapes change, these areas are trend change areas. When the continuous areas scanned from two consecutive years have no overlapping parts, these areas are sudden change areas. When it is necessary to make a prediction for a certain time point in the future, first find the far-infrared scanning maps of previous years corresponding to this time point. In the predicted far-infrared scanning map, the periodic change area remains unchanged, the trend change area continues to change according to the previous trend, and the sudden change area uses the latest far-infrared scanning map.

[0025] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: the present invention equips the drone with a far-infrared band sensor to scan the grazing area multiple times a day. The far-infrared band can monitor the thermal distribution of water bodies, soil and vegetation. By observing the change law of thermal distribution over time, the grazing area is divided into landforms. In the remote sensing scan, infrared scanning images of different bands are used for different landforms, so as to achieve the automatic classification effect of the scanned images without manual judgment.

[0026] By using deep learning technology, the regional position changes of plants and water bodies in a year are counted, and the changing patterns with the seasons are predicted. Most of the time, only one scan is needed to monitor the soil, water bodies and vegetation in the grazing area. In addition, because the utilization rate of infrared band sensors is low, they can be circulated among herders, further saving equipment costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0028] Figure 1 It is a schematic diagram of the overall module structure of the present invention. DETAILED DESCRIPTION

[0029] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0030] See also Figure 1 The present invention provides a technical solution: a remote sensing image automatic classification system and method based on deep learning, including an infrared remote sensing module, an image classification module, and a landform determination module. The infrared remote sensing module is used to scan the ground of the grazing area using infrared remote sensing technology. The image classification module is used to distinguish the types of infrared scanning images and establish a far-infrared scanning map of the grazing area. The landform determination module is used to name the landform according to the analysis results of the thermal distribution and predict the changing rules of various landforms with the seasons.

[0031] The infrared remote sensing module includes a near-infrared band sensor, a mid-infrared band sensor, a far-infrared band sensor, a data acquisition module, a sensor control module, a GPS positioning module, a scanning area definition module, and a drone. The infrared band sensor, the mid-infrared band sensor, and the far-infrared band sensor are all electrically connected to the data acquisition module and the sensor control module. The GPS positioning module is electrically connected to the scanning area definition module. The infrared band sensor is used to receive light in the near-infrared band, the mid-infrared band sensor is used to receive light in the mid-infrared band, and the far-infrared band sensor is used to receive light in the far-infrared band. The data acquisition module is used to collect image signals of the three sensors. The sensor control module is used to control the work of the three sensors. The GPS positioning module is installed on the drone and is used to locate the current scanning area. The scanning area definition module is used to determine the specific location of the current scanning area in the grazing area. The drone is used to carry the three sensors for aerial overhead cruise scanning.

[0032] The image classification module includes a database, a map building module, and an image stitching module. The database is electrically connected to the data acquisition module, the map building module is electrically connected to the image stitching module and the scanning area definition module, the database is used to store image signals of three sensors, the map building module is used to build a far-infrared scanning map of the grazing area, and the image stitching module is used to select scanning images of different sensors according to different landforms, and stitch the images as a whole into the far-infrared scanning map;

[0033] The landform determination module includes a heat distribution analysis module, a landform definition module, a timing module, a change law analysis module, and a regional prediction module. The heat distribution analysis module is electrically connected to the scanning area definition module, the landform definition module is electrically connected to the heat distribution analysis module, the timing module is electrically connected to the landform definition module, the change law analysis module is electrically connected to the timing module, and the regional prediction module is electrically connected to the change law analysis module; the heat distribution analysis module is used to analyze the changes of the heat distribution of the far-infrared scanning map at different times of the day according to the image signal collected by the far-infrared band sensor, the landform definition module is used to define the landforms at different positions of the far-infrared scanning map according to the heat distribution analysis results, the timing module is used to count the time of image signal collection, the change law analysis module uses a deep learning algorithm to analyze the change laws of various landform areas over time, and the regional prediction module is used to predict the changes of future landform areas over time;

[0034] Automatic classification methods for remote sensing images based on deep learning include:

[0035] S1. Install near-infrared band sensors, mid-infrared band sensors and far-infrared band sensors on the drone, record the edge of the grazing area and the GPS coordinates of the drone, and set the drone's cruise route so that the scanning area it passes through just covers the grazing area;

[0036] S2, start the drone, use the infrared remote sensing module to perform aerial overhead cruise scanning of the grazing area, scan x times at fixed intervals within a day, record the image signal collected by the far-infrared band sensor and obtain the change of heat distribution over time, and generate a far-infrared scanning map. According to the analysis results of the heat distribution, define the landforms at different positions of the far-infrared scanning map, and determine the landform areas of water bodies, soil and vegetation;

