Inland pond classification method and device based on watershed scale and electronic equipment

Abstract

The invention relates to a watershed scale inland pond classification method and device and electronic equipment, and the method comprises the steps: obtaining multi-source geographic space data of a target watershed, including remote sensing images, a digital elevation model, land coverage data, JRC global surface water products, urban boundary data and photovoltaic power station distribution data; carrying out radiation calibration, terrain correction, noise removal and space cutting on the remote sensing image; generating an inland pond water body baseline map based on the multi-source geographic space data; and performing differential combination application according to differential characteristics of the aquaculture pond, the urban pond, the photovoltaic pond, the tailing pond and the farmland pond, and outputting identification and classification results of multiple types of inland ponds. According to the method, the detection rate of small-scale ponds is remarkably improved through multi-source geographic space data fusion, the recognition precision of inland ponds is improved, the whole-process classification precision is remarkably improved compared with a single algorithm, consumed time is remarkably reduced, and high-precision recognition and classification of various ponds in the watershed scale are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing identification and classification of small water bodies, and in particular to a method, device and electronic equipment for classifying inland ponds at a watershed scale. Background Art

[0002] Inland ponds, as the majority of the global pond system (accounting for 64.5% and 38.1% in China), play an important role in maintaining biodiversity, supporting aquaculture, and regulating water resources. However, the expansion of ponds driven by human activities has caused environmental risks such as eutrophication, algal blooms, and heavy metal pollution, and requires refined classification and management. Although remote sensing technology provides an efficient means for water body identification, significant technical bottlenecks still exist in the classification of inland ponds: Insufficient recognition accuracy: Traditional water body index methods and pixel-level supervised classification cannot effectively distinguish small-sized, variably shaped inland ponds from natural water bodies, and are easily affected by clouds, shadows, and turbid water. The similarity of pond spectral characteristics leads to missed detections and misclassifications, and there is a lack of robust recognition methods for small-scale water bodies.

[0003] Lack of a classification system: Existing research often focuses on the independent identification of a single type of pond and lacks a unified classification framework covering multiple categories. Global water body datasets only identify the existence of water bodies and are not subdivided by functional type, making them unable to support differentiated ecological governance.

[0004] Limitations of algorithm adaptability: A single machine learning algorithm is difficult to take into account the differentiated characteristics of multiple types of ponds; the lack of a systematic classification framework that integrates multi-source data and multi-algorithm collaboration restricts the accuracy and efficiency of basin-scale applications. Summary of the Invention

[0005] Based on this, it is necessary to provide a basin-scale inland pond classification method, device and electronic equipment to address the above-mentioned problems of insufficient recognition technology, lack of a multi-category classification system and limited algorithm adaptability.

[0006] The present invention provides a method for classifying inland ponds at a watershed scale, the method comprising: Obtain multi-source geospatial data for the target watershed, including remote sensing images, digital elevation models, land cover data, JRC global surface water products, city boundary data, photovoltaic power station distribution data, mining area polygon data, lake datasets, and statistical yearbook data; Perform radiometric calibration, terrain correction, noise removal and spatial cropping on remote sensing images; Generate baseline maps of inland pond water bodies based on multi-source geospatial data; Based on the differentiated characteristics of aquaculture ponds, urban ponds, photovoltaic ponds, tailings ponds, and farmland ponds, differentiated combination applications are carried out to output identification and classification results of multiple categories of inland ponds. The differentiated combination applications include: using object-oriented analysis combined with geometric feature decision trees and neighborhood space discriminant methods to identify aquaculture ponds; using land cover overlay analysis and buffer ratio verification to identify urban ponds; using random forest algorithm to train power station samples to identify photovoltaic ponds; constructing a convolutional neural network model combined with spectral texture features to identify tailings ponds; and using digital elevation models and neighborhood analysis of cultivated land data to identify farmland ponds.

[0007] In one embodiment, performing radiometric calibration, terrain correction, noise removal, and spatial cropping on the remote sensing image includes: Calculate the median backscatter value of Sentinel-1 SAR cross-year series data to eliminate salt and pepper noise; Sentinel-2 images were radiometrically calibrated and terrain corrected, and then clipped according to the boundaries of the study area.

[0008] In one embodiment, generating a baseline map of inland pond water bodies based on multi-source geospatial data includes: The dual-polarization water index was calculated based on the JRC global surface water product and Sentinel-1 SAR data, and the initial water mask was generated by determining the threshold using the water frequency method. Sentinel-2 data were fused to optimize the mask boundaries and form a basin-scale baseline map of inland ponds.

[0009] In one embodiment, the method of identifying aquaculture ponds by using object-oriented analysis combined with a geometric feature decision tree and a neighborhood space discrimination method includes: Use global city boundary data to demarcate non-urban areas, and overlay a baseline map to extract the rough area of ​​aquaculture ponds; Based on Sentinel-2 images and field sampling data, a geometric feature decision tree was constructed to screen aquaculture ponds.

[0010] In one embodiment, the method of constructing a geometric feature decision tree to screen aquaculture ponds includes: The neighborhood expansion method was used to generate buffer zones for potential ponds, and isolated non-aquaculture ponds were identified based on the intersection of the buffer zones. Fish ponds and shrimp and crab ponds were distinguished by the normalized difference pond index and normalized difference vegetation index.

[0011] In one embodiment, identifying urban ponds using land cover overlay analysis and buffer ratio verification includes: The global city boundary data and SinoLC-1 city class layer were overlaid onto the baseline map to screen the candidate set of urban ponds; Independence was verified using a neighborhood-based expansion method and secondarily confirmed by the proportion of impervious surface in the buffer zone.

[0012] In one embodiment, the method of using a random forest algorithm to train power plant samples to identify photovoltaic ponds includes: Collect sample points of photovoltaic power stations in the target area; Set random forest model parameters, train and test the model; The model output was overlaid with the baseline map to determine the PV pond distribution.

