AI analysis method and system applied to land resource investigation

Abstract

The embodiment of the invention provides an AI analysis method and system applied to land resource investigation, and the method comprises the steps: obtaining the multi-source remote sensing image data of a target region, analyzing the spectrum and texture features of the multi-source remote sensing image data, generating a comprehensive land feature map, obtaining an initial land classification map through a land classification model, and obtaining the initial land classification map. And performing time sequence comparative analysis in combination with historical data to determine a dynamic area of land coverage change and generate a land resource change report. The method further comprises the functions of soil quality evaluation, change area visualization, abnormal area real-time monitoring, land classification model training, classification model adjustment according to user feedback and the like, and the efficiency and accuracy of land resource investigation are improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology. Specifically, it relates to an AI analysis method and system for land resource survey. Background Art

[0002] Land resources are an important material basis for human survival and development. Conducting accurate surveys and monitoring of land resources is of crucial significance for rational land use planning, ecological environment protection, and sustainable development. Traditional land resource survey methods often rely on manual on-site exploration, which is not only inefficient but also limited by manpower, material resources, and time, making it difficult to obtain comprehensive, timely, and accurate land information.

[0003] With the development of remote sensing technology, multi-source remote sensing image data provides a rich information source for land resource survey. However, how to efficiently process and analyze this massive amount of data has become a new challenge. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide an AI analysis method and system for land resource survey.

[0005] Combined with the first aspect of this application, an AI analysis method for land resource survey is provided, which is applied to an AI analysis system for land resource survey. The method includes:

[0006] Obtain multi-source remote sensing image data of the target area. The multi-source remote sensing image data includes visible light images, infrared images, and radar images, and contains spectral features and texture features related to land cover types;

[0007] Conduct spectral feature analysis and texture feature extraction on the multi-source remote sensing image data to generate a comprehensive land feature map of the target area. The comprehensive land feature map is used to characterize the spatial distribution and physical attributes of land cover types;

[0008] Based on a preset land classification model, perform pixel-by-pixel classification processing on the comprehensive land feature map to obtain an initial land classification map of the target area. Each pixel point in the initial land classification map is labeled with a corresponding land type label;

[0009] According to historical land survey data, conduct temporal contrast analysis on the land type labels in the initial land classification map, identify the dynamic areas of land cover changes in the target area, and extract the change trajectory information of the dynamic areas;

[0010] Generate a land resource change report based on the position coordinates, area data, and change trajectory information of the dynamic region. The land resource change report includes the boundary vector data of the changed region and the change trend prediction result.

[0011] In a possible implementation manner of the first aspect, the spectral feature analysis and texture feature extraction of the multi-source remote sensing image data to generate the comprehensive land feature map of the target region include:

[0012] Divide the visible light image into multiple spectral bands, calculate the spectral reflectance curve of each spectral band, and extract the peak feature and slope feature of the spectral reflectance curve;

[0013] Conduct thermal radiation intensity analysis on the infrared image to determine the thermal radiation difference parameters of different land cover types within the target region;

[0014] Extract polarization scattering features from the radar image to obtain surface roughness and vegetation height parameters;

[0015] Fuse the peak feature of the spectral reflectance curve, the thermal radiation difference parameter, and the surface roughness parameter to generate an enhanced spatial resolution version of the comprehensive land feature map.

[0016] In a possible implementation manner of the first aspect, the method further includes:

[0017] Determine the humidity parameter, organic matter content parameter, and pH parameter related to soil quality according to the physical properties of different land types in the comprehensive land feature map;

[0018] Match the humidity parameter, organic matter content parameter, and pH parameter with the preset soil quality assessment standard to generate the soil quality grade distribution map of the target region;

[0019] Perform spatial overlay analysis on the soil quality grade distribution map and the changed region in the land resource change report to determine the priority regions of land degradation or improvement.

[0020] In a possible implementation manner of the first aspect, the time-series comparison analysis of the land type labels in the initial land classification map according to the historical land survey data to identify the dynamic region of land cover change in the target region includes:

[0021] Obtain the historical land classification maps of the target region at at least three historical time points, and extract the land type distribution patterns corresponding to each historical time point;

[0022] Calculate the spatial difference between the initial land classification map and the land type distribution patterns at each historical time point to determine the conversion areas where land cover types have changed;

[0023] Classify the conversion types of the conversion areas to generate a set of change types including forest to arable land, arable land to construction land, and water body to wetland.

[0024] In a possible implementation manner of the first aspect, after generating the land resource change report based on the position coordinates, area data, and change trajectory information of the dynamic area, the method further includes:

[0025] According to the boundary vector data of the change area, mark the geometric shape and central coordinates of the change area on the digital map of the target area;

[0026] Based on the change trend prediction result, calculate the area expansion rate and type conversion probability of the change area within a preset future time period;

[0027] Integrate the geometric shape, central coordinates, area expansion rate, and type conversion probability into a visualization layer and overlay it on the interactive interface of the digital map.

[0028] In a possible implementation manner of the first aspect, the method further includes:

[0029] Real-time receive the latest image data transmitted by the satellite remote sensing system, perform spectral feature analysis and texture feature extraction on the latest image data to generate a real-time land feature map;

[0030] Perform difference detection between the real-time land feature map and the comprehensive land feature map to identify the abnormal areas of sudden land cover changes within the target area;

[0031] According to the geographical location and change type of the abnormal area, send a land resource anomaly warning signal to the monitoring terminal, and the warning signal includes the coordinate range of the abnormal area and recommended verification measures.

[0032] In a possible implementation manner of the first aspect, the land classification model is trained through the following steps:

[0033] Obtain a sample image data set containing multiple land cover types and label the true land type labels for each sample in the sample image data set;

[0034] Construct a deep convolutional neural network, input the sample image data set into the deep convolutional neural network for feature learning, and output the predicted land type probability distribution;

[0035] Calculate the cross-entropy loss value based on the difference between the predicted land type probability distribution and the true land type label, and use the backpropagation algorithm to optimize the weight parameters of the deep convolutional neural network;

[0036] When the cross-entropy loss value is lower than the preset threshold, stop training and use the optimized deep convolutional neural network as the land classification model.

[0037] In a possible implementation manner of the first aspect, the method further includes:

[0038] Receive the land survey feedback information submitted by the user through the interactive terminal, where the land survey feedback information includes suggestions for correcting the type annotation of a specific area in the land resource change report;

[0039] Adjust the classification weight parameters of the land classification model according to the type annotation correction suggestions, and re-classify the comprehensive land feature map of the target area to generate an updated land classification map;

[0040] Perform consistency verification on the updated land classification map with historical data, determine the optimization degree of the correction suggestions on the classification result, and feedback the verification result to the interactive terminal.