[0037] S3. Before grazing, the drone is started again to perform aerial overhead cruise scanning, and the image signals collected by the near-infrared band sensor and the mid-infrared band sensor are recorded. The two image signals are loaded into the far-infrared scanning map respectively, and the corresponding image signals are selected according to the types of different landform areas;

[0038] S4. The herders use the far-infrared scanning map processed by observation to plan grazing activities in the grazing area. The drones are used to conduct multiple cruise scans over the grazing area at fixed intervals within a year. The far-infrared band sensor is used to record the change pattern of the thermal distribution of the far-infrared scanning map within a year, and the specific conditions of water bodies, soil and vegetation are updated. The deep learning algorithm is used to analyze the change patterns of various landform areas over time, and the change of future landform areas over time is predicted;

[0039] In S2, the image signal collected by the infrared band sensor generates a thermal distribution map, distinguishes the continuous areas with the same temperature in the thermal distribution map, and collects the temperature scanned in each continuous area within a day. The temperature scanned in a continuous area within a day is , the average temperature scanned at each time is obtained , calculate the variance of temperature , where the variance ranges of continuous regions representing water bodies, vegetation, and soil are , , ,and The specific value is selected according to the actual situation. When the temperature variance of a continuous area is within a certain range, the continuous area is determined to be the landform type corresponding to this range.

[0040] In S3, the process of loading the two image signals into the far-infrared scanning map is as follows:

[0041] S3-1, extract feature points from the infrared images collected by the near-infrared band sensor and the mid-infrared band sensor respectively, perform registration based on the extracted feature points, ensure that the two infrared images are aligned with the pixels of the far-infrared scanning map, and the three images are superimposed on each other;

[0042] S3-2, according to the types of each continuous area divided in the far-infrared scanning map, the corresponding area in the infrared image with matching types is intercepted, and the infrared image collected by the near-infrared band sensor is selected for the continuous area representing water bodies and vegetation, and the infrared image collected by the mid-infrared band sensor is selected for the continuous area representing soil;

[0043] In S4, the specific process of using the far infrared band sensor to record the changing law of the heat distribution of the far infrared scanning map within one year is as follows:

[0044] S4-1. Scan with a drone equipped with a far-infrared band sensor every k days in a year, record the shape and size of each continuous area representing water bodies, vegetation and soil, incorporate the collected images into the far-infrared scanning map, and obtain the landform change pattern within a year. Scan on the same number of days each year in n consecutive years to obtain the landform change pattern within n years;

[0045] In S4, the specific conditions of water, soil and vegetation are updated as follows:

[0046] S4-2. When the drone is performing overhead cruise scanning, the coordinates of the drone's current infrared scanning map are obtained based on GPS positioning, the real-time scanning area is determined, and the latest far-infrared scanning map of the grazing area in the database is read. When there are only continuous areas of water and vegetation in the range corresponding to the real-time scanning area in the far-infrared scanning map, the mid-infrared band sensor is turned off and the near-infrared band sensor is turned on. When there are only continuous areas of soil in the range corresponding to the real-time scanning area in the far-infrared scanning map, the near-infrared band sensor is turned off and the mid-infrared band sensor is turned on.

[0047] S4-3: When the scan area only contains continuous areas of water and vegetation, and the distance between the scan area boundary and the continuous area of ​​soil is less than When the mid-infrared band sensor is turned on, when there is only a continuous area of ​​soil in the scanning area, and the distance between the scanning area boundary and the continuous area of ​​water and vegetation is less than When the near infrared band sensor is turned on, The duration after the latest far-infrared scanning map of this grazing area in the database is obtained. is the coefficient of the speed of landform change over time;

[0048] By selectively shutting down unnecessary sensors, energy can be saved during the drone cruise scanning, making it last longer, saving time for charging back and forth, and improving scanning efficiency. Because the longer the latest far-infrared scanning map of the grazing area is, the greater the variation of the terrain, so it is necessary to turn on the corresponding infrared band sensor at a farther distance in advance to avoid the problem of inaccurate image signal scanning due to terrain changes;

[0049] In S4, the specific method of using deep learning algorithms to analyze the changing patterns of various landform areas over time and predict the changes of future landform areas over time is as follows:

[0050] S4-4. Compare the far-infrared scanning maps collected at the same time in n years horizontally to obtain overlapping continuous areas. These areas are periodic change areas that occur with the seasons. When the continuous areas scanned from two consecutive years have overlapping parts but the shapes change, these areas are trend change areas. When the continuous areas scanned from two consecutive years have no overlapping parts, these areas are sudden change areas. When it is necessary to make a prediction for a certain time point in the future, first find the far-infrared scanning maps of previous years corresponding to this time point. In the predicted far-infrared scanning map, the periodic change area remains unchanged, the trend change area continues to change according to the previous trend, and the sudden change area uses the latest far-infrared scanning map.