[0013] In one embodiment, the method of constructing a convolutional neural network model combining spectral texture features to identify tailings ponds includes: Delineate target tailings pond areas based on DEM, SinoLC-1 bare map layers, and global mining area datasets; A CNN model was constructed, spectral and texture features were input, and tailings ponds were identified through training using convolutional layers, maximum pooling layers, and fully connected layers.

[0014] In one embodiment, the identifying of farmland ponds by using a digital elevation model and neighborhood analysis of farmland data includes: Overlay the DEM, SinoLC-1 cultivated land layer, and non-urban boundary data onto the baseline map; Neighborhood analysis was applied to verify whether ponds were independently distributed in the farmland.

[0015] In one embodiment, the method further includes: Output the spatial distribution map of multi-category inland ponds and construct confusion matrices for fish ponds, shrimp and crab ponds, urban ponds, photovoltaic ponds, tailings ponds, and farmland ponds; The Nash coefficient (NS), correlation coefficient (R²), root mean square error (RMSE) and mean absolute error (MAE) are used to evaluate the accuracy of the framework. The present invention also provides a basin-scale inland pond classification device, comprising: The multi-source data acquisition module is used to obtain multi-source geospatial data of the target watershed, including remote sensing images, digital elevation models, land cover data, JRC global surface water products, urban boundary data, photovoltaic power station distribution data, mining area polygon data, lake datasets and statistical yearbook data.

[0016] The data preprocessing module is used to perform radiometric calibration, terrain correction, noise removal and spatial cropping on remote sensing images.

[0017] The water body baseline map generation module is used to generate the inland pond water body baseline map based on multi-source geospatial data.

[0018] The differentiated classification module is used to perform differentiated combination applications based on the differentiated characteristics of aquaculture ponds, urban ponds, photovoltaic ponds, tailings ponds, and farmland ponds, and output the identification and classification results of multiple categories of inland ponds. The differentiated combination applications include: using object-oriented analysis combined with geometric feature decision trees and neighborhood space discriminant methods to identify aquaculture ponds; using land cover overlay analysis and buffer ratio verification to identify urban ponds; using random forest algorithms to train power station samples to identify photovoltaic ponds; constructing a convolutional neural network model that combines spectral texture features to identify tailings ponds; and using digital elevation models and neighborhood analysis of cultivated land data to identify farmland ponds.

[0019] The present invention also provides an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method for classifying inland ponds at a watershed scale as described above is implemented.

[0020] The above-mentioned basin-scale inland pond classification method, device and electronic equipment significantly improve the detection rate of small-scale ponds through the fusion of multi-source geographic spatial data, and eliminate the interference of clouds and water turbidity; achieve accurate classification through a combination of differentiated algorithms, and use a geometric feature decision tree combined with a neighborhood spatial discrimination method for aquaculture ponds with high accuracy; apply land cover overlay and impervious surface ratio verification to urban ponds to avoid misjudgment; use a lightweight random forest model to capture the coexistence spectral characteristics of water surface and photovoltaic panels for photovoltaic ponds, with high accuracy and short time consumption; construct a CNN to fuse spectral and texture features for tailings ponds, and the recognition rate is improved compared with traditional methods; for farmland ponds, unsupervised and efficient identification is achieved based on neighborhood analysis of digital elevation models and cultivated land data, and the efficiency is improved compared with supervised methods. Finally, the classification accuracy of the entire process is significantly improved compared with a single algorithm, and the time consumption is significantly reduced, achieving high-precision identification and classification of multiple types of ponds at the basin scale. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0022] Figure 1 A flow chart of a method for classifying inland ponds at a watershed scale according to one embodiment; Figure 2 A schematic diagram of a method for classifying inland ponds at a watershed scale according to one embodiment; Figure 3 A flow chart of a method for classifying inland ponds at a watershed scale according to another embodiment; Figure 4Schematic diagram of the classification decision tree extracted for aquaculture ponds; Figure 5 A flow chart of a method for classifying inland ponds at a watershed scale according to another embodiment; Figure 6 A flow chart of a method for classifying inland ponds at a watershed scale according to another embodiment; Figure 7 Schematic diagram of the random forest extracted for the photovoltaic pond; Figure 8 A flow chart of a method for classifying inland ponds at a watershed scale according to another embodiment; Figure 9 Schematic diagram of the convolutional neural network extracted for the tailings pond; Figure 10 A flow chart of a method for classifying inland ponds at a watershed scale according to another embodiment; Figure 11 A schematic diagram of a watershed-scale inland pond classification device according to one embodiment; Figure 12 FIG. 1 is a diagram showing the internal structure of a computer device according to an embodiment. DETAILED DESCRIPTION

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0024] Inland ponds, as the majority of the global pond system (accounting for 64.5% of the total, with China accounting for 38.1%), play an important role in maintaining biodiversity, supporting aquaculture, and regulating water resources. However, the expansion of ponds driven by human activities (such as food production and energy development) has triggered environmental risks such as eutrophication, algal blooms, and heavy metal pollution, necessitating refined classification and management. While remote sensing technology provides an efficient means for identifying water bodies, significant technical bottlenecks remain in the classification of inland ponds: Insufficient recognition accuracy: Traditional water body index methods (such as NDWI) and pixel-level supervised classification (MLC, SVM) have difficulty in effectively distinguishing small-area, multi-shaped inland ponds from natural water bodies, and are easily affected by clouds, shadows, and turbid water bodies. The similarity of pond spectral characteristics (such as fish ponds and shrimp and crab ponds, photovoltaic ponds and ordinary ponds) leads to missed detections and misclassifications, and there is a lack of robust recognition methods for small-scale water bodies.

[0025] Lack of a classification system: Existing research often focuses on the independent identification of a single type of pond (such as aquaculture ponds and tailings ponds), and lacks a unified classification framework covering multiple categories (aquaculture / urban / photovoltaic / tailings / farmland ponds); global water body datasets (such as JRC-GSW) only identify the existence of water bodies and are not subdivided by functional type, which cannot support differentiated ecological governance.

