Pit-pond counting method based on multi-source remote sensing image
By combining optical and radar remote sensing images and using edge detection and random forest classifiers to identify the surface of ponds, the problem of identifying and counting pond boundaries in large-scale areas was solved, and rapid and accurate statistics on the number of ponds were achieved.
Patent Information
- Application Number
- CN202511471943.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-15
- Publication Date
- 2026-01-09
AI Technical Summary
Existing technologies struggle to accurately identify and count the boundaries of pits and ponds over large-scale areas. In particular, medium-resolution remote sensing images have difficulty identifying the boundaries of individual pits and ponds and distinguishing the lines between them, resulting in complex distribution characteristics of pits and ponds and severe limitations in mixed pixels.
By combining optical and radar remote sensing, the spatial distribution of pond water surfaces is identified through median composite images. The pond water bodies are marked using edge detection and connection component labeling algorithms. The pond water surfaces are identified by combining random forest classifiers and multiple spectral indices. The data is then filtered and converted into vector surface data, and finally, individual pond patches are segmented and counted.
It enables rapid and relatively accurate automatic counting of the number of pits and ponds in a large-scale area, with an accuracy of less than 20%, effectively overcoming the problems of mixed pixels and blurred boundaries, and improving the accuracy of pit and pond monitoring.
Smart Images
Figure CN121305348A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of land use remote sensing identification and relates to a method for counting pits and ponds based on multi-source remote sensing images. Background Technology
[0002] Ponds, as a major form of aquaculture, are widely distributed in river and lake plains, coastal mudflats, and other areas, and have seen rapid development in many countries and regions in recent decades. However, the expansion of ponds has also brought many environmental challenges, such as disorderly construction and unreasonable planning, which may lead to reduced water and soil resource utilization efficiency, eutrophication of water bodies, and excessive greenhouse gas emissions. At the same time, the construction of a large number of ponds in coastal and lakeside areas has also exacerbated the degradation of wetland ecosystems such as mangroves, salt marshes, and mudflats, resulting in the loss of animal and plant habitats and a decline in ecosystem stability.
[0003] Monitoring ponds helps provide a macro-level perspective for their planning and management, thereby curbing their disorderly expansion and achieving green and circular development.
[0004] Currently, remote sensing technology is gradually becoming an important means of monitoring ponds and pits, breaking through the limitations of traditional survey and statistical methods in terms of spatiotemporal scope, and its effectiveness has been proven in numerous studies. There are diverse types of remote sensing data, each with its own advantages and disadvantages. In high-resolution images (<5 m), such as Quickbird, PlanetScope, and Gaofen-2, the morphological features of ponds and pits are obvious and their boundaries are clear, but due to the high cost of image acquisition and limited coverage, they are difficult to apply to large-scale regional monitoring. Synthetic Aperture Radar (SAR) images are not affected by weather conditions, have the ability to penetrate clouds and fog, and can provide complete time-series data, but their imaging mechanism is complex and easily affected by topographic relief and speckle noise. Medium-resolution optical images, such as Landsat-5 / 7-9 and Sentinel-2, have been widely used in pond identification research and have achieved certain results. However, current research on pond identification usually only focuses on the water surface of the pond and is difficult to identify individual ponds. This is because most studies use medium-resolution remote sensing images with a resolution of 10-30 meters, while the water surface area of ponds is small and the boundaries mostly show mixed pixels. At the same time, large-scale aquaculture ponds exist in contiguous areas, making the boundaries between ponds even more blurred and difficult to identify.
[0005] In summary, methods for identifying ponds and pools have been developed, but the distribution characteristics of ponds and pools are complex and severely limited by mixed pixels, and there is still no method for accurate identification of pond and pool boundaries and for counting ponds and pools. Summary of the Invention
[0006] The primary objective of this invention is to provide a method for counting ponds based on multi-source remote sensing images. This method combines optical remote sensing and radar remote sensing, and utilizes theoretical principles from multiple disciplines such as optics and morphology to gradually identify pond embankments and divide individual pond patches, ultimately achieving the calculation of the number of ponds.
[0007] To achieve the above objectives, the present invention adopts the following technical solution:
[0008] A method for counting pits and ponds based on multi-source remote sensing imagery, the method comprising:
[0009] Median composite was performed on radar and optical remote sensing images of the study area, and the spatial distribution of pond water surfaces was identified in the composite image.
[0010] Non-water pixels are extracted from the surface of the pit using edge detection, and these non-water pixels are then assigned as land pixels.
[0011] The assigned image is then used to mark the complete pond water body using a connection component marking algorithm.
