Natural resource dynamic monitoring system and method based on multi-source remote sensing image

Abstract

The invention discloses a natural resource dynamic monitoring system and method based on a multi-source remote sensing image. The system comprises a multi-source data acquisition module, a preprocessing module, a feature extraction module, a dynamic monitoring module, a visual output module and a distributed storage module. According to the method, the precision and timeliness of natural resource change detection are remarkably improved by constructing a multi-source remote sensing image intelligent fusion framework. The system adopts a parallelization preprocessing assembly line to effectively eliminate radiation and geometric differences among multi-source data; a feature fusion module based on deep learning automatically extracts multi-scale change features, and overcomes the environmental adaptability defect of a traditional threshold method; and the dynamic monitoring module realizes automatic extraction and classification of sub-meter change pattern spots, and the monitoring efficiency is improved. Meanwhile, the system supports collaborative analysis of multi-platform data of satellites, aviation, unmanned aerial vehicles and the like, the cloud shielding influence is greatly reduced, and total-factor and high-frequency decision support is provided for natural resource supervision.

Description

Technical Field

[0001] The present application relates to the technical field of monitoring systems, and in particular to a natural resource dynamic monitoring system and method based on multi-source remote sensing images. Background Art

[0002] Current natural resource monitoring mainly relies on a single satellite data source, which has problems such as long data update cycle, severe cloud obstruction and interference, and incomplete spectral information.

[0003] Among traditional methods, for example, the Chinese invention patent with application number 2021103077857, named A method and device for automatic extraction of mines and solid waste landfills based on remote sensing data, realizes the automatic extraction of mines and solid waste landfills when implemented, greatly reduces labor costs and time costs, improves work efficiency, and provides technical support for environmental pollution source supervision; the application number is 2025105355573, named A dynamic monitoring of urban green space coverage, realizes high-precision dynamic monitoring of green space coverage when implemented, and improves monitoring efficiency and accuracy.

[0004] However, when the vegetation around the target area changes with the seasons, the above technical solution cannot identify it, which may lead to insufficient detection accuracy due to small color differences. Summary of the Invention

[0005] To this end, the present application provides a natural resource dynamic monitoring system and method based on multi-source remote sensing images to solve the problem of inaccurate recognition accuracy when seasonal changes occur.

[0006] In order to achieve the above objectives, this application provides the following technical solutions:

[0007] First aspect:

[0008] A natural resource dynamic monitoring system based on multi-source remote sensing images, including a multi-source data acquisition module, a pre-processing module, a feature extraction module, a dynamic monitoring module, a visual output module, and a distributed storage module;

[0009] The multi-source data acquisition module is used to receive satellite remote sensing images, aerial remote sensing images and UAV remote sensing images. The satellite remote sensing images include multispectral data and synthetic aperture radar data, and the aerial remote sensing images include high-resolution RGB images and lidar point cloud data.

[0010] The preprocessing module is connected to the multi-source data acquisition module, and the preprocessing module includes a radiation correction unit, a geometric correction unit, and a registration and fusion unit. The radiation correction unit uses a sensor response model to eliminate image radiation distortion, the geometric correction unit performs terrain correction based on a digital elevation model, and the registration and fusion unit realizes spatial alignment of multi-source images through a feature point matching algorithm.

[0011] The feature extraction module includes a temporal feature analysis unit and a change detection unit. The temporal feature analysis unit uses a three-dimensional convolutional neural network to extract the spatiotemporal change characteristics of resource coverage, and the change detection unit identifies the changed area by constructing a dual-temporal remote sensing image feature difference matrix.

[0012] The feature extraction module also includes a same-season comparison submodule, a meteorological interface submodule, and a regional strategy engine. The same-season comparison submodule is used to call historical same-season images in the distributed storage unit, calculate the inter-annual difference value of the vegetation index, and generate a seasonal mask. The meteorological interface submodule is used to access the temperature and humidity data stream of the meteorological bureau in real time. When continuous freeze-thaw cycles are monitored in the target area, the cold zone change detection parameter set is automatically switched. The regional strategy submodule is used to preload the latitude zone rule library, activate the fallen leaf compensation algorithm in the temperate zone, and enable the snow scatter correction model in the cold zone.

[0013] The dynamic monitoring module is connected to the feature extraction module. The dynamic monitoring module includes a change patch generation submodule and a resource classification submodule. The change patch generation submodule aggregates the change pixels into vector patches through an edge connection algorithm, and the resource classification submodule identifies the patches based on the random forest model.