[0026] Limitations of algorithm adaptability: A single machine learning algorithm (such as random forest and CNN) is difficult to take into account the differentiated characteristics of multiple types of ponds (such as geometric regularity, spatial location, and texture complexity); the lack of a systematic classification framework that integrates multi-source data (optical / radar remote sensing, topography, and land use) and multi-algorithm collaboration (spatial analysis + decision tree + deep learning) restricts the accuracy and efficiency of basin-scale applications.

[0027] Therefore, developing an inland pond classification method that integrates multi-source remote sensing data, couples spatial analysis and machine learning algorithms, and adapts to the watershed scale is of great significance to improving the accuracy of small and micro-water system mapping and supporting ecological protection decision-making.

[0028] The following combination Figures 1-12 The present invention describes a method, apparatus, and electronic device for classifying inland ponds at a watershed scale.

[0029] like Figure 1 and Figure 2 As shown, in one embodiment, a method for classifying inland ponds at a watershed scale comprises the following steps: Step S110, multi-source data collaborative processing: obtain multi-source geospatial data of the target watershed, including remote sensing images, digital elevation models, land cover data, JRC global surface water products, city boundary data, photovoltaic power station distribution data, mining area polygon data, lake datasets and statistical yearbook data.

[0030] The specific requirements for remote sensing images (including Sentinel-1 SAR data and Sentinel-2 optical images), digital elevation models (DEMs), land cover data (SinoLC-1), JRC global surface water products, urban boundary data, photovoltaic power station distribution data, mining area polygon data, lake datasets, and statistical yearbook data are as follows: Sentinel-1 long-term series of aperture radar data spanning multiple years in the target study area, namely Sentinel-1 Synthetic Aperture Radar (SAR), Sentinel-2 satellite remote sensing image data of the corresponding years with a spatial resolution of 30 m, SRTMGL1 global 1 arc-second DEM data, and China 1 The national-scale land cover map with m resolution, namely land cover data (SinolC-1), the China photovoltaic power station polygon geospatial dataset is the spatial distribution information data of China's photovoltaic power stations from 2010 to 2022, the global-scale mining area polygon (version 2) includes mining area polygon data from 2000 to 2019, vector boundary data of the target study area and global city boundary vector data, the Chinese lake dataset within the corresponding study area, and statistical data from the "China Fisheries Statistical Yearbook" and "China Statistical Yearbook" of the corresponding years.

[0031] Step S120 , performing radiometric calibration, terrain correction, noise removal, and spatial cropping on the remote sensing image.

[0032] Remote sensing imagery noise removal utilizes a time series median composite method. This involves calculating the median backscatter value for Sentinel-1 SAR data across multiple years to eliminate salt-and-pepper noise caused by sensor, atmospheric particles, and the imaging environment. Because inland ponds maintain relatively continuous water storage over relatively long periods, backscatter values ​​remain stable within a specific range year-round. Compared to seasonal water bodies, median composites further enhance low-backscatter signatures. By leveraging the stability of backscatter values ​​throughout the year, median composites eliminate transient outliers (such as bird flocks and vessel interference), thereby enhancing the low-scatter signature of water bodies. Sentinel-2 images are then radiometrically and topographically corrected. They are then cropped to the study area boundary. Radiometric calibration corrects for sensor errors, while topographic correction eliminates shadow distortion. This ensures that subsequent classification is unaffected by imaging conditions and reduces errors in extracting small pond boundaries.

[0033] By fusing the spectral details of optical imagery (Sentinel-2) with the cloud-rain interference resistance of radar data (Sentinel-1), combined with DEM terrain constraints, the detection capability of small-scale ponds is significantly improved, overcoming the defects of traditional single optical data vulnerable to environmental interference; through iterative optimization of JRC global land surface water products and localized remote sensing data, seasonal and temporary water body interference is eliminated to ensure the spatial reference accuracy of inland ponds. Radiometric calibration and terrain correction reduce atmospheric distortion of Sentinel-2 to ensure that the input data quality meets the needs of watershed-scale analysis.

[0034] Step S130, dynamic construction of classification framework: generate inland pond water baseline map based on multi-source geospatial data.

[0035] As an option, based on JRC global land surface water products and Sentinel-1 SAR data, calculate the dual-polarization water index, determine the threshold to generate the initial water mask through the water body frequency method, and through the fusion of Sentinel-1 dual-polarization data (VV+VH), enhance the water body and vegetation / bare soil discrimination (20% higher water body recognition rate than traditional NDWI index), use the water body frequency method to exclude seasonal water bodies (such as rainwater) through time series statistics, ensure that the baseline map only retains stable inland ponds. Subsequently, fuse Sentinel-2 data to optimize the mask boundary to form a watershed-scale inland pond baseline map. For Sentinel-2 optimization, supplement the 10m resolution visible light band to refine the pond boundary (remove river / lake misjudgment), and provide a pure spatial mask for classification.

[0036] Step S140, for the differentiated characteristics of aquaculture ponds, urban ponds, photovoltaic ponds, tailings ponds, and farmland ponds, perform differentiated combination application to output multi-class inland pond recognition and classification results, including: using object-oriented analysis combined with geometric feature decision tree and neighborhood spatial discrimination method to identify aquaculture ponds; using land cover overlay analysis and buffer ratio verification to identify urban ponds; using random forest algorithm to train power station samples to identify photovoltaic ponds; constructing a convolutional neural network model combining spectral texture features to identify tailings ponds; applying digital elevation model and neighborhood analysis of farmland data to identify farmland ponds.