[0041] Combined with the second aspect of the present application, there is provided an AI analysis system for land resource survey. The AI analysis system for land resource survey includes a machine-readable storage medium and a processor. The machine-readable storage medium stores machine-executable instructions, and when the processor executes the machine-executable instructions, the AI analysis system for land resource survey implements the foregoing AI analysis method for land resource survey.

[0042] Combined with the third aspect of the present application, there is provided a computer-readable storage medium. The computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed, the foregoing AI analysis method for land resource survey is implemented.

[0043] Combined with any of the above aspects, in the embodiments of the present application, by obtaining multi-source remote sensing image data of the target area, analyzing its spectral and texture features to generate a comprehensive land feature map, using the land classification model to obtain an initial land classification map, and performing temporal comparison analysis in combination with historical data to determine the dynamic area of land cover change and generate a land resource change report. It also includes functions such as soil quality assessment, visualization of change areas, real-time monitoring of abnormal areas, training of land classification models, and adjustment of classification models according to user feedback, improving the efficiency and accuracy of land resource surveys. Description of the Drawings

[0044] To more clearly illustrate the technical solutions of the embodiments of the present application, the accompanying drawings required for the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present application and should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained in combination with these drawings.

[0045] Figure 1 Flow schematic diagram of the AI analysis method applied to land resource survey provided by the embodiments of the present application. Specific embodiments

[0046] To enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0047] The terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, device, product or terminal that includes a series of steps or units is not limited to the listed steps or units, but may optionally further include steps or units not listed, or may optionally further include other steps or units inherent to these processes, methods, products or terminals.

[0048] Referring to "embodiments" herein means that specific features, structures or characteristics described in connection with the embodiments can be included in at least one embodiment of the present invention. The appearance of this phrase in various positions in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art will explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0049] Figure 1 The flow schematic diagram of the AI analysis method applied to land resource survey provided by the embodiments of the present application is shown. It should be understood that in other embodiments, the order of some steps of the AI analysis method applied to land resource survey in this embodiment can be shared based on actual needs, or some of the steps can also be omitted or maintained. The details of the AI analysis method applied to land resource survey include:

[0050] Step S110: Obtain multi-source remote sensing image data of the target area. The multi-source remote sensing image data includes visible light images, infrared images, and radar images, and contains spectral features and texture features related to land cover types.

[0051] In an actual land resource survey scenario, assume that the target area is a large city and its surrounding areas. This target area contains various different land cover types, such as construction land in the city, surrounding cultivated land, forests, water bodies, and wetlands. To comprehensively obtain the land information of this area, satellite remote sensing technology can be used to obtain multi-source remote sensing image data. Among them, visible light images can reflect the reflection characteristics of surface objects in different visible light bands. For example, green vegetation has a relatively high reflectance in the green light band, while the reflectance characteristics of buildings in each visible light band are significantly different from those of vegetation and water bodies. Infrared images mainly reflect the thermal radiation situation of the surface. Due to differences in factors such as their material composition and water content, different land cover types have different thermal radiation intensities. For example, water bodies have a large heat capacity and show relatively low thermal radiation intensity in infrared images; while concrete buildings in the city absorb heat during the day and show relatively high thermal radiation intensity in infrared images. Radar images obtain surface information by transmitting and receiving radar waves, and its polarization scattering characteristics are closely related to factors such as the roughness of the surface and the height of vegetation. For example, due to the relatively high vegetation height and rough surface in forest areas, the polarization scattering characteristics in radar images are significantly different from those in flat cultivated land areas. These different types of remote sensing image data together contain rich spectral features and texture features related to land cover types, providing an ample data basis for subsequent analysis.

[0052] Step S120: Conduct spectral feature analysis and texture feature extraction on the multi-source remote sensing image data to generate a comprehensive land feature map of the target area. The comprehensive land feature map is used to characterize the spatial distribution and physical properties of land cover types.

[0053] First, divide the visible light image into multiple spectral bands, calculate the spectral reflectance curve of each spectral band, and extract the peak feature and slope feature of the spectral reflectance curve. In the target area mentioned in the previous example of this embodiment, taking the vegetation-covered area as an example, in the blue and red bands of visible light, its spectral reflectance curve has a specific shape. The reflectance in the blue band is relatively low, reaching a peak as the wavelength increases to the green band, and then decreasing in the red band. By calculating and analyzing the peak features and slope features of these spectral reflectance curves, different types of vegetation can be distinguished. For example, there may be slight differences in the peaks and slopes of the spectral reflectance curves between coniferous forests and broad-leaved forests. For non-vegetation areas such as buildings and roads, their spectral reflectance curves show different characteristics from those of vegetation in each band, and their peaks and slopes can reflect the characteristics of building materials.

[0054] Next, perform thermal radiation intensity analysis on the infrared image to determine the thermal radiation difference parameters of different land cover types within the target area. Still taking the water body and urban area in the target area as an example, due to its high specific heat capacity, the thermal radiation intensity of the water body is relatively stable at different times of the day, and the overall thermal radiation intensity is relatively low. While in the urban area, due to the heat generated by buildings, roads, and human activities, the thermal radiation intensity is high and there are obvious changes between day and night. By performing thermal radiation intensity analysis on the infrared image, the thermal radiation difference parameters of different land cover types can be accurately determined, which helps to distinguish different types of land.

[0055] Then, extract the polarization scattering characteristics of the radar image to obtain the surface roughness and vegetation height parameters. In the forest area of the target area, the tall trees increase the surface roughness, and the polarization scattering characteristics of the radar wave show strong scattering characteristics. While in the cultivated area, the surface is relatively flat and the polarization scattering characteristics are weak. By analyzing the polarization scattering characteristics of the radar image, the surface roughness and vegetation height parameters can be accurately obtained, and these parameters are also of great significance for distinguishing different land cover types.

[0056] Finally, fuse the peak feature of the spectral reflectance curve, the thermal radiation difference parameter, and the surface roughness parameter to generate an enhanced version of the spatial resolution of the comprehensive land feature map. In this process, various features extracted from the visible light image, infrared image, and radar image are fused, so that the comprehensive land feature map can more comprehensively and accurately reflect the spatial distribution and physical properties of land cover types within the target area. For example, in the generated comprehensive land feature map, the boundaries of different land cover types such as urban areas, forest areas, cultivated areas, and water body areas, as well as their respective characteristic distributions, can be clearly seen.

[0057] Step S130: Perform per-pixel classification processing on the comprehensive land feature map based on a preset land classification model to obtain an initial land classification map of the target area, where each pixel in the initial land classification map is labeled with a corresponding land type label.

