Dynamic threshold water surface extraction method based on remote sensing data and air-ground collaborative observation

Abstract

The invention discloses a dynamic threshold water surface extraction method based on remote sensing data and air-ground collaborative observation, and relates to the technical field of hydrological remote sensing, and the method comprises the steps: building a satellite-unmanned aerial vehicle-ground triple observation system, and supplementing a high-risk region which cannot be covered by a remote sensing image through a high-resolution image of an unmanned aerial vehicle and ground measured data; meanwhile, a self-adaptive NDWI threshold value adjustment algorithm is established, actual measurement data are used for counting true positive rates and false positive rates under different threshold values, and an optimal threshold value is determined by constructing an ROC curve; and then, combining space-time statistical analysis to generate an intermittent surface water space-time distribution diagram of the high-altitude region. Therefore, the dynamic threshold water surface extraction method based on remote sensing data and air-ground collaborative observation can adapt to dynamic changes of different regions, seasons and water turbidity, so that the extraction precision and space continuity of the surface water surface boundary are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of hydrological remote sensing technology, and in particular to a dynamic threshold water surface extraction method based on remote sensing data and air-ground collaborative observation. Background Art

[0002] Remote sensing surface water extraction is a technical method based on satellite, aerial or ground sensor technology to identify and quantify the spatial distribution of surface water through spectral characteristics, spatial texture and other information. Its core lies in utilizing the physical properties of the interaction between electromagnetic waves and surface water (such as the strong absorption characteristics of surface water in the near-infrared band) to achieve automatic identification and dynamic monitoring of the earth's surface water (compared with traditional manual surveys or statistical surveys).

[0003] Remote sensing technology, by capturing high-frequency, multi-scale surface water information, can analyze the potential connections between surface water and groundwater, revealing the coupling mechanisms between surface water area changes and local climate feedbacks (such as the evaporation-precipitation cycle). However, mis-extraction issues caused by cloud obscuration, mixed pixels, or seasonal vegetation interference can lead to misjudgment of surface water boundaries and missed detection of small surface water areas, thus affecting the accuracy of water resource assessments and ecological compensation decisions.

[0004] Currently, the Normalized Difference Water Index (NDWI) is widely used to extract surface water. Although NDWI has a good effect in extracting surface water, it still faces new challenges: (1) Current research focuses on mapping the static surface water range, while there is a lack of methods for extracting dynamic processes such as water level fluctuations, temporal changes in water quality parameters (such as turbidity and chlorophyll concentration), and inundation frequency, which makes it difficult to support the refined analysis of hydrological processes and ecological functions; (2) Existing methods generally use global or fixed thresholds to segment surface water, ignoring the spatiotemporal heterogeneity of surface water spectral characteristics caused by seasonal, geographical environment, and sensor differences; (3) Existing methods mostly use pixels or regions as units for threshold segmentation. The former is easily affected by mixed pixel effects and salt and pepper noise, while the latter cannot depict the fine boundaries of small surface water surfaces, resulting in difficulty in effectively matching the extraction results with actual hydrological management units. Summary of the Invention

[0005] The purpose of the present invention is to provide a dynamic threshold water surface extraction method based on remote sensing data and air-ground collaborative observation. By integrating multi-source data of "air-ground" multi-platform collaborative observation, it can form effective complementarity at the temporal and spatial resolution level, realize temporal and spatial dynamic monitoring of surface water surface, and provide effective data support for the subsequent drawing of intermittent surface water surface maps.

[0006] To achieve the above objectives, the present invention provides a dynamic threshold water surface extraction method based on remote sensing data and air-ground collaborative observation, comprising:

[0007] Step 1: Obtain multi-temporal and multi-source remote sensing data, drone image data, and measured surface water sample data of the study area, and generate a raster dataset with a unified geographic spatial reference, matching spatiotemporal resolution, and pixel alignment after preprocessing;

[0008] Step 2: Based on the generated raster dataset, obtain the normalized water index (NDWI) of the raster data of the same or adjacent dates at the corresponding coordinates of the measured location, construct a scatter plot of the ROC curve under different thresholds in each region, and obtain the optimal threshold by maximizing the Youden index value;

[0009] Step 3: Establish an NDWI dynamic threshold determination algorithm based on measured data, set a surface water surface identification rule set based on NDWI changes, and draw a spatiotemporal distribution map of intermittent surface water surfaces in the study area.

[0010] Furthermore, step 1 includes implementing spatiotemporal benchmark unification processing on the acquired multi-source heterogeneous data: geographic coordinate system conversion, spatial resolution resampling, temporal consistency correction and pixel-level spatial alignment to establish a raster dataset with a unified spatiotemporal benchmark.

[0011] Furthermore, step 2 also includes drawing an F1 score curve based on the F1 scores corresponding to different thresholds with the threshold as the horizontal axis to evaluate the stability of the threshold under class imbalance.

[0012] Furthermore, step 3 includes applying the algorithm for determining the NDWI dynamic threshold in different areas based on the measured data, constructing an adaptive NDWI threshold rule set, and mapping the intermittent surface water distribution in the study area.