[0037] Based on literature and previous research, the main classification categories and descriptions of inland ponds at the watershed scale are summarized. Pond categories include: aquaculture ponds (fish ponds, shrimp and crab ponds), urban ponds, photovoltaic ponds, tailings ponds, farmland ponds, and other ponds. Specifically, aquaculture ponds: fish ponds are used for fish farming and have regular shapes, are usually concentrated, have regular arrangements, and have uniform color distribution. Shrimp and crab ponds are primarily used for shrimp and crab farming and have regular shapes, are usually concentrated, have regular arrangements, and cultivate plants within the ponds. Urban ponds: ponds are located within urban boundaries, are usually isolated, and have landscaping functions. Photovoltaic ponds: ponds with photovoltaic modules located above the water surface. Tailings ponds: ponds in mining areas that store mine waste, located at high altitudes, have high heavy metal content, and have unique spectral characteristics. Farmland ponds: ponds located around farmland and scattered. Other ponds: primarily natural, abandoned, or unused ponds with minimal human interference, except for those in the above-specified categories.

[0038] Construct a framework for identifying and classifying inland ponds at the watershed scale. Use multi-source data and multi-machine learning algorithms to identify and classify inland ponds. For example, use spatial analysis methods in geography (overlay analysis, buffer analysis, neighborhood analysis) combined with decision tree, random forest algorithm (RF), convolutional neural network (CNN) machine learning algorithms, and obtain the best identification and classification results according to the characteristics of each type of pond, and perform accuracy analysis. As an option, for aquaculture ponds, urban ponds, photovoltaic ponds, tailings ponds, and farmland ponds, according to the shape, location, and spectral texture features of each type of pond, dynamically select and combine the following methods: spatial neighborhood analysis and geometric feature decision tree, land cover overlay and buffer ratio verification, machine learning classification model training, and deep learning feature extraction. By integrating parallel classification frameworks for aquaculture ponds (fish ponds / shrimp and crab ponds), urban ponds, photovoltaic ponds, tailings ponds, and farmland ponds, we can achieve collaborative identification of ponds. For regular geometric characteristics of aquaculture ponds, construct a geometric feature decision tree (such as area, shape index) and a spatial neighborhood discrimination method. The geometric feature decision tree locks regular shapes and excludes irregular boundaries of natural water bodies. The neighborhood spatial discrimination method (buffer inflation + intersection test) filters isolated water bodies. Combine NDPI (phytoplankton enrichment) and NDVI (underwater vegetation) to distinguish fish ponds (NDPI > 0.2) and shrimp and crab ponds (NDVI > 0.3), and improve classification accuracy. For urban ponds, use land cover overlay analysis and buffer ratio verification. Overlay urban boundary data + SinoLC-1 to quickly locate urban water bodies. Verify the accuracy of the buffer impermeable surface ratio to improve accuracy. For photovoltaic ponds, use a random forest model to train photovoltaic station samples to learn the spectral characteristics of photovoltaic panels and water surfaces coexisting. Reduce time consumption at high accuracy. For tailings ponds, use a convolutional neural network model to input DEM elevation (tailings ponds are located in high and steep slope areas), Sentinel-2 spectrum (heavy metal pollution causes short-wave infrared drop), and texture features (mud patchy texture). Automatically extract pollution markers through convolutional layers (32 / 16 filter groups) to improve recognition rate from 30% to 64%. For farmland ponds, overlay the SinoLC-1 cultivated land layer with the non-urban boundary, and combine 50m buffer analysis to ensure that the pond is completely surrounded by farmland, achieving fast identification without samples, with an efficiency improvement of 70% compared to supervised classification.

[0039] The basin-scale inland pond classification method of this embodiment significantly improves the detection rate of small-scale ponds by fusing multi-source geospatial data, eliminating interference from clouds and water turbidity; it achieves accurate classification through a combination of differentiated algorithms. For aquaculture ponds, a geometric feature decision tree combined with a neighborhood spatial discriminant method is used, supplemented by NDPI (plankton enrichment) and NDVI (underwater vegetation) to distinguish fish ponds from shrimp and crab ponds, with high accuracy; land cover overlay and impervious surface ratio verification are applied to urban ponds to avoid misjudgment; a lightweight random forest model is used to capture the spectral characteristics of the coexistence of water surface and photovoltaic panels in photovoltaic ponds, with high accuracy and short time consumption; a CNN is constructed for tailings ponds to fuse spectral and texture features, and the recognition rate is improved compared with traditional methods; for farmland ponds, unsupervised and efficient identification is achieved based on neighborhood analysis of digital elevation models and cultivated land data, with improved efficiency compared with supervised methods. Ultimately, the classification accuracy of the entire process is significantly improved compared with a single algorithm, and the time consumption is significantly reduced, achieving high-precision identification and classification of multiple types of ponds at the basin scale.

[0040] like Figure 3 As shown, in one embodiment, the method of identifying aquaculture ponds by using object-oriented analysis combined with a geometric feature decision tree and a neighborhood space discrimination method includes the following steps: Step S310: Use global city boundary data to divide non-urban areas, and overlay the baseline map to extract the rough range of aquaculture ponds.

[0041] The non-urban boundaries of the target area were demarcated using global city boundary data, and the rough range of aquaculture ponds was extracted by combining the inland pond baseline map at the watershed scale.

[0042] Step S320: Based on Sentinel-2 images and field sampling data, a geometric feature decision tree is constructed to screen aquaculture ponds.

[0043] Using Sentinel-2 satellite remote sensing imagery combined with statistical data from field aquaculture pond sampling, we selected multiple geometric features as sensitive parameters for aquaculture ponds and determined thresholds for these parameters. A decision tree was constructed using these geometric features and parameters to screen aquaculture ponds. Ponds were screened based on regular shape, an area threshold (>0.1 hectares), and agglomeration (neighborhood pond density >5 ponds / km²), addressing misclassification caused by irregular shapes of natural water bodies.

[0044] Alternatively, a geometric feature decision tree was constructed to screen for aquaculture ponds. This included using a neighborhood expansion method to generate buffer zones for potential ponds and identifying isolated non-aquaculture ponds based on the intersection of these buffer zones. The Normalized Difference Pool Index (NDPI) and the Normalized Difference Vegetation Index (NDVI) were used to distinguish fish ponds from shrimp and crab ponds. Using the neighborhood expansion method, isolated non-aquaculture ponds were identified based on the intersection of these buffer zones, achieving a 93% exclusion rate. NDPI enhanced the extraction of plankton-rich water bodies (fish ponds), while NDVI detected underwater vegetation (a core characteristic of shrimp and crab ponds). This achieved an 87.67% accuracy in distinguishing fish from shrimp and crab ponds (a 35% improvement over single-spectrum classification).