[0012] Filter out ponds with sizes exceeding a preset size value from the complete pond water body, calculate the mean difference of the pond water body using a moving window, mark the pixels with mean values exceeding a preset threshold, and assign them the value of land pixels;
[0013] All land pixels are converted into vector surface data based on the location of the land pixel's centerline;
[0014] The spatial distribution of the pond water surface is segmented using the vector surface data, and individual pond patches are marked and counted.
[0015] In some embodiments of the present invention, the VV polarization backscattering coefficient, VH polarization backscattering coefficient, band spectral index, optical reflectance of each band, altitude, slope and surface temperature are extracted from the synthetic image as input features, and a random forest classifier is used to identify the surface of the pond.
[0016] In some embodiments of the present invention, the band spectral indices include NDVI index, EVI index, MNDWI index, NDSI index, NDBI index, SDWI index, and AWEI index. sh Index and AWEI nsh index.
[0017] In some embodiments of the present invention, post-processing is performed on the pond surface identification results obtained by the classifier, including:
[0018] Identify and label anomalous pixels;
[0019] A sliding window is used to perform neighborhood statistics on each abnormal cell, and the mode value of the neighborhood statistics is used to replace the original abnormal cell value.
[0020] Remove small, scattered patches with an area smaller than the preset area value;
[0021] Based on the lake and reservoir ranges obtained from land surveys, the identification results of pond water surfaces are masked, the shape index of water patches is calculated, and water bodies with a shape index greater than a preset threshold are removed.
[0022] In some embodiments of the present invention, the SDWI index and the MNDWI index are combined with the Canny operator to identify non-water body pixels at the edge.
[0023] The union of the recognition results of the two indices is used as the recognition result of non-water body pixels.
[0024] In some embodiments of the present invention, the mean difference of water bodies in ponds is calculated based on the SDWI index.
[0025] In some embodiments of the present invention, after the land pixels are dilated in an independent layer, the dilation result is converted into vector line data according to the position of the center line, and then the vector line data is converted into vector surface data according to a fixed pixel length.
[0026] In some embodiments of the present invention, after erosion processing is performed on the spatial distribution results of the segmented pit and pond water surface, individual pit and pond patches are marked.
[0027] In some embodiments of the present invention, the method further includes counting the remaining individual pond patches after removing those with an area smaller than a preset area threshold.
[0028] In some embodiments of the present invention, the radar remote sensing images and optical remote sensing images are medium-resolution remote sensing images.
[0029] The method of this invention combines optical remote sensing and radar remote sensing data sources, and uses classification features and morphological methods to identify the spatial distribution of pond water surfaces, identify pond embankments and divide individual pond patches, enabling rapid and relatively accurate automatic counting of the number of ponds in large-scale areas.
[0030] It should be understood that all combinations of the foregoing concepts and the additional concepts described in more detail below may be considered part of the inventive subject matter of this disclosure, provided that such concepts do not contradict each other. Furthermore, all combinations of the claimed subject matter are considered part of the inventive subject matter of this disclosure.
[0031] The foregoing and other aspects, embodiments, and features of the teachings of the present invention will be more fully understood from the following description in conjunction with the accompanying drawings. Other additional aspects of the invention, such as features and / or beneficial effects of exemplary embodiments, will become apparent from the following description or may be learned through practice of specific embodiments according to the teachings of the present invention. Attached Figure Description
[0032] The accompanying drawings are not intended to be drawn to scale. In the drawings, each identical or nearly identical component shown in the various figures may be denoted by the same reference numeral. For clarity, not every component is labeled in each figure. Embodiments of various aspects of the invention will now be described by way of example and with reference to the accompanying drawings, wherein:
[0033] Figure 1 This is a flowchart of the method of the present invention.
[0034] Figure 2 These are the results of water extraction in the study area.
[0035] Figure 3 These are the results extracted from aquaculture ponds in the study area.
[0036] Figure 4 The results were extracted from aquaculture ponds in a typical area of the study region.
[0037] Figure 5 These are the pixel recognition results of the pit and pond boundaries in the study area.
[0038] Figure 6 These are the vectorization results of the pit and pond boundary pixels in the study area.
[0039] Figure 7 These are the results of pit and pond patch segmentation in the study area.
[0040] Figure 8 This is the result of erosion at the boundaries of pits and ponds in the study area.
[0041] In the aforementioned Figures 1-8, the coordinates, symbols, or other representations expressed in English are all well-known in the field and will not be elaborated upon in this example. Detailed Implementation
[0042] To better understand the technical content of the present invention, specific embodiments are described below in conjunction with the accompanying drawings.