[0013] Furthermore, step 3 includes analyzing regional surface water information from multiple dimensions and drawing a spatiotemporal distribution map of intermittent surface water in the study area.

[0014] Therefore, the present invention adopts the above-mentioned dynamic threshold water surface extraction method based on remote sensing data and air-ground collaborative observation, which has the following technical effects:

[0015] The present invention takes into account seasonal turbidity changes and regional background interference, and utilizes drone images compared to manual photography, which have comprehensive advantages such as high resolution, rich spectral information, good imaging consistency, flexible flight and automated data acquisition. It is particularly suitable for efficient monitoring of complex terrain and small water bodies. At the same time, the dynamic threshold value of drone and ground measured data analysis is used to replace the fixed threshold value in the traditional method, and accurate calculation is performed for the dynamic changes of turbidity in different regions, seasons and surface water surfaces, which significantly improves the accuracy of surface water boundary extraction and spatial continuity, becoming an efficient and robust surface water detection method in complex environments, thereby more accurately drawing intermittent surface water distribution maps in high-altitude areas.

[0016] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a schematic diagram of a dynamic threshold water surface extraction method based on remote sensing data and air-ground collaborative observation;

[0018] Figure 2 It is a flow chart of multi-source data preprocessing in an embodiment of a dynamic threshold water surface extraction method based on remote sensing data and air-ground collaborative observation;

[0019] Figure 3 It is a flow chart of determining the NDWI dynamic threshold value of a high altitude area based on a ROC curve scatter plot in an embodiment of a dynamic threshold water surface extraction method based on remote sensing data and air-ground collaborative observation;

[0020] Figure 4 This is an intermittent surface water distribution map of Lhasa in an embodiment of a dynamic threshold water surface extraction method based on remote sensing data and air-ground collaborative observation. DETAILED DESCRIPTION

[0021] The present invention can be explained in more detail by the following examples. The purpose of disclosing the present invention is to protect all changes and improvements within the scope of the present invention. The present invention is not limited to the following examples.

[0022] like Figure 1 As shown in the figure, the present invention provides a dynamic threshold water surface extraction method based on remote sensing data and air-ground collaborative observation, including the acquisition and preprocessing of multi-source remote sensing data and measured surface water surface data, the establishment of a ROC curve scatter plot with FPR as the horizontal axis and TPR as the vertical axis under different thresholds, the determination of the optimal threshold by maximizing the Youden index, the construction of a surface water surface identification rule set based on NDWI changes, and the statistics and mapping of intermittent information of surface water surfaces.

[0023] Step 1: Obtain time series remote sensing data (including Sentinel-2 images, DEM terrain data) and auxiliary data (drone images, measured surface water samples), and pre-process them to generate raster data that is temporally and spatially aligned and has consistent georeferences. Figure 2 Using the Sentinel-2 spatial and temporal resolution as a benchmark, we determined parameters such as the geographic spatial reference R, spatial resolution S, and temporal resolution T. Based on this, we first performed cloud masking on the time series optical images. We then converted the time series raster dataset and surface water surface vectors into the geographic spatial reference R. The time series raster data was then resampled to the spatial resolution S, and the spatial and temporal positions of the pixels were strictly aligned. Finally, the mean characteristic value of the raster dataset pixels within the region was calculated and used as the characteristic value of the surface water surface to generate the surface water surface NDWI dataset.

[0024] Step 2: To address the problems of the existing technology, this embodiment provides a method for determining NDWI dynamic thresholds. This method uses drone imagery and ground data analysis to set dynamic thresholds for different regions, adapting them to the dynamic changes in different regions, seasons, and water turbidity. This significantly improves the extraction accuracy and spatial continuity of surface water boundaries, making it particularly suitable for surface water extraction in high-altitude areas. Figure 3 ,The determination of NDWI dynamic threshold includes three steps:

[0025] (1) Based on the surface water NDWI dataset generated by preprocessing, the NDWI values ​​were classified into two categories according to different fixed thresholds, and the intervals were divided into intervals of 0.01 or 0.05 to generate a classification result matrix; all thresholds were traversed, and the true positive rate (TPR) and false positive rate (FPR) under each threshold were calculated, and a receiver operating characteristic (ROC) curve scatter plot was drawn with FPR as the horizontal axis and TPR as the vertical axis.

[0026] Among them, the true positive rate (TPR):

[0027]

[0028] This indicator reflects the ability to correctly identify water bodies, TP is the true positive sample, and FN is the number of missed water bodies.

[0029] False Positive Rate (FPR):

[0030]

[0031] This indicator reflects the error rate of misjudging non-water bodies as water bodies, FP is the number of false positive samples, and TN is the number of non-water bodies correctly excluded.