[0045] In the aquaculture pond classification, see Figure 4 Object-oriented classification decision trees have the potential to address the need for accurate identification of aquaculture ponds and can effectively support classic mapping techniques. Global city boundary data was used to delineate the non-urban boundaries of the target area. Combined with a basin-scale baseline map of inland ponds, a rough outline of aquaculture ponds was extracted. Sentinel-2 satellite remote sensing imagery, combined with field data from aquaculture pond sampling, was used to select multiple geometric features as sensitive parameters for aquaculture ponds and determine thresholds for these parameters. A decision tree for aquaculture pond selection was constructed using these geometric features and parameters to screen aquaculture ponds. Because aquaculture ponds are often concentrated in low-altitude and flat rural areas, a neighborhood-based expansion method and spatial discriminant methods were used. All potential aquaculture ponds were expanded with a buffer of a certain size and then superimposed based on whether they intersected with other buffers. If there was no intersection, the pond was identified as an isolated non-aquaculture pond. To further distinguish fish ponds from shrimp and crab ponds, the Normalized Difference Pond Index (NDPI) and the Normalized Difference Vegetation Index (NDVI) were used. The NDPI reflects the status of surface plankton and flora, enhancing the identification of aquaculture ponds with low plankton mobility and abundant plankton. More importantly, shrimp and crab ponds usually have a large amount of underwater vegetation (distinguished by NDVI) to facilitate molting and avoid enemies, which is obviously different from fish ponds.

[0046] like Figure 5 As shown, in one embodiment, land cover overlay analysis and buffer ratio verification are used to identify urban ponds, including the following steps: In step S510, the global city boundary data and the SinoLC-1 city layer are overlaid onto the baseline map to screen the candidate set of urban ponds. By overlaying the city boundaries, non-urban water bodies are directly excluded, and the candidate set of urban ponds is narrowed down.

[0047] In step S520, the independence is verified using a neighborhood-based expansion method and reconfirmed using the impervious surface ratio of the buffer zone.

[0048] The impervious surface ratio is used for verification, and the proportion of surrounding impervious surfaces is calculated through buffer zone analysis (>60% is determined to be urban landscape water bodies), with the accuracy increased to 75.75%.

[0049] In the urban pond classification, urban ponds are important resources for environmental aesthetics, maintaining ecology, protecting environmental quality, and safeguarding public health. A three-step classification method is used to identify urban ponds. In the first classification step, the city boundary and SinoLC-1 are superimposed on the basin-scale inland pond baseline map to screen urban ponds. Secondly, a neighborhood-based expansion method is used to ensure their isolation. The final step is to expand these potential ponds by a buffer of a certain size, and then calculate the proportion of different land cover categories within the intersection range. If there is a large proportion of impervious surface area or grassland in the overlapping area, it is judged as an urban pond.

[0050] like Figure 6 As shown, in one embodiment, a random forest algorithm is used to train power station samples to identify photovoltaic ponds, including the following steps: Step S610: Collect PV power station sample points in the target area. 1,000 PV power station sample points are used to cover different regions and scales to address sample bias.

[0051] Step S620: Set the random forest model parameters, train, and test the model. A lightweight random forest model (number of trees = 20) was used. While maintaining 87% accuracy, the training efficiency was five times higher than that of CNN, meeting the requirements for rapid watershed-scale mapping.

[0052] Step S630: Overlay the model output with the baseline map to determine the photovoltaic pond distribution.

[0053] The PV power station sample points in the target study area were collected, the parameters of the random forest model were set, and training and testing were performed to obtain the final results. The accurate PV pond distribution results were determined by combining the inland pond baseline map at the watershed scale.

[0054] In the photovoltaic pond classification, see Figure 7 The innovative fishery-photovoltaic model involves installing photovoltaic modules on the water surface while conducting aquaculture activities, harnessing solar power. The random forest algorithm (RF) is an ensemble classifier that uses a set of decision trees to predict classification or regression, offering advantages such as high accuracy, high efficiency, and good stability. RF is sensitive to sampling design, and appropriate training samples are crucial for the classification accuracy and stable performance of the RF model. Samples were labeled as photovoltaic power plants, and 1,000 sample points were collected. The number of trees was set to 20, and the remaining parameters remained at their default values. The constructed RF model was trained and tested to obtain the final results.

[0055] like Figure 8As shown, in one embodiment, a convolutional neural network model combining spectral texture features is constructed to identify tailings ponds, including the following steps: Step S810: Delineate the target area of ​​the tailings pond based on the DEM, the SinoLC-1 bare map layer, and the global mining area dataset.

[0056] Step S820: Build a CNN model, input spectral and texture features, and train the model to identify tailings ponds through convolutional layers (32 / 16 filters), maximum pooling layers, and fully connected layers.

[0057] The spectrum and texture are input together, and the spectral anomalies caused by heavy metal pollution (sudden drop in shortwave infrared band reflectance) and the unique texture of tailings slurry (CNN automatically extracts mottled features) are used to resolve the confusion between tailings ponds and ordinary bare land ponds. A multi-layer convolution structure is adopted to capture multi-scale features through a 32 / 16 filter group, and the maximum pooling layer is used to enhance the spatial pattern recognition of contaminated areas, thereby increasing the detection rate of tailings ponds from 30% of traditional methods to 64%.