[0058] For example, a pre-constructed and trained land classification model can be utilized. This land classification model is obtained by training with a large number of sample image datasets containing various land cover types. On the comprehensive land feature map of the target area, the land classification model classifies each pixel. Taking a pixel as an example, the land classification model can determine the land type corresponding to the pixel, such as construction land, cultivated land, forest, water body, or other types, based on various feature information of the pixel in the comprehensive land feature map, such as spectral reflectance characteristics, thermal radiation characteristics, surface roughness characteristics, etc. For instance, if a pixel shows a high peak in the green light band of spectral reflectance, low thermal radiation intensity, and low surface roughness, the model may label this pixel as the cultivated land type. After performing per-pixel classification processing on the entire comprehensive land feature map, an initial land classification map of the target area is finally obtained, and each pixel in this initial land classification map is labeled with the corresponding land type label.

[0059] Step S140: Conduct temporal contrast analysis on the land type labels in the initial land classification map according to historical land survey data, identify the dynamic areas of land cover change in the target area, and extract the change trajectory information of the dynamic areas.

[0060] First, obtain the historical land classification maps of the target area at at least three historical time points, and extract the land type distribution patterns corresponding to each historical time point. Assume that in this embodiment, the historical land classification maps of the target area 10 years ago, 5 years ago, and 1 year ago are obtained. 10 years ago, the urban scale of the target area was small, with large areas of cultivated land and forests around, and the water body area was relatively stable. 5 years ago, the city began to expand outwards, and some cultivated land was converted into construction land, and the forest area also decreased slightly. 1 year ago, the expansion speed of the city accelerated, more cultivated land and some forests were converted into construction land, and at the same time, some wetlands around the water body were also affected to a certain extent.

[0061] Then, calculate the spatial differences between the initial land classification map and the land type distribution patterns at each historical time point to determine the conversion areas where the land cover types have changed. For example, by comparing the current initial land classification map with the historical land classification map 10 years ago, it is found that the expanded part of the city center area is the conversion area where the land cover type has changed, and the original cultivated land or forest has now become construction land. Similarly, by comparing with the historical land classification maps 5 years ago and 1 year ago, other areas where land type conversions have occurred can also be determined.

[0062] Finally, classify the conversion types of the conversion regions to generate a set of change types including forest to cultivated land, cultivated land to construction land, and water body to wetland. In the target region, through comparative analysis, it is found that there are many areas where forests are cut down and reclaimed as cultivated land, and these areas are classified as the change type of forest to cultivated land; during the urban expansion process, a large amount of cultivated land is occupied for building houses, roads, etc., and these areas belong to the change type of cultivated land to construction land; in addition, due to the influence of human activities around some water bodies, the wetland area decreases and the water body area increases, and this situation is classified as the change type of water body to wetland. By classifying the conversion types of these conversion regions, the dynamic situation of land cover change in the target region can be understood more clearly, and the change trajectory information of these dynamic regions can be extracted, such as the specific time process and spatial range of a certain region changing from forest to cultivated land, etc.

[0063] Step S150: Generate a land resource change report based on the position coordinates, area data, and change trajectory information of the dynamic regions, where the land resource change report includes the boundary vector data of the change regions and the change trend prediction results.

[0064] After determining the dynamic regions of land cover change in the target region and obtaining the relevant position coordinates, area data, and change trajectory information, start generating the land resource change report. Taking the example of the reduction of cultivated land caused by urban expansion in the target region, the land resource change report will clearly indicate the boundary vector data of the urban expansion region (i.e., the change region), and these data can accurately depict the scope of cultivated land occupied by urban expansion. At the same time, according to the trend of historical land cover change, such as the speed of urban expansion and the rate of cultivated land reduction, etc., predict the future change trend. For example, if the average area of urban outward expansion in the past 5 years is 10 square kilometers per year, according to the current urban development plan and population growth trend, it is predicted that the city may continue to expand outward at a speed of 12 square kilometers per year in the next 5 years, and this expansion will mainly continue to occupy the surrounding cultivated land. The prediction results will be included in the land resource change report to provide a basis for the reasonable planning and management of land resources.

[0065] Next, step S120 can include:

[0066] Step S121: Divide the visible light image into multiple spectral bands, calculate the spectral reflectance curve of each spectral band, and extract the peak feature and slope feature of the spectral reflectance curve.

[0067] In the target area, the processing of visible light images is very meticulous. Suppose in this embodiment, the visible light band is divided into several main bands such as the blue light band (400 - 500 nm), the green light band (500 - 600 nm), and the red light band (600 - 700 nm). For an area mixed with different vegetation types, in the blue light band, the reflectivity of vegetation is relatively low because blue light is easily absorbed by substances such as chlorophyll in vegetation. When entering the green light band, the reflection effect of chlorophyll in vegetation on green light is enhanced, making the reflectivity reach a peak. For example, the peak reflectivity of grassland in this area in the green light band may be around 550 nm, while for the forest area, due to the more complex vegetation structure, its peak reflectivity in the green light band may be slightly higher, about 560 nm. Looking at the red light band again, the reflectivity of vegetation starts to decline. By accurately calculating the spectral reflectivity curve of each spectral band, the peak characteristics and slope characteristics can be further extracted. Taking grassland as an example, its slope from the blue light band to the green light band gradually increases, while the slope from the green light band to the red light band gradually decreases. These characteristics can help this embodiment distinguish different types of vegetation and can also be used to distinguish from other non-vegetation land cover types (such as buildings, water bodies, etc.). The reflectivity curve of buildings in each visible light band is relatively stable, without obvious peak and slope changes like those of vegetation.

[0068] Step S122: Analyze the thermal radiation intensity of the infrared image to determine the thermal radiation difference parameters of different land cover types within the target area.

[0069] Continuing with different land cover types in the target area as an example, when there is sufficient sunlight during the day, the asphalt roads in the city absorb a large amount of solar heat due to their dark color and show a high thermal radiation intensity in the infrared image. While the adjacent park green space has a relatively low thermal radiation intensity due to the transpiration of vegetation and its own heat dissipation characteristics. For water bodies, such as lakes, their thermal radiation intensity changes relatively little during the day and night and is generally at a low level. By conducting a detailed analysis of the thermal radiation intensity of the infrared image, the thermal radiation difference parameters corresponding to each pixel point or each small area can be accurately determined. For example, in an area at the junction of a city and the surrounding farmland, it can be found that the average thermal radiation intensity of the urban area may reach a certain value, while the average thermal radiation intensity of the farmland area is significantly lower than that of the urban area due to soil moisture evaporation and the influence of vegetation. These thermal radiation difference parameters provide an important basis for distinguishing different land cover types.