[0043] Various aspects of the invention are described in this disclosure with reference to the accompanying drawings, in which numerous illustrative embodiments are shown. The embodiments of this disclosure are not necessarily intended to encompass all aspects of the invention. It should be understood that the various concepts and embodiments described above, as well as those described in more detail below, can be implemented in any of many ways, as the concepts and embodiments disclosed herein are not limited to any particular implementation. Furthermore, some aspects of the invention disclosed may be used alone or in any suitable combination with other aspects of the invention disclosed.
[0044] Example 1
[0045] This embodiment uses a coastal region A as an example to further describe the technical method of the present invention. The study area covers approximately 144,200 km².
[0046] The flowchart of the pit and pond counting method based on multi-source remote sensing imagery shown in the embodiment is as follows: Figure 1 As shown, the image data source used in this embodiment is medium-resolution Sentinel-1 radar remote sensing image and Sentinel-2 optical remote sensing image, and specifically includes the following steps:
[0047] Step 1: Preliminary extraction of the spatial distribution of pond water surface: Based on the backscattering coefficient of Sentinel-1 GRD data and the optical reflectance based on Sentinel-2 MSI data, a multi-dimensional feature variable library is constructed. The spatial distribution of pond water surface is obtained using a random forest classifier, as detailed below:
[0048] Based on the GEE platform, Sentinel-1 SAR GRD image data and Sentinel-2 MSI L2A remote sensing image data were acquired. Image data from April to October of the corresponding region were selected, and after median composite, the backscattering coefficients and optical reflectivities of the VV and VH polarization channels were extracted, and characteristic spectral indices such as NDVI, MNDWI, NDSI, and SDWI were calculated. The calculation methods are shown in Table 1.
[0049] Table 1. Formulas for Calculating Characteristic Spectral Indices
[0050]
[0051] Note: ρ blue ρ green ρ red ρ nir ρ swir ρ swir2 These refer to the reflectivity values for the corresponding wavebands; VV and VH refer to the backscattering coefficient values of the VV and VH polarization channels, respectively.
[0052] Data on elevation, slope, and surface temperature parameters of the study area were obtained. Based on the extracted characteristic spectral indices, backscattering coefficients of VV and VH polarization channels, spectral reflectance of each band, and elevation, slope, and surface temperature parameters of the study area, a multidimensional set of characteristic variables was constructed.
[0053] Random forest was used as the machine learning model for identifying aquaculture ponds. A classification model for the study area was trained using a multidimensional feature variable set, and preliminary classification results were obtained using the model, such as... Figure 2 The initial classification image may contain scattered small patches or isolated pixels with a class anomalous characteristics, whose class is significantly inconsistent with surrounding pixels. To eliminate these anomalies, post-processing of the preliminary results is necessary. Specifically, firstly, these fragmented pixels (i.e., the aforementioned scattered small patches or isolated pixels) are identified and labeled; then, a sliding window is used to perform neighborhood statistics on each anomalous pixel, and the original anomalous pixel value is replaced with the mode of the neighborhood statistics (the value with the highest frequency of raster pixel values within the statistical area). Finally, pixels with areas too small (<0.0025 km²) are removed. 2 The small, scattered patches were used to obtain a distribution map of the aquaculture ponds.
[0054] The distribution map of aquaculture ponds was further corrected using land survey results and shape indices. Since lakes and reservoirs do not show significant changes in a short period, the water body extraction results were masked using the reservoir and lake areas obtained from the most recent national land survey. The Third National Land Survey was conducted in 2017, and the results from various regions were compiled and released in 2018; therefore, the results of the Third National Land Survey were used to mask lakes and reservoirs. Simultaneously, the shape index of water body patches was calculated. When the shape index was greater than 8.00, it was identified as a river or ditch, and this part of the water body was removed. The final processed result is as follows: Figure 3 , Figure 4 This is the result of processing a typical area.
[0055] Finally, the verification accuracy of pit and pond extraction is shown in Table 2. Based on the confusion matrix, relevant indicators were further calculated. The overall accuracy was 0.9637, the Kappa coefficient was 0.9271, the mapping accuracy was 0.95068, and the user accuracy was 0.97198.
[0056] Table 2 Confusion Matrix
[0057]
[0058] Step 2, Pond Boundary Vectorization: First, edge detection is performed on the pond surface image. SDWI and MNDWI indices are used to identify mixed land and water pixels at the edges. At the same time, mean difference is used to identify pond embankment pixels in contiguous ponds. The mixed land and water pixels and pond embankment pixels are then converted into vector data, as follows:
[0059] First, the edges of the ponds were detected using SDWI and MNDWI indices combined with the Canny algorithm. The union of the SDWI and MNDWI indices was then used as the identification result for mixed water and land pixels. Considering the high turbidity and high reflectivity of the water in high-density aquaculture ponds, the MNDWI threshold was set to 0.1 and the SDWI threshold was set to 0.25 to obtain non-water pixels (including mixed water and land pixels and pure land pixels).