[0058] In the tailings pond classification, based on the tailings pond target area obtained from DEM, SinoLC-1 (bare land) and global mining datasets, a CNN model combining spectral and texture features is established to further identify tailings ponds. Figure 9 , the CNN architecture consists of two convolutional layers, a maximum pooling layer, and three fully connected layers. Using a segmented image with a pixel size of 350×150, inputs of different sizes are fed into two convolutional layers with filter sizes of 32 and 16 and a kernel size of 3×3. The maximum pooling layer extracts features by summing up the maximum values ​​of the convolution filter inputs. After the convolutional layers extract features, the pooling layers downsample, and then map the features to the CNN output through a flattening layer and a subset of fully connected layers. After the last three fully connected layers are applied, the output of the network is obtained by scaling the features to a 1×1 size of the tailings pool sample.

[0059] like Figure 10 As shown, in one embodiment, the identification of farmland ponds by applying a digital elevation model and neighborhood analysis of cultivated land data includes the following steps: Step S1010: Overlay the DEM, SinoLC-1 cultivated land layer, and non-urban boundary data onto the baseline map. By overlaying the SinoLC-1 cultivated land layer and non-urban boundary data, it is ensured that the pond is completely within the farmland.

[0060] Step S1020: Apply neighborhood analysis to verify whether ponds are independently distributed within the farmland. Using the neighborhood expansion method, we verify the independence of ponds (distance to the nearest water body > 50m), excluding linear water bodies such as irrigation canals, and achieve rapid identification without sample annotation (70% efficiency improvement over machine learning). In the farmland pond classification, farm ponds are generally widespread, scattered around low-lying, flat farmland, and primarily used for irrigation and drainage. Therefore, a basin-scale baseline map of inland ponds was overlaid and analyzed using DEM, SinoLC-1 (cultivated land), and non-urban boundary data. The aforementioned domain analysis was applied to examine whether ponds were isolated within farmland, thereby identifying farmland ponds.

[0061] All ponds outside the categories of aquaculture ponds, urban ponds, photovoltaic ponds, tailings ponds, and farmland ponds are considered "other ponds". They may be unused or abandoned ponds, or ponds with multiple functions (whose characteristics are not obvious or easy to confuse), and therefore need to be listed separately.

[0062] After classifying multiple inland pond categories using a differentiated combination of methods, a spatial distribution map of these multiple categories of inland ponds was generated. Confusion matrices were also constructed for fish ponds, shrimp and crab ponds, urban ponds, photovoltaic ponds, tailings ponds, and farmland ponds. The confusion matrices quantified the accuracy of the six pond categories (e.g., overall accuracy for fish ponds was 91.50% and the Kappa coefficient was 79.27%), and major misclassifications (e.g., tailings ponds misidentified as bare land) were identified.

[0063] The Nash coefficient (NS), correlation coefficient (R²), root mean square error (RMSE), and mean absolute error (MAE) were used to evaluate the accuracy of the framework. Hydrological indicators were cross-validated, with the Nash coefficient and correlation coefficient used to verify the rationality of the spatial distribution. The RMSE was used to constrain the area estimation error, ensuring that the results can directly support the quantitative management of water resources in the basin.

[0064] Specifically, confusion matrices for fish ponds, shrimp and crab ponds, urban ponds, photovoltaic ponds, tailings ponds, and farmland ponds were established, and the accuracy of the inland pond identification and classification framework at the watershed scale was analyzed. The indicators included: Nash coefficient (NS), correlation coefficient (R 2 ), root mean square error (RMSE), and mean absolute error (MAE).

[0065] Among them, the overall accuracy of fish ponds, shrimp and crab ponds, urban ponds, photovoltaic ponds, tailings ponds, and farmland ponds were 91.50%, 87.67%, 75.75%, 87.00%, 64.00%, and 62.00%, respectively; the Kappa coefficients (%) were 79.27, 75.34, 48.27, 41.23, 27.42, and 10.02, respectively. The Nash coefficient (NS) and correlation coefficient (R) of the inland pond identification and classification framework at the watershed scale were 2 ), root mean square error (RMSE), and mean absolute error (MAE) are 0.96, 0.96, 9.91, and 8.83, respectively. The development, validation, and application of the proposed framework demonstrate that the estimation of inland ponds at the watershed scale is accurate and up-to-date.

[0066] This basin-scale inland pond classification method utilizes multi-source data collaboration. By fusing Sentinel-1SAR (which penetrates clouds and is unaffected by sunlight) with Sentinel-2 optical data (high spectral resolution), combined with DEM terrain constraints, it significantly improves the detection capability of small-scale ponds and overcomes the defect that traditional single optical data is susceptible to environmental interference. A refined processing flow is adopted to eliminate transient noise through time series median synthesis. JRC water body products are superimposed with localized Sentinel data to optimize the baseline map, ensuring that the water body mask accuracy meets the requirements of basin-scale applications, thus breaking through the bottleneck of precision identification.

[0067] For the first time, a parallel classification framework for aquaculture ponds (fish ponds / shrimp and crab ponds), urban ponds, photovoltaic ponds, tailings ponds, and farmland ponds was integrated, filling the gap in existing technologies that only focus on a single type. Based on the regular geometric characteristics of aquaculture ponds, a geometric feature decision tree (such as area and shape index) and a spatial neighborhood discrimination method were constructed, and NDPI / NDVI were used to distinguish fish ponds from shrimp and crab ponds (with an accuracy of 87.67%). Based on the location characteristics of urban ponds, the spectral confusion problem was solved by superimposing land cover data and verifying the proportion of impervious surfaces in the buffer zone. For tailings ponds with complex spectral textures, CNN was used to automatically extract deep features (such as special reflection patterns caused by heavy metal pollution) to improve the recognition rate of hard-to-distinguish targets, thus establishing a full-category classification system.