[0070] Step S123: Extract the polarization scattering characteristics of the radar image to obtain the surface roughness and vegetation height parameters.

[0071] In the forest area of the target region, the height and density of trees are important factors affecting the polarimetric scattering characteristics of radar images. In a tall and dense forest, when radar waves propagate through it, multiple scattering occurs, resulting in strong scattering characteristics in the polarimetric scattering characteristics. In this embodiment, a specific mathematical model can be established to invert the vegetation height in the forest area based on the polarimetric scattering data of the radar image. For example, for a mature forest, by analyzing the polarimetric scattering characteristics of the radar image, it can be calculated that its average vegetation height may be between 20 and 30 meters. For obtaining the surface roughness, taking the comparison between cultivated land and wasteland as an example, the cultivated land is cultivated and leveled, and the surface is relatively smooth, showing weak scattering characteristics in the polarimetric scattering characteristics of the radar image. While the wasteland, without artificial leveling, has more undulations and stones on the surface, and the scattering intensity of the polarimetric scattering characteristics is higher than that of the cultivated land. By extracting the polarimetric scattering characteristics of the radar image, the surface roughness and vegetation height parameters can be accurately obtained, and these parameters help to more accurately reflect the characteristics of land cover types in the comprehensive land feature map.

[0072] Step S124: Integrate the peak characteristics of the spectral reflectance curve, the thermal radiation difference parameter, and the surface roughness parameter to generate an enhanced version of the spatial resolution of the comprehensive land feature map.

[0073] In the entire data processing process of the target region, integrating the characteristics from these different sources is a very crucial step. For example, in a mixed region containing cities, forests, and cultivated land, the peak characteristics of the spectral reflectance curve obtained from visible light images (such as the reflectance peak of the forest in the green light band, the reflectance peak of the cultivated land in the red light band, etc.), the thermal radiation difference parameter obtained from infrared images (such as the high thermal radiation intensity of the city, the relatively low thermal radiation intensity of the forest, etc.), and the surface roughness parameter obtained from radar images (such as the high surface roughness of the forest, the low surface roughness of the cultivated land) can be integrated. Through a specific integration algorithm, these characteristics are combined according to certain weights and rules. The enhanced version of the spatial resolution of the comprehensive land feature map generated in this way can more clearly display the boundaries and internal structures between different land cover types. For example, in the boundary area between the city and the forest, the boundary that may have been blurred due to the limitations of a single data source can be accurately displayed in the integrated map, and it can accurately reflect the characteristic changes of the land cover types on both sides of the boundary.

[0074] In a possible implementation manner, the method may further include:

[0075] Step S125: Determine the humidity parameter, organic matter content parameter, and pH parameter related to soil quality according to the physical properties of different land types in the comprehensive land feature map.

[0076] In the target area, for cultivated land, its soil quality is very important. From the comprehensive land characteristic atlas, some physical property information related to cultivated land can be obtained, and then parameters related to soil quality can be inferred. For example, according to the characteristics of the spectral reflectance curve in certain bands, the soil humidity can be judged. If the reflectance in the near-infrared band is low, it may indicate that the soil humidity is high because water absorbs light in the near-infrared band. For the parameter of organic matter content, it can be inferred by analyzing the texture characteristics and color characteristics of the land. Soils with higher organic matter content usually have darker colors and relatively finer textures. For example, a cultivated land with long-term organic farming has a darker soil color than ordinary cultivated land, and the texture characteristics in the comprehensive land characteristic atlas also show a more delicate state, indicating that the soil organic matter content in this area is high. The pH parameter can be judged by the spectral characteristics related to certain mineral components in the soil. For example, the absorption peaks of certain minerals in specific bands can reflect the pH of the soil.

[0077] Step S126: Match the humidity parameter, organic matter content parameter, and pH parameter with the preset soil quality assessment criteria to generate the soil quality grade distribution map of the target area.

[0078] Assume that there is a set of preset soil quality assessment criteria in this embodiment. According to parameters such as soil humidity, organic matter content, and pH, the soil quality is divided into four grades: excellent, good, medium, and poor. In the target area, for each cultivated land area, the previously determined humidity parameter, organic matter content parameter, and pH parameter can be matched with this assessment criteria. For example, a certain cultivated land has moderate humidity, high organic matter content, and a pH close to neutral. According to the assessment criteria, the soil quality of this cultivated land is rated as "excellent" grade. By conducting such assessments on all cultivated lands and other land types related to soil quality in the entire target area, a soil quality grade distribution map can be generated. In this soil quality grade distribution map, it can be clearly seen which areas have good soil quality and which areas have poor soil quality, providing a basis for the rational use of land resources and soil improvement.

[0079] Step S127: Conduct a spatial overlay analysis of the soil quality grade distribution map and the changed areas in the land resource change report to determine the priority areas for land degradation or improvement.

[0080] In the target area, when urban expansion leads to the occupation of some cultivated land, the occupied cultivated land is marked as a changed area in the land resource change report. At the same time, there can be a distribution map of soil quality grades. By performing a spatial overlay analysis on the two, it can be found that some of the occupied cultivated land originally had very good soil quality (such as the soil quality grade being "excellent"). If these areas are occupied by urban construction, they belong to the high-priority areas of land degradation because they originally had high agricultural production value. For some areas with relatively poor original soil quality (such as the soil quality grade being "poor"), if their soil quality is improved through some measures (such as land improvement and soil amendment) during the land cover change process, then these areas are the priority areas for land improvement. Through this spatial overlay analysis, more scientific and reasonable strategies can be formulated for the management and protection of land resources.

[0081] In one possible implementation manner, step S140 may include:

[0082] Step S141: Obtain the historical land classification maps of the target area at at least three historical time points, and extract the land type distribution patterns corresponding to each historical time point.

[0083] In the example of the target area, the historical land classification maps at three historical time points, namely 10 years ago, 5 years ago, and 1 year ago, may have been determined. For the historical land classification map 10 years ago, the urban area was relatively small at that time, mainly concentrated in certain areas, with large areas of cultivated land and forests around, the water body distribution was relatively stable, and the wetland also maintained a certain area. This is the land type distribution pattern corresponding to 10 years ago, where each land type has its specific distribution range and shape in space. The historical land classification map 5 years ago shows that the city began preliminary expansion, extending along the main traffic arteries to the surrounding areas. Some of the cultivated land on the original urban fringe was incorporated into the urban area and became construction land, and a small part of the forest around the city was also cut down for the construction of small industrial parks or residential areas. At this time, the areas of the water body and the wetland did not change much, but in some local areas, due to the influence of agricultural irrigation and urban water use, the distribution and water volume of the water body had some subtle adjustments. This is the land type distribution pattern corresponding to 5 years ago, which reflects the spatial layout and interrelationships of the land cover types in the target area at this time point.