[0060] After assigning non-water pixels to land pixels, the connectedComponents function is used to mark complete pond water bodies. Pond water bodies with both length and width exceeding 5 pixels are filtered out. The mean difference is calculated using a moving window, and the threshold is set to 0.1 (SDWI value). Pixels with SDWI values exceeding the threshold of 0.1 are identified as mixed pixels in the pond water body, which are the pixels where the pond embankment is located.
[0061] After assigning the pond embankment pixel to the land pixel, for all land pixels (such as...) Figure 5 Dilation is performed in units of 1.42 pixels to connect the embankment pixels, ensuring the continuity of subsequent vector lines. Then, it is converted into vector line data based on the center position, such as... Figure 6 Then, it is converted into vector surface data with a length of 0.5 pixels. This operation is done in a new separate layer to prevent the pond from disappearing due to the expansion of the pond embankment pixels. The vector surface is then loaded into the pond water surface spatial distribution data layer.
[0062] Step 3, Patch Segmentation and Counting: The spatial distribution of pond water surfaces is segmented using vector surface data, such as... Figure 7 The segmented pit vectors are subjected to erosion processing, such as... Figure 8 Independent individual pond patches are marked with connected regions, and the number of ponds is finally obtained.
[0063] The spatial distribution of the pond water surface was first converted into a vector and then segmented, with an erosion treatment length of 1.42. Then, areas less than 0.0025 km² were considered. 2 After removing broken patches, the final number of pits and ponds was 342,538. The number obtained by visual counting using the corresponding Google high-resolution image was 428,010, with an error of 20%. The method of this invention has good accuracy and can effectively count pits and ponds.
[0064] While the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the invention. Those skilled in the art to which this invention pertains can make various modifications and refinements without departing from the spirit and scope of the invention.
Claims
1. A method for counting pits and ponds based on multi-source remote sensing imagery, characterized in that, The method includes: Median composite was performed on radar and optical remote sensing images of the study area, and the spatial distribution of pond water surfaces was identified in the composite image. Non-water pixels are extracted from the surface of the pit using edge detection, and these non-water pixels are then assigned as land pixels. The assigned image is then used to mark the complete pond water body using a connection component marking algorithm. Filter out ponds with sizes exceeding a preset size value from the complete pond water body, calculate the mean difference of the pond water body using a moving window, mark the pixels with mean values exceeding a preset threshold, and assign them the value of land pixels; All land pixels are converted into vector surface data based on the location of the land pixel's centerline; The spatial distribution of the pond water surface is segmented using the vector surface data, and individual pond patches are marked and counted.
2. The method according to claim 1, characterized in that, The VV polarization backscattering coefficient, VH polarization backscattering coefficient, band spectral index, optical reflectance of each band, altitude, slope and surface temperature are extracted from the synthetic image as input features, and a random forest classifier is used to identify the surface of the pond.
3. The method according to claim 2, characterized in that, The spectral indices for this band include NDVI index, EVI index, MNDWI index, NDSI index, NDBI index, SDWI index, and AWEI index. sh Index and AWEI nsh index.
4. The method according to claim 1, characterized in that, Post-processing is performed on the pond surface identification results obtained using the classifier, including: Identify and label anomalous pixels; A sliding window is used to perform neighborhood statistics on each abnormal cell, and the mode value of the neighborhood statistics is used to replace the original abnormal cell value. Remove small, scattered patches with an area smaller than the preset area value; Based on the lake and reservoir ranges obtained from land surveys, the identification results of pond water surfaces are masked, the shape index of water patches is calculated, and water bodies with a shape index greater than a preset threshold are removed.
5. The method according to claim 1, characterized in that, The SDWI index and MNDWI index, combined with the Canny operator, were used to identify non-water body edge pixels. The union of the recognition results of the two indices is used as the recognition result of non-water body pixels.
6. The method according to claim 1, characterized in that, The difference in mean water quality in ponds was calculated based on the SDWI index.
7. The method according to claim 1, characterized in that, After dilating the land pixels in an independent layer, the dilution result is converted into vector line data according to the position of the center line, and then the vector line data is converted into vector surface data according to a fixed pixel length.
8. The method according to claim 1 or 7, characterized in that, After performing erosion treatment on the spatial distribution of the segmented pits and ponds, individual pit and pond patches are marked.
9. The method according to claim 8, characterized in that, It also includes counting the remaining individual pond patches after removing those with an area smaller than a preset area threshold.
10. The method according to claim 1, characterized in that, The radar remote sensing images and optical remote sensing images are medium-resolution remote sensing images.