[0014] The visualization output module is used to generate dynamic monitoring maps containing information on change locations, change areas, and land type conversions, and output change heat maps and statistical reports;

[0015] The distributed storage module uses a NoSQL database to store multi-temporal remote sensing images and change detection results, and accelerates data retrieval through spatial indexing.

[0016] Preferably, the pre-processing module further comprises a cloud detection unit, which uses a combination of band threshold segmentation and texture analysis to identify and mask cloud coverage areas.

[0017] Preferably, the feature extraction module includes a multi-scale fusion unit, which performs weighted fusion of visible light, near infrared and radar feature maps through a channel attention mechanism.

[0018] Preferably, the dynamic monitoring module further includes a change verification unit, which evaluates the credibility of the change results by superimposing historical monitoring results and manually annotated samples.

[0019] Preferably, a task scheduler is further included, which dynamically allocates GPU computing resources according to data priority, wherein the change detection task enjoys the highest computing priority.

[0020] Second aspect:

[0021] The method for dynamic monitoring of natural resources based on multi-source remote sensing images includes the following steps:

[0022] Step S1: synchronously acquiring multi-source images of the target area, the multi-source images including satellite images, aerial images, and drone images, wherein the satellite image time series covers at least three growing seasons;

[0023] Step S2: Perform radiometric normalization on the multi-source images, unify the radiometric benchmarks of different sensors using the histogram matching method, and achieve sub-pixel registration between images through affine transformation to generate the processed image;

[0024] Step S3: Construct a time series image stack, use principal component analysis to extract the first three principal components in the processed image, and use them as change-sensitive features. Simultaneously load the historical image database of the same season, and generate seasonal change marker areas by calculating the difference in vegetation index between the current image and the historical image of the same season.

[0025] Step S4: inputting the current phase remote sensing image and the reference phase remote sensing image into a pre-trained change detection model to output a pixel-level change probability map;

[0026] Step S5: using a region growing algorithm to aggregate the changed pixels to form an initial change patch, and combining it with morphological filtering to remove isolated patches caused by noise;

[0027] Step S6: Based on the spectral characteristics, texture characteristics and geometric characteristics of the spots, a classifier is used to determine the natural resource type of the changed spots;

[0028] Step S7: Overlay the change map with the natural resource background database to automatically generate a statistical report on changes in cultivated land occupation, forest land reduction, and water body expansion.

[0029] Preferably, the change detection model in step S4 adopts a U-Net network structure, the encoder part of which includes a ResNet50 backbone network, and the decoder part integrates multi-level feature maps.

[0030] Preferably, the classifier in step S6 adopts the XGBoost algorithm, and the input features include the NDVI mean, blue band standard deviation and edge density index of the image spot.

[0031] Compared with the prior art, this application has at least the following beneficial effects:

[0032] When the present invention is implemented, it can filter out periodic changes such as the winter withering and summer blooming of deciduous forests by automatically comparing the current image with the data of the same month in previous years, thereby improving the accuracy of recognition; by constructing an intelligent fusion framework for multi-source remote sensing images, the accuracy and timeliness of natural resource change detection are significantly improved. The system uses a parallel preprocessing pipeline to effectively eliminate the radiation and geometric differences between multi-source data; the feature fusion module based on deep learning automatically extracts multi-scale change features, overcoming the environmental adaptability defects of the traditional threshold method; the dynamic monitoring module realizes the automatic extraction and classification of sub-meter change patterns, improving monitoring efficiency. At the same time, the system supports the collaborative analysis of multi-platform data such as satellites, aviation, and drones, greatly reducing the impact of cloud cover, and providing full-factor, high-frequency decision support for natural resource supervision. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is a module diagram of the natural resources dynamic monitoring system based on multi-source remote sensing images in this application. DETAILED DESCRIPTION

[0034] The present application will be further described below in detail through specific embodiments in conjunction with the accompanying drawings.

[0035] like Figure 1 As shown, the present application discloses a natural resource dynamic monitoring system based on multi-source remote sensing images, including a multi-source data acquisition module, a preprocessing module, a feature extraction module, a dynamic monitoring module, a visualization output module and a distributed storage module;

[0036] The multi-source data acquisition module is used to receive satellite remote sensing images, aerial remote sensing images, and UAV remote sensing images. The satellite remote sensing images contain multispectral data and synthetic aperture radar data, and the aerial remote sensing images contain high-resolution RGB images and lidar point cloud data. The multi-source data acquisition module constructs a spatiotemporal complementary observation network by synchronously collecting visible light, radar, and lidar data from satellite, aerial, and UAV platforms. Images of natural resources (forestland, cultivated land, or water bodies) are collected through satellite remote sensing images, aerial remote sensing images, and UAV remote sensing images. Multiple image acquisition methods can ensure the accuracy of data acquisition.