[0068] Differentiated model deployment is adopted, and the optimal algorithm combination is selected according to the characteristics of the pond type. For example, random forest (RF) is adapted to the sample-driven characteristics of photovoltaic ponds, and CNN is adapted to the texture complexity of tailings ponds, to avoid the insufficient generalization ability of a single algorithm. Decision-making is enhanced through spatial analysis, and geospatial methods such as neighborhood analysis and buffer zone validation are introduced to strengthen the use of contextual information (such as the independent distribution determination of farmland ponds) and solve the problem of spatial misjudgment in pixel-level classification. A full-process verification mechanism is adopted, and the framework robustness is ensured through dual verification of confusion matrix (various types of accuracy 62%~91%) and hydrological model indicators (NS=0.96, R²=0.96), thereby achieving multi-algorithm collaborative optimization performance.

[0069] Through this invention, high-precision automated mapping of five major types of inland ponds (aquaculture / urban / photovoltaic / tailings / farmland) is achieved at the watershed scale, with an overall average accuracy of 81.18% (an improvement of 32% compared to a single algorithm), and classification time is reduced by 50%, providing standardized technical support for the management of small and micro water resources.

[0070] The following describes the watershed-scale inland pond classification device provided by the present invention. The watershed-scale inland pond classification device described below and the watershed-scale inland pond classification method described above can be referenced to each other.

[0071] likeFigure 11 As shown, in one embodiment, a basin-scale inland pond classification device includes a multi-source data acquisition module 1110 , a data preprocessing module 1120 , a water body baseline map generation module 1130 and a differential classification module 1140 .

[0072] The multi-source data acquisition module 1110 is used to obtain multi-source geospatial data of the target watershed, including remote sensing images, digital elevation models, land cover data, JRC global surface water products, urban boundary data, photovoltaic power station distribution data, mining area polygon data, lake data sets and statistical yearbook data.

[0073] The data preprocessing module 1120 is used to perform radiometric calibration, terrain correction, noise removal, and spatial cropping on remote sensing images.

[0074] The water body baseline map generating module 1130 is used to generate an inland pond water body baseline map based on multi-source geospatial data.

[0075] The differentiated classification module 1140 is used to perform differentiated combination applications based on the differentiated characteristics of aquaculture ponds, urban ponds, photovoltaic ponds, tailings ponds, and farmland ponds, and output identification and classification results of multiple categories of inland ponds. The differentiated combination applications include: using object-oriented analysis combined with geometric feature decision trees and neighborhood space discriminant methods to identify aquaculture ponds; using land cover overlay analysis and buffer ratio verification to identify urban ponds; using random forest algorithms to train power station samples to identify photovoltaic ponds; constructing a convolutional neural network model that combines spectral texture features to identify tailings ponds; and using digital elevation models and neighborhood analysis of cultivated land data to identify farmland ponds.

[0076] Figure 12 The following is a schematic diagram of the physical structure of an electronic device. The electronic device may be a smart terminal, and its internal structure diagram may be as follows: Figure 12 As shown. The electronic device includes a processor, a memory, and a network interface connected via a system bus. The processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the electronic device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a method for classifying inland ponds at a watershed scale is implemented, the method comprising: Obtain multi-source geospatial data for the target watershed, including remote sensing images, digital elevation models, land cover data, JRC global surface water products, city boundary data, photovoltaic power station distribution data, mining area polygon data, lake datasets, and statistical yearbook data; Perform radiometric calibration, terrain correction, noise removal and spatial cropping on remote sensing images; Generate baseline maps of inland pond water bodies based on multi-source geospatial data; Based on the differentiated characteristics of aquaculture ponds, urban ponds, photovoltaic ponds, tailings ponds, and farmland ponds, differentiated combination applications are carried out to output identification and classification results of multiple categories of inland ponds. The differentiated combination applications include: using object-oriented analysis combined with geometric feature decision trees and neighborhood space discriminant methods to identify aquaculture ponds; using land cover overlay analysis and buffer ratio verification to identify urban ponds; using random forest algorithm to train power station samples to identify photovoltaic ponds; constructing a convolutional neural network model combined with spectral texture features to identify tailings ponds; and using digital elevation models and neighborhood analysis of cultivated land data to identify farmland ponds.

[0077] Those skilled in the art will understand that the structure shown in the classification of inland ponds at the watershed scale in the figure is merely a block diagram of a portion of the structure related to the scheme of the present invention, and does not constitute a limitation on the electronic device to which the scheme of the present invention is applied. The specific electronic device may include more or fewer components than shown in the figure, or combine certain components, or have a different arrangement of components.

[0078] In another aspect, the present invention further provides a computer storage medium storing a computer program, wherein when the computer program is executed by a processor, a method for classifying inland ponds at a watershed scale is implemented, the method comprising: Obtain multi-source geospatial data for the target watershed, including remote sensing images, digital elevation models, land cover data, JRC global surface water products, city boundary data, photovoltaic power station distribution data, mining area polygon data, lake datasets, and statistical yearbook data; Perform radiometric calibration, terrain correction, noise removal and spatial cropping on remote sensing images; Generate baseline maps of inland pond water bodies based on multi-source geospatial data; Based on the differentiated characteristics of aquaculture ponds, urban ponds, photovoltaic ponds, tailings ponds, and farmland ponds, differentiated combination applications are carried out to output identification and classification results of multiple categories of inland ponds. The differentiated combination applications include: using object-oriented analysis combined with geometric feature decision trees and neighborhood space discriminant methods to identify aquaculture ponds; using land cover overlay analysis and buffer ratio verification to identify urban ponds; using random forest algorithm to train power station samples to identify photovoltaic ponds; constructing a convolutional neural network model combined with spectral texture features to identify tailings ponds; and using digital elevation models and neighborhood analysis of cultivated land data to identify farmland ponds.