[0084] The land classification historical map from one year ago indicates that the urban expansion rate has further accelerated. More arable land has been converted into construction land, and the commercial and residential areas of the city have continuously expanded outwards. The forest area continues to decrease, mainly due to urban expansion and some illegal logging activities. In terms of water bodies, some small rivers and ponds have dried up or shrunk in area due to the construction of the urban drainage system and the change of land use patterns. Wetlands have also been threatened to a certain extent and their area has decreased. This is the land type distribution pattern corresponding to one year ago, reflecting the distribution characteristics of land cover types in the target area in a relatively recent period.

[0085] Step S142: Calculate the spatial differences between the initial land classification map and the land type distribution patterns at each historical time point to determine the conversion areas where the land cover types have changed.

[0086] Taking a certain area in the target area as an example, the current initial land classification map can be compared with the land type distribution pattern from 10 years ago. Suppose in the initial land classification map, a certain area is marked as construction land, but in the land classification historical map from 10 years ago, this area was arable land. Through precise spatial difference calculation, it can be determined that this area is the conversion area where the land cover type has changed. This calculation involves comparing the positions and land type labels of each pixel point in the two maps, calculating the set of pixel points that have changed spatially, and the area composed of these pixel points is the place where the land cover type has changed.

[0087] Similarly, the spatial differences between the initial land classification map and the land type distribution patterns from 5 years ago and 1 year ago can be calculated. For example, in the comparison with that from 5 years ago, it is found that an area that was originally forest on the urban fringe is now marked as construction land and some wasteland (due to the undeveloped state after deforestation) in the current initial land classification map. This area is the conversion area where the land cover type has changed during these 5 years. In the comparison with that from 1 year ago, the conversion areas such as the arable land newly occupied by urban expansion and the areas where the forest has been further damaged can also be determined.

[0088] Step S143: Classify the conversion types of the conversion areas to generate a set of change types including forest to arable land, arable land to construction land, and water body to wetland.

[0089] Within the target area, after determining multiple conversion areas through the previous steps, the conversion types of these conversion areas are classified in detail. For example, in some mountainous areas around the city, due to the need for agricultural development, some forests have been cut down and reclaimed as arable land, and these areas belong to the type of forest to arable land. During the urban expansion process, a large amount of arable land has been used for building urban infrastructure, housing, and commercial facilities, etc. This part of the conversion area belongs to the type of arable land to construction land.

[0090] For the conversion of water bodies and wetlands, in some corners of the target area, due to human reclamation activities or water conservancy project construction, some areas that were originally water bodies gradually dried up, and the surrounding wetland vegetation was also damaged. The area of the water body decreased, and the area of the wetland also decreased accordingly; while in other places, due to the enhanced awareness of wetland protection and the adoption of some restoration measures, the land around some water bodies gradually evolved into wetlands. The change types of these areas belong to water body to wetland (negative transformation) and wetland to water body (positive transformation) respectively. By classifying the conversion types of all conversion areas, a complete set containing various change types can be generated, and this set can comprehensively reflect the dynamic characteristics of land cover change in the target area.

[0091] Furthermore, in a possible implementation manner, step S150 may include:

[0092] Step S151: According to the boundary vector data of the change area, mark the geometric shape and central coordinates of the change area on the digital map of the target area.

[0093] Taking the change area of urban expansion occupying cultivated land as an example on the digital map of the target area. According to the boundary vector data of the change area in the previously generated land resource change report, these data accurately describe the boundary position of the change area. The geometric shape of the change area can be accurately drawn on the digital map. For example, if it is an irregular polygon, the vertices and sides of the polygon are depicted according to the vector data. At the same time, calculate the central coordinates of the change area, and the central coordinates can be obtained by performing specific mathematical calculations on the vertex coordinates of the polygon. For example, take the centroid coordinates of the polygon as the central coordinates. Mark the central coordinate point on the digital map, so that the position and shape characteristics of the change area can be intuitively displayed on the map.

[0094] Step S152: Based on the change trend prediction result, calculate the area expansion rate and type conversion probability of the change area within a preset future time period.

[0095] Continuing with the example of the changing area of urban expansion occupying arable land, if it is predicted based on previous land resource change reports that the city will continue to expand outwards in the next 5 years and mainly occupy the surrounding arable land. According to the historical data of past urban expansion, such as the area data of urban expansion in the past 5 years, calculate the average annual area expansion rate. Suppose the city has occupied 10 square kilometers of arable land on average per year in the past 5 years, then this 10 square kilometers / year is the past area expansion rate. Based on existing urban development plans, population growth trends, land use policies and other factors, predict the area expansion rate for the next 5 years. If it is predicted that the urban expansion speed will accelerate in the next 5 years, it may occupy 12 square kilometers of arable land on average per year, which is the area expansion rate within the preset future time period.

[0096] For the calculation of the type conversion probability, the possibility of conversion between different land types within the changing area can be considered. For example, during the process of urban expansion, in addition to the main conversion type of arable land to construction land, there may also be a small amount of cases where arable land is converted to wasteland (due to land idleness or temporary undevelopment). Based on historical data and various current influencing factors, calculate that the probability of arable land being converted to construction land may be 0.8 (indicating an 80% possibility), and the probability of arable land being converted to wasteland may be 0.2 (indicating a 20% possibility). These calculation results of the area expansion rate and type conversion probability provide a quantitative basis for the planning and management of land resources.

[0097] Step S153: Integrate the geometric shape, central coordinates, area expansion rate, and type conversion probability into a visualization layer and overlay it on the interactive interface of the digital map.

[0098] In the interactive interface of the digital map of the target area, the geometric shape (such as a polygon) of the previously marked changing area, the calculated central coordinates, area expansion rate, and type conversion probability and other information can be integrated into a visualization layer. For the geometric shape, it is clearly displayed on the map with a specific color and line; the central coordinates are displayed on the map with a special marker (such as a small circle or a cross); the area expansion rate can be represented by adding a data label in the visualization layer or by the depth of different colors to indicate the magnitude of the expansion rate (for example, the greater the expansion rate, the darker the color); the type conversion probability can also be represented by similar data labels or color differentiation to indicate the possibility of different type conversions. Then overlay this visualization layer on the interactive interface of the digital map, so that land resource managers or other relevant personnel can intuitively obtain various detailed information about the changing area when viewing the digital map, which is convenient for land resource planning, decision-making, and management.