[0037] The preprocessing module is connected to the multi-source data acquisition module, and the preprocessing module includes a radiation correction unit, a geometric correction unit and a registration and fusion unit. The radiation correction unit uses a sensor response model to eliminate image radiation distortion, the geometric correction unit performs terrain correction based on a digital elevation model, and the registration and fusion unit realizes spatial alignment of multi-source images through a feature point matching algorithm. The preprocessing module first performs radiation correction (eliminating sensor differences) and geometric correction (eliminating terrain distortion based on DEM) on heterogeneous data, and then realizes sub-pixel registration through SIFT feature point matching. For example, when taking high-altitude photos of the target area, clouds will be captured in the image, and the occlusion of clouds may cause misjudgment, so the clouds in the image need to be erased;

[0038] The feature extraction module includes a temporal feature analysis unit and a change detection unit. The temporal feature analysis unit uses a three-dimensional convolutional neural network to extract the spatiotemporal variation characteristics of resource coverage, and the change detection unit identifies the changed area by constructing a dual-temporal remote sensing image feature difference matrix. The core of the feature extraction module is that the temporal analysis unit uses 3D-CNN to extract the long-term evolution characteristics of resource coverage (such as the slow degradation of forest land), while the change detection unit constructs a dual-temporal feature difference matrix to capture sudden changes (such as illegal land occupation). The land occupation of natural resources such as forest land and water bodies will change slowly over time, for example, expanding or decreasing with the seasons. The feature extraction module can identify these changes. It can also identify sudden situations such as illegal land occupation, mudslides, floods, etc.

[0039] The feature extraction module also includes a same-season comparison submodule, a meteorological interface submodule, and a regional strategy engine. The same-season comparison submodule is used to call historical same-season images in the distributed storage unit, calculate the inter-annual difference value of the vegetation index, and generate a seasonal mask. The meteorological interface submodule is used to access the temperature and humidity data stream of the meteorological bureau in real time. When continuous freeze-thaw cycles are monitored in the target area, the cold zone change detection parameter set is automatically switched. The regional strategy submodule is used to preload the latitude zone rule library, activate the fallen leaf compensation algorithm in the temperate zone, and enable the snow scatter correction model in the cold zone.

[0040] The dynamic monitoring module is connected to the feature extraction module. The dynamic monitoring module includes a change patch generation submodule and a resource classification submodule. The change patch generation submodule aggregates the changed pixels into vector patches through an edge connection algorithm. The resource classification submodule identifies the patches based on the random forest model. The dynamic monitoring engine inputs the pixel-level change results into the edge connection algorithm to generate vector patches. The random forest model classifies the land change events into cultivated land, woodland, etc.

[0041] The visualization output module is used to generate dynamic monitoring maps of natural resources (forest land, water bodies or cultivated land), including information on change locations, change areas and land type conversions, and output change heat maps and statistical reports to help users focus on areas with high frequency changes.

[0042] The distributed storage module uses a NoSQL database to store multi-temporal remote sensing images and change detection results, and accelerates data retrieval through spatial indexing to achieve rapid historical data backtracking.

[0043] During the implementation of this system, the multi-source data acquisition module integrates heterogeneous data from satellite, aviation and UAV platforms to form a spatiotemporal complementary observation network, solving the cloud occlusion and update delay problems of a single data source; the pre-processing module implements radiation-geometry-space triple correction, which can compress the registration error to the sub-pixel level, providing a high-precision data base for change detection; the feature extraction module uses a deep learning model to automatically capture the evolution laws of resources, and combines the dual-phase difference matrix to accurately locate the change area, thereby improving the detection rate of subtle changes (<0.5 acres); the dynamic monitoring engine aggregates change information into sub-meter vector patches and intelligently classifies them, and finally outputs heat maps and statistical reports that can directly support decision-making, thereby replacing the traditional manual interpretation process.