[0051] Since far-infrared sensors usually need to use some specific materials (such as germanium-doped silicon, lead telluride, quantum well materials, etc.) to detect radiation in the far-infrared band, the cost of these materials is usually high and the production process is complicated, resulting in the cost of far-infrared sensors being significantly higher than mid-infrared sensors and near-infrared sensors. Therefore, renting and circulating far-infrared sensors among herders is a lower-cost option. The present invention records the data collected by the far-infrared sensor during each scan, analyzes the changing patterns of water bodies, soil and vegetation over time within a year, and predicts future changes, which can reduce the demand for far-infrared sensors and effectively save equipment costs.

[0052] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants 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 other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0053] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. 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.

Claims

1. A remote sensing image automatic classification system based on deep learning, characterized by: It includes an infrared remote sensing module, an image classification module, and a landform determination module. The infrared remote sensing module is used to scan the ground of the grazing area using infrared remote sensing technology. The image classification module is used to distinguish the types of infrared scanning images and establish a far-infrared scanning map of the grazing area. The landform determination module is used to name the landform according to the analysis results of the thermal distribution and predict the changing rules of various landforms with the seasons. The infrared remote sensing module includes a near-infrared band sensor, a mid-infrared band sensor, a far-infrared band sensor, a data acquisition module, a sensor control module, a GPS positioning module, a scanning area definition module, and a drone. The infrared band sensor, the mid-infrared band sensor, and the far-infrared band sensor are all electrically connected to the data acquisition module and the sensor control module. The GPS positioning module is electrically connected to the scanning area definition module. The infrared band sensor is used to receive light in the near-infrared band, the mid-infrared band sensor is used to receive light in the mid-infrared band, and the far-infrared band sensor is used to receive light in the far-infrared band. The data acquisition module is used to collect image signals of the three sensors. The sensor control module is used to control the operation of the three sensors. The GPS positioning module is installed on the drone to locate the current scanning area. The scanning area definition module is used to determine the specific position of the current scanning area in the grazing area. The drone is used to carry the three sensors for aerial overhead cruise scanning. The image classification module includes a database, a map building module, and an image stitching module. The database is electrically connected to the data acquisition module, the map building module is electrically connected to the image stitching module and the scanning area definition module, the database is used to store image signals of three sensors, the map building module is used to build a far-infrared scanning map of the grazing area, and the image stitching module is used to select scanning images of different sensors according to different landforms, and stitch the images as a whole into the far-infrared scanning map; The landform determination module includes a thermal distribution analysis module, a landform definition module, a timing module, a change law analysis module, and a regional prediction module. The thermal distribution analysis module is electrically connected to the scanning area definition module, the landform definition module is electrically connected to the thermal distribution analysis module, the timing module is electrically connected to the landform definition module, the change law analysis module is electrically connected to the timing module, and the regional prediction module is electrically connected to the change law analysis module; the thermal distribution analysis module is used to analyze the changes in the thermal distribution of the far-infrared scanning map at different times of the day based on the image signal collected by the far-infrared band sensor, the landform definition module is used to define the landform at different locations of the far-infrared scanning map based on the thermal distribution analysis results, the timing module is used to count the time of image signal collection, the change law analysis module uses a deep learning algorithm to analyze the change laws of various landform areas over time, and the regional prediction module is used to predict the changes in future landform areas over time.

2. A remote sensing image automatic classification method based on deep learning, characterized by: The method works according to the system of claim 1, comprising: S1. Install near-infrared band sensors, mid-infrared band sensors and far-infrared band sensors on the drone, record the edge of the grazing area and the GPS coordinates of the drone, and set the drone's cruise route so that the scanning area it passes through just covers the grazing area; S2, start the drone, use the infrared remote sensing module to perform aerial overhead cruise scanning of the grazing area, scan x times at fixed intervals within a day, record the image signal collected by the far-infrared band sensor and obtain the change of heat distribution over time, and generate a far-infrared scanning map. According to the analysis results of the heat distribution, define the landforms at different positions of the far-infrared scanning map, and determine the landform areas of water bodies, soil and vegetation; S3. Before grazing, the drone is started again to perform aerial overhead cruise scanning, and the image signals collected by the near-infrared band sensor and the mid-infrared band sensor are recorded. The two image signals are loaded into the far-infrared scanning map respectively, and the corresponding image signals are selected according to the types of different landform areas; S4. The far-infrared scanning map processed by herders is used to plan grazing activities in the grazing area. The drone is used to conduct multiple cruise scans above the grazing area at fixed intervals within a year. The far-infrared band sensor is used to record the change pattern of the thermal distribution of the far-infrared scanning map within a year, and the specific conditions of water bodies, soil and vegetation are updated. The deep learning algorithm is used to analyze the change patterns of various landform areas over time, and the change of future landform areas over time is predicted.