[0099] Furthermore, the method may further include:

[0100] Step S161: Receive the latest image data transmitted by the satellite remote sensing system in real time, perform spectral feature analysis and texture feature extraction on the latest image data, and generate a real-time land feature map.

[0101] During the real-time monitoring of land resources in the target area, the satellite remote sensing system continuously transmits the latest image data. These image data contain the land information of the target area at the current moment. Processing these latest image data is similar to the processing of multi-source remote sensing image data before. First, perform spectral feature analysis, divide the visible light image part into multiple spectral bands, calculate the spectral reflectance curve of each spectral band, and extract its peak feature and slope feature; perform thermal radiation intensity analysis on the infrared image to determine the thermal radiation difference parameters of different land cover types; perform polarization scattering feature extraction on the radar image to obtain surface roughness and vegetation height parameters. Then fuse the features extracted from different image types to generate a real-time land feature map. For example, at a certain moment, the satellite transmits the latest image data containing the urban, forest, cultivated land, and water body in the target area. After the above processing, the characteristics of each land cover type at this moment can be accurately reflected in the real-time land feature map, such as the performance of buildings in the city in terms of spectral reflectance curve, thermal radiation intensity, and polarization scattering characteristics, and the vegetation height, thermal radiation, and spectral reflectance characteristics of the forest area.

[0102] Step S162: Perform difference detection between the real-time land feature map and the comprehensive land feature map to identify the abnormal areas of sudden land cover changes in the target area.

[0103] Compare the newly generated real-time land feature map with the previously generated comprehensive land feature map. Taking the forest area in the target area as an example, if the characteristics such as vegetation height and spectral reflectance of a certain forest area in the comprehensive land feature map are normal, but in the real-time land feature map, it is found that the vegetation height of this forest area suddenly decreases and the spectral reflectance curve also changes greatly, this may be due to forest fires or large-scale illegal logging. The area with a large difference in the comparison of the two maps is the abnormal area of sudden land cover change. Another example is in the cultivated land area. If it is found in the real-time land feature map that the originally flat cultivated land appears in a large area of irregular shapes and the spectral reflectance curve is very different from the characteristics in the comprehensive land feature map, it may be due to natural disasters such as floods and mudslides or illegal occupation by humans. Through this difference detection, the abnormal areas of sudden land cover changes in the target area can be quickly and accurately identified.

[0104] Step S163: Send a land resource anomaly warning signal to the monitoring terminal according to the geographical location and change type of the anomaly area, where the warning signal includes the coordinate range of the anomaly area and recommended verification measures.

[0105] After identifying the anomaly area, determine its specific location according to its geographical location within the target area, such as its longitude and latitude coordinate range. If the anomaly is caused by a possible forest fire in a forest area and the change type is vegetation damage, the land resource anomaly warning signal sent to the monitoring terminal will include the precise coordinate range of the anomaly area and give recommended verification measures, such as dispatching drones or ground personnel to the area to conduct fire hazard inspections or forest damage investigations. If the anomaly is in an arable land area due to possible natural disasters or illegal occupation, the warning signal will also include the coordinate range, and the recommended verification measures may include further detailed analysis using satellite images, dispatching land supervision personnel to conduct on-site inspections, etc. This can timely notify relevant personnel to handle the anomaly of land resources and protect the safety and reasonable utilization of land resources.

[0106] Further, the method may further include:

[0107] Step S171: Obtain a sample image dataset containing multiple land cover types and label each sample in the sample image dataset with a true land type label.

[0108] In the process of constructing a land classification model, it is first necessary to collect a large number of sample image datasets. These sample images should cover various land cover types that may appear in the target area, such as different types of forests (coniferous forests, broad-leaved forests, etc.), arable lands with different planting patterns (paddy fields, dry lands, etc.), urban construction lands of various scales, water bodies of different forms (rivers, lakes, ponds, etc.), and wetlands in different ecological environments. For each sample image, the true land type label can be marked through on-site inspections, historical data queries, or high-precision reference images. For example, if a sample image shows a forest area full of pine trees, the land type label of "coniferous forest" can be marked for this sample; if it is a paddy field growing rice, the land type label of "paddy field" can be marked. By marking a large number of sample images in this way, an accurate reference basis is provided for subsequent model training.

[0109] Step S172: Construct a deep convolutional neural network, input the sample image dataset into the deep convolutional neural network for feature learning, and output the predicted land type probability distribution.

[0110] Construct a deep convolutional neural network, which includes multiple convolutional layers, pooling layers, fully connected layers and other structures. Input the previously labeled sample image dataset into this deep convolutional neural network. In the convolutional layer of the network, the convolutional kernel slides on the sample image to extract local features in the image, such as edge, texture and other features. The pooling layer downsamples the features extracted by the convolutional layer, reducing the amount of data while retaining important feature information. After being processed by multiple convolutional layers and pooling layers, the features are integrated in the fully connected layer, and finally the predicted probability distribution of land types is output. For example, for an input sample image, the deep convolutional neural network may output that the probability of this sample image belonging to "coniferous forest" is 0.3, the probability of belonging to "broad-leaved forest" is 0.2, the probability of belonging to "paddy field" is 0.1, the probability of belonging to "dry land" is 0.15, the probability of belonging to "urban construction land" is 0.1, the probability of belonging to "water body" is 0.05, the probability of belonging to "wetland" is 0.1, etc., which is the predicted probability distribution of land types.

[0111] Step S173: Calculate the cross-entropy loss value according to the difference between the predicted probability distribution of land types and the true land type label, and use the backpropagation algorithm to optimize the weight parameters of the deep convolutional neural network.

[0112] Calculate the cross-entropy loss value based on the predicted probability distribution of land types output by the network and the true land type label originally marked for the sample image. The cross-entropy loss value is an index to measure the difference between the predicted result and the true result. For example, if the true label of a sample image is "coniferous forest", and the probability of "coniferous forest" in the predicted result output by the network is 0.3, then there is a certain difference between this predicted result and the true result, and the cross-entropy loss value corresponding to this difference is calculated through a specific formula. Then, use the backpropagation algorithm. According to the calculated cross-entropy loss value, starting from the output layer of the network, gradually adjust the weight parameters of the network towards the input layer direction. The adjustment of the weight parameters is to make the predicted result of the network closer to the true result and reduce the cross-entropy loss value. For example, if after adjusting the weight parameters of a certain convolutional kernel in a certain convolutional layer, the predicted probability of the network for the "coniferous forest" sample image is closer to 1 (indicating more accurately predicting as "coniferous forest"), then the adjustment of this weight parameter is effective.