[0044] The preprocessing module also includes a cloud detection unit, which uses a combination of band threshold segmentation and texture analysis to identify and mask cloud coverage areas. The cloud area is initially screened through shortwave infrared band threshold segmentation, and then combined with GLCM texture analysis to identify broken clouds. After generating a cloud mask, the historical data replacement mechanism is automatically triggered to improve the monitoring accuracy of cloudy areas.

[0045] When the above solution is implemented, the cloud detection unit can detect clouds above the target area. Once clouds are identified, the monitoring accuracy is increased by replacing them with historical data, thereby avoiding misjudgment due to image differentiation when clouds block the view.

[0046] The feature extraction module includes a multi-scale fusion unit, which uses a channel attention mechanism to weightedly fuse visible light, near-infrared and radar feature maps. The channel attention mechanism is used to dynamically assign weights to visible light, near-infrared and radar feature maps. For example, near-infrared features are enhanced in vegetation monitoring, and radar scattering features are enhanced in building monitoring, thereby significantly improving the ability to identify mixed land types.

[0047] The dynamic monitoring module also includes a change verification unit, which evaluates the credibility of the change results by superimposing historical monitoring results with manually annotated samples, superimposes the automatically detected change patterns with the historical database, filters out patterns with confidence levels less than a predetermined value, and pushes them for manual review. At the same time, the manually annotated samples are fed back to the deep learning model for online fine-tuning, so that the system's false alarm rate continues to decrease.

[0048] It also includes a task scheduler that dynamically allocates GPU computing resources based on data priority, with change detection tasks enjoying the highest computing priority, thereby ensuring that the delay in disaster event monitoring is shortened.

[0049] The method for dynamic monitoring of natural resources based on multi-source remote sensing images includes the following steps:

[0050] Step S1: synchronously acquiring multi-source images of the target area, the multi-source images including satellite images, aerial images, and drone images, wherein the satellite image time series covers at least three growing seasons;

[0051] Step S2: Perform radiometric normalization on the multi-source images, unify the radiometric benchmarks of different sensors using the histogram matching method, and achieve sub-pixel registration between images through affine transformation to generate the processed image;

[0052] Step S3: Construct a time series image stack and use principal component analysis to extract the first three principal components of the processed image as change-sensitive features. Simultaneously load the historical image database of the same season and calculate the vegetation index difference between the current image and the historical image of the same season to generate seasonal change markers, thereby marking areas that change with the seasons.

[0053] Step S4: inputting the current phase remote sensing image and the reference phase remote sensing image into a pre-trained change detection model to output a pixel-level change probability map;

[0054] Step S5: using a region growing algorithm to aggregate the changed pixels to form an initial change patch, and combining it with morphological filtering to remove isolated patches caused by noise;

[0055] Step S6: Based on the spectral characteristics, texture characteristics and geometric characteristics of the spots, a classifier is used to determine the natural resource type of the changed spots;

[0056] Step S7: Overlay the change map with the natural resource background database to automatically generate a statistical report on changes in cultivated land occupation, forest land reduction, and water body expansion.

[0057] The change detection model in step S4 adopts a U-Net network structure, the encoder part of which includes a ResNet50 backbone network, and the decoder part integrates multi-level feature maps.

[0058] In step S6, the classifier adopts the XGBoost algorithm, and the input features include the NDVI mean, blue band standard deviation and edge density index of the patch. The NDVI mean of the patch is used to distinguish vegetation, the blue band standard deviation is used to identify buildings, and the edge density index is used to delineate the boundaries of water bodies. The above three features are combined to improve the accuracy of cultivated land / grassland classification.

[0059] The technical features of the above embodiments can be combined arbitrarily (as long as there is no contradiction in the combination of these technical features). In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described; these embodiments that are not explicitly written should also be considered to be within the scope of this specification.