3. The method for automatic classification of remote sensing images based on deep learning according to claim 2, characterized in that: In S2, the image signal collected by the infrared band sensor generates a thermal distribution map, and the continuous areas with the same temperature in the thermal distribution map are distinguished from each other, and the temperature scanned in each continuous area in one day is collected. The temperature scanned in a continuous area in one day is , the average temperature scanned at each time is obtained , calculate the variance of temperature , where the variance ranges of continuous regions representing water bodies, vegetation, and soil are , , ,and The specific value is selected according to the actual situation. When the temperature variance of a continuous area is within a certain range, the continuous area is determined to be the landform type corresponding to this range.

4. The method for automatic classification of remote sensing images based on deep learning according to claim 3, characterized in that: In S3, the process of loading the two image signals into the far-infrared scanning map is specifically as follows: S3-1, extract feature points from the infrared images collected by the near-infrared band sensor and the mid-infrared band sensor respectively, perform registration based on the extracted feature points, ensure that the two infrared images are aligned with the pixels of the far-infrared scanning map, and the three images are superimposed on each other; S3-2. According to the types of each continuous area divided in the far-infrared scanning map, the corresponding area in the infrared image of the matching type is intercepted, and the infrared image collected by the near-infrared band sensor is used for the continuous area representing water bodies and vegetation, and the infrared image collected by the mid-infrared band sensor is used for the continuous area representing soil.

5. The method for automatic classification of remote sensing images based on deep learning according to claim 4, characterized in that: In S4, the specific process of using the far infrared band sensor to record the changing law of the heat distribution of the far infrared scanning map within one year is: S4-1. Use a UAV equipped with a far-infrared band sensor to scan once every k days in a year, record the shape and size of each continuous area representing water bodies, vegetation and soil, incorporate the collected images into the far-infrared scanning map, and derive the pattern of landform changes within a year. Perform scans on the same number of days each year for n consecutive years to derive the pattern of landform changes within n years.

6. The method for automatic classification of remote sensing images based on deep learning according to claim 5, characterized in that: In S4, the specific conditions of water bodies, soil and vegetation are updated in the following manner: S4-2. When the drone is performing overhead cruise scanning, the coordinates of the drone's current infrared scanning map are obtained based on GPS positioning, the real-time scanning area is determined, and the latest far-infrared scanning map of the grazing area in the database is read. When there are only continuous areas of water and vegetation in the range corresponding to the real-time scanning area in the far-infrared scanning map, the mid-infrared band sensor is turned off and the near-infrared band sensor is turned on. When there are only continuous areas of soil in the range corresponding to the real-time scanning area in the far-infrared scanning map, the near-infrared band sensor is turned off and the mid-infrared band sensor is turned on. S4-3: When the scan area only contains continuous areas of water and vegetation, and the distance between the scan area boundary and the continuous area of ​​soil is less than When the mid-infrared band sensor is turned on, when there is only a continuous area of ​​soil in the scanning area, and the distance between the scanning area boundary and the continuous area of ​​water and vegetation is less than When the near infrared band sensor is turned on, The duration after the latest far-infrared scanning map of this grazing area in the database is obtained. is the coefficient of how fast the landform changes over time.

7. The method for automatic classification of remote sensing images based on deep learning according to claim 6, characterized in that: In S4, the specific method of using the deep learning algorithm to analyze the change patterns of various landform areas over time and predict the change of future landform areas over time is: S4-4. Compare the far-infrared scanning maps collected at the same time in n years horizontally to obtain overlapping continuous areas. These areas are periodic change areas that occur with the seasons. When the continuous areas scanned from two consecutive years have overlapping parts but the shapes change, these areas are trend change areas. When the continuous areas scanned from two consecutive years have no overlapping parts, these areas are sudden change areas. When it is necessary to make a prediction for a certain time point in the future, first find the far-infrared scanning maps of previous years corresponding to this time point. In the predicted far-infrared scanning map, the periodic change area remains unchanged, the trend change area continues to change according to the previous trend, and the sudden change area uses the latest far-infrared scanning map.

Images