[0079] In yet another aspect, a computer program product or computer program is provided, the computer program product or computer program comprising computer instructions stored in a computer readable storage medium. A processor of an electronic device reads the computer instructions from the computer readable storage medium, and the processor implements a method for inland pond classification at a catchment scale when executing the computer instructions, the method comprising: obtaining multi-source geospatial data of a target catchment, including remote sensing images, digital elevation model, land cover data, JRC global land surface water product, urban boundary data, photovoltaic power station distribution data, mining area polygon data, lake data set and statistical yearbook data; performing radiation calibration, terrain correction, noise removal and spatial clipping on the remote sensing images; generating an inland pond water body baseline map based on the multi-source geospatial data; performing differential combination application for the differentiated characteristics of aquaculture ponds, urban ponds, photovoltaic ponds, tailings ponds and farmland ponds, and outputting multi-class inland pond recognition and classification results, the differential combination application including: using object-oriented analysis combined with geometric feature decision tree and neighborhood spatial discriminant method to identify aquaculture ponds; using land cover overlay analysis and buffer zone proportion verification to identify urban ponds; using random forest algorithm to train power station samples to identify photovoltaic ponds; constructing a convolutional neural network model combining spectral texture features to identify tailings ponds; and applying digital elevation model and neighborhood analysis of farmland data to identify farmland ponds.

[0080] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by a computer program instructing related hardware, and the computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments. Any reference to memory, storage, database or other medium used in the embodiments provided by the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory.

[0081] By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0082] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0083] The above-described embodiments merely illustrate several embodiments of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make various modifications and improvements without departing from the spirit of the present invention, and these modifications and improvements fall within the scope of the present invention. Therefore, the scope of the present invention shall be determined by the appended claims.

Claims

1. A method for classifying inland ponds at a watershed scale, characterized in that: The method comprises: Obtain multi-source geospatial data for the target watershed, including remote sensing images, digital elevation models, land cover data, JRC global surface water products, city boundary data, photovoltaic power station distribution data, mining area polygon data, lake datasets, and statistical yearbook data; Perform radiometric calibration, terrain correction, noise removal and spatial cropping on remote sensing images; Generate baseline maps of inland pond water bodies based on multi-source geospatial data; Based on the differentiated characteristics of aquaculture ponds, urban ponds, photovoltaic ponds, tailings ponds, and farmland ponds, differentiated combination applications are carried out to output identification and classification results of multiple categories of inland ponds. The differentiated combination applications include: using object-oriented analysis combined with geometric feature decision trees and neighborhood space discriminant methods to identify aquaculture ponds; using land cover overlay analysis and buffer ratio verification to identify urban ponds; using random forest algorithm to train power station samples to identify photovoltaic ponds; constructing a convolutional neural network model combined with spectral texture features to identify tailings ponds; and using digital elevation models and neighborhood analysis of cultivated land data to identify farmland ponds.

2. The method for classifying inland ponds at a watershed scale according to claim 1, wherein: The radiometric calibration, terrain correction, noise removal and spatial cropping of the remote sensing image include: Calculate the median backscatter value of Sentinel-1 SAR cross-year series data to eliminate salt and pepper noise; Sentinel-2 images were radiometrically calibrated and terrain corrected, and then clipped according to the boundaries of the study area.

3. The method for classifying inland ponds at a watershed scale according to claim 1, wherein: The method of generating an inland pond water baseline map based on multi-source geospatial data includes: The dual-polarization water index was calculated based on the JRC global surface water product and Sentinel-1 SAR data, and the initial water mask was generated by determining the threshold using the water frequency method. Sentinel-2 data were fused to optimize the mask boundaries and form a basin-scale baseline map of inland ponds.

4. The method for classifying inland ponds at a watershed scale according to claim 1, wherein: The method of identifying aquaculture ponds by using object-oriented analysis combined with a geometric feature decision tree and a neighborhood space discrimination method includes: Use global city boundary data to demarcate non-urban areas, and overlay a baseline map to extract the rough area of ​​aquaculture ponds; Based on Sentinel-2 images and field sampling data, a geometric feature decision tree was constructed to screen aquaculture ponds.

5. The method for classifying inland ponds at a watershed scale according to claim 4, wherein: The method of constructing a geometric feature decision tree to screen aquaculture ponds comprises: The neighborhood expansion method was used to generate buffer zones for potential ponds, and isolated non-aquaculture ponds were identified based on the intersection of the buffer zones. Fish ponds and shrimp and crab ponds were distinguished by the normalized difference pond index and normalized difference vegetation index.

6. The method for classifying inland ponds at a watershed scale according to claim 1, wherein: The identification of urban ponds using land cover overlay analysis and buffer ratio validation includes: The global city boundary data and SinoLC-1 city class layer were overlaid onto the baseline map to screen the candidate set of urban ponds; Independence was verified using a neighborhood-based expansion method and secondarily confirmed by the proportion of impervious surface in the buffer zone.

7. The method for classifying inland ponds at a watershed scale according to claim 1, wherein: The random forest algorithm is used to train power station samples to identify photovoltaic ponds, including: Collect sample points of photovoltaic power stations in the target area; Set random forest model parameters, train and test the model; The model output was overlaid with the baseline map to determine the PV pond distribution.

8. The method for classifying inland ponds at a watershed scale according to claim 1, wherein: The method of constructing a convolutional neural network model combining spectral texture features to identify tailings ponds includes: Delineate target tailings pond areas based on DEM, SinoLC-1 bare map layers, and global mining area datasets; A CNN model was constructed, spectral and texture features were input, and tailings ponds were identified through training using convolutional layers, maximum pooling layers, and fully connected layers.

9. The method for classifying inland ponds at a watershed scale according to claim 1, wherein: The method of identifying farmland ponds by applying a digital elevation model and neighborhood analysis of cultivated land data includes: Overlay the DEM, SinoLC-1 cultivated land layer, and non-urban boundary data onto the baseline map; Neighborhood analysis was applied to verify whether ponds were independently distributed in the farmland.

10. The method for classifying inland ponds at a watershed scale according to any one of claims 1 to 9, characterized in that: The method further comprises: Output the spatial distribution map of multi-category inland ponds and construct confusion matrices for fish ponds, shrimp and crab ponds, urban ponds, photovoltaic ponds, tailings ponds, and farmland ponds; The Nash coefficient (NS), correlation coefficient (R²), root mean square error (RMSE) and mean absolute error (MAE) are used to evaluate the accuracy of the framework.

Images