Claims

1. A dynamic monitoring system for natural resources based on multi-source remote sensing images, characterized by: It includes multi-source data acquisition module, pre-processing module, feature extraction module, dynamic monitoring module, visual output module and distributed storage module; The multi-source data acquisition module is used to receive satellite remote sensing images, aerial remote sensing images and UAV remote sensing images. The satellite remote sensing images include multispectral data and synthetic aperture radar data, and the aerial remote sensing images include high-resolution RGB images and lidar point cloud data. The preprocessing module is connected to the multi-source data acquisition module, and the preprocessing module includes a radiation correction unit, a geometric correction unit, and a registration and fusion unit. The radiation correction unit uses a sensor response model to eliminate image radiation distortion, the geometric correction unit performs terrain correction based on a digital elevation model, and the registration and fusion unit realizes spatial alignment of multi-source images through a feature point matching algorithm. The feature extraction module includes a temporal feature analysis unit and a change detection unit. The temporal feature analysis unit uses a three-dimensional convolutional neural network to extract the spatiotemporal change characteristics of resource coverage, and the change detection unit identifies the changed area by constructing a dual-temporal remote sensing image feature difference matrix. The feature extraction module also includes a same-season comparison submodule, a meteorological interface submodule, and a regional strategy engine. The same-season comparison submodule is used to call historical same-season images in the distributed storage unit, calculate the inter-annual difference value of the vegetation index, and generate a seasonal mask. The meteorological interface submodule is used to access the temperature and humidity data stream of the meteorological bureau in real time. When continuous freeze-thaw cycles are monitored in the target area, the cold zone change detection parameter set is automatically switched. The regional strategy submodule is used to preload the latitude zone rule library, activate the fallen leaf compensation algorithm in the temperate zone, and enable the snow scatter correction model in the cold zone. The dynamic monitoring module is connected to the feature extraction module. The dynamic monitoring module includes a change patch generation submodule and a resource classification submodule. The change patch generation submodule aggregates the change pixels into vector patches through an edge connection algorithm, and the resource classification submodule identifies the patches based on the random forest model. The visualization output module is used to generate dynamic monitoring maps containing information on change locations, change areas, and land type conversions, and output change heat maps and statistical reports; The distributed storage module uses a NoSQL database to store multi-temporal remote sensing images and change detection results, and accelerates data retrieval through spatial indexing.

2. The natural resource dynamic monitoring system based on multi-source remote sensing images according to claim 1 is characterized in that: The pre-processing module further includes a cloud detection unit, which uses a combination of band threshold segmentation and texture analysis to identify and mask cloud coverage areas.

3. The natural resource dynamic monitoring system based on multi-source remote sensing images according to claim 1 is characterized in that: The feature extraction module includes a multi-scale fusion unit, which weightedly fuses visible light, near infrared and radar feature maps through a channel attention mechanism.

4. The natural resource dynamic monitoring system based on multi-source remote sensing images according to claim 1 is characterized in that: The dynamic monitoring module also includes a change verification unit, which evaluates the credibility of the change results by superimposing historical monitoring results and manually annotated samples.

5. The natural resource dynamic monitoring system based on multi-source remote sensing images according to claim 1 is characterized in that: It also includes a task scheduler that dynamically allocates GPU computing resources based on data priority, with change detection tasks enjoying the highest computing priority.

6. The method for a natural resource dynamic monitoring system based on multi-source remote sensing images according to claim 1, characterized in that: The following steps are involved: Step S1: synchronously acquiring multi-source images of the target area, the multi-source images including satellite images, aerial images, and drone images, wherein the satellite image time series covers at least three growing seasons; Step S2: Perform radiometric normalization on the multi-source images, unify the radiometric benchmarks of different sensors using the histogram matching method, and achieve sub-pixel registration between images through affine transformation to generate the processed image; Step S3: Construct a time series image stack, use principal component analysis to extract the first three principal components in the processed image, and use them as change-sensitive features. Simultaneously load the historical image database of the same season, and generate seasonal change marker areas by calculating the difference in vegetation index between the current image and the historical image of the same season. Step S4: inputting the current phase remote sensing image and the reference phase remote sensing image into a pre-trained change detection model to output a pixel-level change probability map; Step S5: using a region growing algorithm to aggregate the changed pixels to form an initial change patch, and combining it with morphological filtering to remove isolated patches caused by noise; Step S6: Based on the spectral characteristics, texture characteristics and geometric characteristics of the spots, a classifier is used to determine the natural resource type of the changed spots; Step S7: Overlay the change map with the natural resource background database to automatically generate a statistical report on changes in cultivated land occupation, forest land reduction, and water body expansion.

7. The method for dynamic monitoring of natural resources based on multi-source remote sensing images according to claim 6, characterized in that: The change detection model in step S4 adopts a U-Net network structure, the encoder part of which includes a ResNet50 backbone network, and the decoder part integrates multi-level feature maps.

8. The method for dynamic monitoring of natural resources based on multi-source remote sensing images according to claim 6, characterized in that: In step S6, the classifier adopts the XGBoost algorithm, and the input features include the NDVI mean, blue band standard deviation and edge density index of the image spot.

Images