[0032] (2) Draw a TPR-FPR scatter plot (ROC curve), and the area under the curve (AUC) measures the overall performance of the classifier. The closer the AUC is to 1, the stronger the model's discrimination ability. In the process of determining the threshold, it is necessary to prioritize the integrity and accuracy of water body extraction to avoid extreme cases of missed classification (such as small water bodies in arid areas) and misclassification (such as shadows or bare soil). To this end, this embodiment uses a method of maximizing the Youden index to quantify the comprehensive performance of the threshold, thereby determining the optimal threshold, that is, selecting the threshold closest to the upper left corner on the ROC curve, which can balance sensitivity (TPR) and specificity (1-FPR).

[0033] (3) To address the problem of sample imbalance (e.g., low proportion of water bodies), this embodiment also calculates the following indicators, which are mainly used to verify the generalization degree of the optimal threshold:

[0034] Precision: This indicator characterizes the proportion of actual water bodies among the results predicted as water bodies, focusing on reducing false positives.

[0035] Recall rate (TPR): R = TPR. This indicator focuses on reducing missed detections.

[0036] F1 score: This metric, the harmonic mean of precision and recall, is used to assess the stability of the threshold under class imbalance. During validation, it is important to ensure that the F1 score curve is close to its peak to ensure generalization across time and space. The horizontal axis of the F1 score curve corresponds to different thresholds.

[0037] Step 3: First, based on measured data, an NDWI dynamic threshold determination algorithm is applied to different regions (e.g., Lhasa, Nyingchi, Shannan, Shigatse, and other cities). A set of surface water surface identification rules based on NDWI changes is then established. For example, a set of rules is used to count the number of days D during a natural year when the NDWI value is below a threshold. If D is less than 5 days, it indicates normal flow; if D is greater than or equal to 5 days, it indicates drought. In other embodiments, more rules can be added based on actual conditions.

[0038] Then, based on the optimal NDWI thresholds in different regions, regional surface water information is statistically analyzed from multiple dimensions. From a temporal analysis perspective, the optimal NDWI thresholds for a region are calculated year by year (month by month), and intermittent surface water distribution maps are produced year by year (month by month). From a spatial analysis perspective, the optimal NDWI thresholds for different regions are calculated, and intermittent surface water distribution maps are produced for the corresponding regions. Figure 4 In this embodiment, the intermittent surface water distribution map of Lhasa City is produced using the above-mentioned optimal threshold. Compared with the fixed NDWI threshold algorithm, the method proposed in the present invention can more accurately depict the intermittent situation of the intermittent surface water surface in Lhasa City, which is closer to the actual situation and reduces the misidentification and omission of surface water surfaces.

[0039] Therefore, the present invention adopts the above-mentioned dynamic threshold water surface extraction method based on remote sensing data and air-ground collaborative observation, which can quickly and accurately calculate the NDWI threshold for different regions, seasons and dynamic changes in surface water turbidity, and generate intermittent river distribution maps in altitude areas. It is suitable for the fine identification of surface water surfaces at different altitudes in the Yarlung Zangbo River basin on the Qinghai-Tibet Plateau, so as to improve the accuracy of water resource monitoring in complex terrain areas and support decisions such as water resource management and agricultural production activities.

[0040] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A dynamic threshold water surface extraction method based on remote sensing data and air-ground collaborative observation is characterized by: include: Step 1: Obtain multi-temporal and multi-source remote sensing data, drone image data, and measured surface water sample data of the study area, and generate a raster dataset with a unified geographic spatial reference, matching spatiotemporal resolution, and pixel alignment after preprocessing; Step 2: Based on the generated raster dataset, obtain the normalized water index (NDWI) of the raster data of the same or adjacent dates at the corresponding coordinates of the measured location, construct a scatter plot of the ROC curve under different thresholds in each region, and obtain the optimal threshold by maximizing the Youden index value; Step 3: Establish an NDWI dynamic threshold determination algorithm based on measured data, set a surface water surface identification rule set based on NDWI changes, and draw a spatiotemporal distribution map of intermittent surface water surfaces in the study area.

2. The dynamic threshold water surface extraction method based on remote sensing data and air-ground collaborative observation according to claim 1 is characterized in that: Step 1 involves implementing spatiotemporal benchmark unification processing on the acquired multi-source heterogeneous data: geographic coordinate system conversion, spatial resolution resampling, temporal consistency correction, and pixel-level spatial alignment to establish a raster dataset with a unified spatiotemporal benchmark.

3. The dynamic threshold water surface extraction method based on remote sensing data and air-ground collaborative observation according to claim 1 is characterized in that: Step 2 also includes drawing an F1 score curve based on the F1 scores corresponding to different thresholds with the threshold as the horizontal axis to evaluate the stability of the threshold under class imbalance.

4. The dynamic threshold water surface extraction method based on remote sensing data and air-ground collaborative observation according to claim 1 is characterized in that: Step 3 involves applying the algorithm for determining the NDWI dynamic threshold in different areas based on the measured data, constructing an adaptive NDWI threshold rule set, and mapping the intermittent surface water distribution in the study area.

5. The dynamic threshold water surface extraction method based on remote sensing data and air-ground collaborative observation according to claim 1 is characterized in that: Step 3 involves analyzing regional surface water information from multiple dimensions and drawing a spatiotemporal distribution map of intermittent surface water in the study area.

Images