[0113] Step S174: When the cross-entropy loss value is lower than the preset threshold, stop training and use the optimized deep convolutional neural network as the land classification model.

[0114] During the training process, a preset threshold can be set, such as 0.01. When the calculated cross-entropy loss value is lower than this preset threshold, it indicates that the prediction result of the network is already close enough to the true result, and at this time, the training is stopped. The optimized deep convolutional neural network is used as the final land classification model. This land classification model can be used to perform per-pixel classification processing on the comprehensive land feature map of the target area, such as accurately predicting the land type of each pixel point in the target area in practical applications, providing reliable classification results for land resource surveys.

[0115] Furthermore, the method may further include:

[0116] Step S181: Receive the land survey feedback information submitted by the user through the interactive terminal, where the land survey feedback information includes a type annotation correction suggestion for a specific area in the land resource change report.

[0117] During the land resource survey process, the user (who may be a land resource management personnel, on-site inspection personnel, or other relevant experts) submits land survey feedback information through the interactive terminal (such as a computer, tablet, or dedicated land survey equipment, etc.). For example, when viewing the land resource change report, the user finds that a certain area is marked as "wasteland" in the report, but based on the user's own on-site inspection or other reliable information, believes that this area is actually "cultivated land during fallow period", so the user submits land survey feedback information containing this correction suggestion (correcting "wasteland" to "cultivated land during fallow period") through the interactive terminal.

[0118] Step S182: Adjust the classification weight parameters of the land classification model according to the type annotation correction suggestion, and re-classify the comprehensive land feature map of the target area to generate an updated land classification map.

[0119] When receiving the user's correction suggestion, adjust the classification weight parameters of the land classification model according to this correction suggestion. For example, if the user's correction suggestion is about the type annotation of "wasteland" and "cultivated land during fallow period", then in the land classification model, the feature extraction and classification weight parameters related to these two land types will be adjusted. The adjustment method may be to increase the weight of the features related to "cultivated land during fallow period" (such as specific spectral reflectance features, texture features, etc.) and reduce the weight of the features related to "wasteland". Then, use the adjusted land classification model to re-classify the comprehensive land feature map of the target area. In this process, the land classification model can re-classify each pixel point according to the new weight parameters, and finally generate an updated land classification map. In this updated land classification map, the area that was previously mislabeled as "wasteland" may be correctly labeled as "cultivated land during fallow period".

[0120] Step S183: Perform consistency verification on the updated land classification map and historical data, determine the optimization degree of the classification result by the correction suggestions, and feedback the verification result to the interactive terminal.

[0121] Perform consistency verification on the updated land classification map and previous historical data (such as historical land classification maps, historical land survey records, etc.). For example, compare the land type markings of the same area in the updated land classification map and the land classification historical map 5 years ago. If in the updated land classification map, a certain area is correctly marked as "cultivated land during fallow period", and in the historical map 5 years ago, this area was also "cultivated land" (just in a different cultivation state), this indicates that the correction suggestions have a positive optimization effect on the classification result. By comparing and analyzing multiple areas and multiple historical data, determine the overall optimization degree of the classification result by the correction suggestions. Finally, feedback this verification result (such as qualitative results like "high", "medium", "low", etc. or specific quantitative data) to the interactive terminal, so that users can understand the impact of their feedback information on the land classification result, in order to make more accurate judgments and decisions in subsequent land resource surveys and management.

[0122] In the above embodiments, the AI analysis system applied to land resource surveys for executing the above method embodiments has at least one processor, a control module (chipset) coupled to at least one of the (at least one) processors, a memory coupled to the control module, a non-volatile memory (NVM) / storage device coupled to the control module, at least one loading to / output device coupled to the control module, and a network interface coupled to the control module.

[0123] The processor may include at least one single-core or multi-core processor, and the processor may include any combination of general-purpose processors or dedicated processors (such as graphics processors, application processors, baseband processors, etc.). For some alternative embodiments, the AI analysis system applied to land resource surveys can be an electronic device such as the gateway described in the embodiments of the present application.

[0124] For some alternative embodiments, the AI analysis system applied to land resource surveys may include at least one computer-readable medium having instructions (such as a memory or NVM / storage device) and at least one processor integrated with the at least one computer-readable medium and configured to execute the instructions to implement modules so as to perform the actions described in the present disclosure.

[0125] For one embodiment, the control module may include any suitable interface controller to provide any suitable interface to at least one of the (at least one) processors and / or any suitable device or component communicating with the control module.

[0126] The control module may include a memory controller module to provide an interface to the memory. The memory controller module may be a hardware module, a software module, and / or a firmware module.

[0127] The memory may be used to load and store data and / or instructions for, e.g., an AI analysis system applied to land resource survey. For one embodiment, the memory may include any suitable volatile memory, e.g., suitable DRAM.

[0128] For one embodiment, the control module may include at least one load-to / output controller to provide an interface to the NVM / storage device and the (at least one) load-to / output device.

[0129] For example, the NVM / storage device may be used to store data and / or instructions. The NVM / storage device may include any suitable non-volatile memory (e.g., flash memory) and / or may include any suitable (at least one) non-volatile storage device (e.g., at least one hard disk drive (HDD), at least one compact disc (CD) drive, and / or at least one digital versatile disc (DVD) drive).

[0130] The NVM / storage device may include storage resources that are physically part of the device on which the AI analysis system applied to land resource survey is installed, or it may be accessible by the device without being part of the device. For example, the NVM / storage device may be accessed via the (at least one) load-to / output device based on a network.

[0131] The (at least one) load-to / output device may provide an interface for the AI analysis system applied to land resource survey to communicate with any other suitable device. The load-to / output device may include communication components, spelling components, sensor components, etc. The network interface may provide an interface for the AI analysis system applied to land resource survey to communicate based on at least one network. The AI analysis system applied to land resource survey may wirelessly communicate with at least one component of the wireless network based on any prior and / or protocol in at least one wireless network prior and / or protocol, e.g., access a wireless network based on a communication prior.

[0132] For one embodiment, at least one of the (at least one) processors may be loaded with the logic of at least one controller of the control module (e.g., a memory controller module). For one embodiment, at least one of the (at least one) processors may be loaded with the logic of at least one controller of the control module to form a system-level load. For one embodiment, at least one of the (at least one) processors may be integrated with the logic of at least one controller of the control module on the same die. For one embodiment, at least one of the (at least one) processors may be integrated with the logic of at least one controller of the control module on the same die to form a system-on-chip (SoC).

[0133] The embodiments of the present application have been introduced in detail above. Specific examples are used in this article to elaborate on the principles and implementation manners of the present application. The descriptions of the above embodiments are only used to help understand the method and its core idea of the present application; at the same time, for those of ordinary skill in the art, based on the idea of the present application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present application.

[0134] An embodiment of the present invention discloses a computer-readable storage medium that stores a computer program for electronic data exchange, wherein the computer program causes a computer to execute the steps in the AI analysis method applied to land resource survey described in the foregoing embodiments.

[0135] An embodiment of the present invention discloses a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to execute the steps in the AI analysis method applied to land resource survey described in the foregoing embodiments.

[0136] The device embodiments described above are only illustrative. The modules described as separate components may or may not be physically separated. The components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed to multiple modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative labor.

[0137] Through the above specific description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the above technical solutions, in essence, or the part that contributes to the prior art can be embodied in the form of a software product, and this computer software product can be stored in a computer-readable storage medium. The storage medium includes read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc memories, magnetic disk memories, tape memories, or any other medium that can be used for a computer to read or store data.

[0138] Finally, it should be noted that: the above-disclosed is only the preferred embodiment of the present invention, which is only used to illustrate the technical solution of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An AI analysis method applied to land resource survey, characterized in that, The method includes: Obtaining multi-source remote sensing image data of the target area, where the multi-source remote sensing image data includes visible light images, infrared images, and radar images, and contains spectral features and texture features related to land cover types; Performing spectral feature analysis and texture feature extraction on the multi-source remote sensing image data to generate a comprehensive land feature map of the target area, where the comprehensive land feature map is used to characterize the spatial distribution and physical properties of land cover types; Performing per-pixel classification processing on the comprehensive land feature map based on a preset land classification model to obtain an initial land classification map of the target area, where each pixel point in the initial land classification map is labeled with a corresponding land type label; Performing temporal contrast analysis on the land type labels in the initial land classification map according to historical land survey data to identify dynamic areas of land cover change in the target area, and extracting change trajectory information of the dynamic areas; Generating a land resource change report based on the position coordinates, area data, and change trajectory information of the dynamic areas, where the land resource change report includes boundary vector data of the changed area and a change trend prediction result.

2. The AI analysis method applied to land resource survey according to claim 1, characterized in that, The performing spectral feature analysis and texture feature extraction on the multi-source remote sensing image data to generate the comprehensive land feature map of the target area includes: Dividing the visible light image into multiple spectral bands, calculating the spectral reflectance curve of each spectral band, and extracting the peak feature and slope feature of the spectral reflectance curve; Performing thermal radiation intensity analysis on the infrared image to determine the thermal radiation difference parameters of different land cover types in the target area; Performing polarization scattering feature extraction on the radar image to obtain surface roughness and vegetation height parameters; Fusing the peak feature of the spectral reflectance curve, the thermal radiation difference parameter, and the surface roughness parameter to generate an enhanced version of the spatial resolution of the comprehensive land feature map.

3. The AI analysis method applied to land resource survey according to claim 2, characterized in that, The method further includes: Determining humidity parameters, organic matter content parameters, and pH parameters related to soil quality according to the physical properties of different land types in the comprehensive land feature map; Matching the humidity parameters, organic matter content parameters, and pH parameters with a preset soil quality assessment standard to generate a soil quality grade distribution map of the target area; Performing spatial overlay analysis on the soil quality grade distribution map and the changed area in the land resource change report to determine the priority areas of land degradation or improvement.

4. The AI analysis method applied to land resource survey according to claim 1, characterized in that The performing temporal contrast analysis on the land type labels in the initial land classification map according to historical land survey data to identify the dynamic areas of land cover change in the target area includes: Obtaining historical land classification maps of the target area at at least three historical time points, and extracting the land type distribution patterns corresponding to each historical time point; Performing spatial difference calculation on the initial land classification map and the land type distribution patterns of each historical time point to determine the conversion areas where land cover types have changed; Classify the conversion types of the conversion areas to generate a set of change types including forest to arable land, arable land to construction land, and water body to wetland.

5. The AI analysis method applied to land resource survey according to claim 1, wherein After generating the land resource change report based on the position coordinates, area data, and change trajectory information of the dynamic area, the method further includes: According to the boundary vector data of the change area, mark the geometric shape and central coordinates of the change area on the digital map of the target area; Based on the change trend prediction result, calculate the area expansion rate and type conversion probability of the change area within a preset future time period; Integrate the geometric shape, central coordinates, area expansion rate, and type conversion probability into a visualization layer and overlay it on the interactive interface of the digital map.

6. The AI analysis method applied to land resource survey according to claim 1, wherein The method further includes: Receive the latest image data transmitted by the satellite remote sensing system in real time, perform spectral feature analysis and texture feature extraction on the latest image data, and generate a real-time land feature map; Perform difference detection between the real-time land feature map and the comprehensive land feature map to identify the abnormal areas of sudden land cover changes in the target area; According to the geographical location and change type of the abnormal area, send a land resource abnormal warning signal to the monitoring terminal, and the warning signal includes the coordinate range of the abnormal area and the recommended verification measures.

7. The AI analysis method applied to land resource survey according to claim 1, characterized in that, The land classification model is trained through the following steps: Obtain a sample image data set containing multiple land cover types, and label each sample in the sample image data set with a true land type label; Construct a deep convolutional neural network, input the sample image data set into the deep convolutional neural network for feature learning, and output the predicted land type probability distribution; Calculate the cross-entropy loss value according to the difference between the predicted land type probability distribution and the true land type label, and use the backpropagation algorithm to optimize the weight parameters of the deep convolutional neural network; When the cross-entropy loss value is lower than the preset threshold, stop training and use the optimized deep convolutional neural network as the land classification model.

8. The AI analysis method applied to land resource survey according to claim 1, characterized in that, The method further includes: Receive the land survey feedback information submitted by the user through the interactive terminal, and the land survey feedback information includes the type annotation correction suggestions for specific areas in the land resource change report; Adjust the classification weight parameters of the land classification model according to the type annotation correction suggestions, and re-classify the comprehensive land feature map of the target area to generate an updated land classification map; Perform consistency verification on the updated land classification map and the historical data to determine the optimization degree of the classification result by the correction suggestions, and feedback the verification result to the interactive terminal.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores machine-executable instructions, and when the machine-executable instructions are executed by a computer, the AI analysis method for land resource survey described in any one of claims 1-8 is implemented.

10. An AI analysis system applied to land resource survey, characterized in that, It includes a processor and a computer-readable storage medium, and machine-executable instructions are stored in the computer-readable storage medium. When the machine-executable instructions are executed by a computer, the AI analysis method for land resource survey described in any one of claims 1-8 is implemented.

Images