Coastline classification method based on land-sea overall planning
By using UAV aerial photography and aerial triangulation optimization techniques to select training samples, combined with clustering processing and manual interpretation, the problem of low efficiency in coastline classification in existing technologies has been solved, achieving high-precision automated classification that is suitable for coastal zone management under the integrated land and sea approach.
Patent Information
- Application Number
- CN202510757274.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-09
- Publication Date
- 2025-10-28
AI Technical Summary
Existing coastline classification methods rely on manual selection of training samples, resulting in low efficiency and accuracy affected by subjective experience, making it difficult to meet the needs of large-scale refined coastal zone management under the integrated land and sea approach.
High-precision remote sensing images were acquired through drone aerial photography. Combined with aerial triangulation optimization and supervised classification techniques, a training sample set with a separation value greater than 1.8 was selected. Clustering was used to optimize the classification patches, and manual visual interpretation was combined with iterative adjustments to ensure the consistency of the classification results.
It has significantly improved the automation level and interpretation accuracy of coastal land feature classification, reduced subjective errors, improved classification efficiency, and met the needs of large-scale coastal zone refined management.
Smart Images
Figure CN120852837A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of coastline classification technology, and in particular to a coastline classification method based on land-sea integration. Background Technology
[0002] Existing technologies for coastline classification have significant limitations. Currently, mainstream methods rely heavily on high-resolution imagery acquired through UAV aerial photography, extracting ground feature information through a combination of supervised classification and visual interpretation. While these techniques can achieve a certain level of accuracy, their core bottleneck lies in the selection and optimization of training samples. Existing methods require manual visual selection of training samples, with repeated adjustments to sample distribution and quantity to improve classification accuracy. This process is not only time-consuming and labor-intensive, but the balance and representativeness of sample selection are also easily influenced by subjective experience. This deficiency directly restricts the efficiency and reliability of coastline classification, making it difficult to meet the needs of large-scale, refined coastal zone management under integrated land-sea management. Therefore, a technical method that optimizes the generation and classification of training samples is urgently needed to improve the automation level and interpretation accuracy of coastline classification. Summary of the Invention
[0003] The purpose of this invention is to address the shortcomings of existing technologies by providing a land-sea integrated coastline classification method, thereby solving the problem that the efficiency and reliability of existing coastline classification methods need to be improved.
[0004] This invention provides a coastline classification method based on land-sea integration, comprising the following steps:
[0005] S1. Aerial photography of the target coastal area is carried out by an UAV equipped with an optical camera to obtain remote sensing images of a preset range in the direction of the vertical coastline of the sea and land, and simultaneously record positioning and attitude data.
[0006] S2. Perform aerial triangulation on the aerial images, import the spatial coordinates of pre-deployed image control points for point optimization, and generate a two-dimensional orthophoto.
[0007] S3. Based on orthophotos, automatically interpret coastal land features using supervised classification methods, specifically including:
[0008] S31. In orthophotos, select training samples of various land cover types in a balanced manner, calculate the Jeffries-Matusta distance and transform divergence between samples, and filter the sample set with a separation value greater than 1.8.
[0009] S32. Classify the orthophoto using the maximum likelihood method and output the initial classification results;
[0010] S4. Perform clustering on the initial classification results, merge classification patches with poor spatial continuity, and generate an optimized classification map;
[0011] S5. Verify the accuracy of the optimized classification map through manual visual interpretation. If the difference between the classification result and the manual interpretation exceeds the threshold, readjust the training samples and iterate through steps S3 to S4 until the accuracy requirements are met.
[0012] Further, step S1 includes:
[0013] S11. Evenly deploy image control points at preset intervals in the target coastal zone area, and measure and obtain the spatial coordinates of each image control point through real-time dynamic positioning technology of the global navigation satellite system.
[0014] S12. Set the UAV's flight altitude and heading / lateral overlap, and generate an aerial survey line file covering the target area;
[0015] S13. Perform aerial photography missions during periods of non-direct sunlight and check the integrity of the aerial photography data.
[0016] Furthermore, in step S11, the control point markers have significant spectral differences from the surrounding ground features.
[0017] Furthermore, in step S31, the selection of training samples satisfies the following conditions:
[0018] S311. The sample size for each land cover type is evenly distributed;
[0019] S312. The sample area avoids the extreme spectral response regions of the coastal transition zone.
[0020] Further, step S4 includes:
[0021] S41. Merge classified patches using clustering tools;
[0022] S42. Calculate the number of pixels and standard deviation of each object category after merging, and generate a classification statistical report.
[0023] Furthermore, in step S41, the clustering tool merges small patches from neighboring similar classification regions.
[0024] Furthermore, in step S5, manual visual interpretation is verified using quadrats of a preset size, with 3 to 5 quadrats per type of land cover.
[0025] Furthermore, the sample plot is a square region with a side length that matches the spatial distribution characteristics of land cover types.
[0026] Furthermore, in step S32, the maximum likelihood method is used to perform classification according to the default parameters.
[0027] Furthermore, in step S2, the aerial triangulation optimization process includes iterative correction of the control point residuals after the puncture point.
[0028] The beneficial effects of this invention are as follows: This invention acquires high-precision remote sensing images through UAV aerial photography and combines aerial triangulation optimization and supervised classification techniques, solving the problems of existing technologies where training sample selection relies on human experience and classification accuracy is affected by sample balance. By selecting training sample sets with a dispersibility value greater than 1.8 and using clustering to optimize classification patches, the automation level and interpretation accuracy of coastal zone land cover classification are significantly improved. Simultaneously, through manual visual interpretation and iterative adjustments, the consistency between classification results and actual features is ensured. This method reduces subjective errors, improves coastline classification efficiency, and can meet the needs of large-scale, refined coastal zone management under integrated land and sea management. Attached Figure Description
[0029] Figure 1 This is a flowchart of the coastline classification method based on land-sea integration of the present invention. Detailed Implementation
[0030] The following non-limiting embodiments are intended to enable those skilled in the art to more fully understand the present invention, but do not limit the present invention in any way. It should be noted that the following embodiments should not be construed as limiting the scope of protection of the present invention. If those skilled in the art make some non-essential improvements and adjustments to the present invention based on the above description, they shall still fall within the scope of protection of the present invention.
[0031] Please combine Figure 1 This invention provides a method for coastline classification based on land-sea integration, comprising the following steps:
[0032] S1. Aerial photography of the target coastal area is carried out by an UAV equipped with an optical camera to obtain remote sensing images of a preset range in the direction of the vertical coastline of the sea and land, and the positioning and attitude data are recorded simultaneously.
[0033] Specifically, step S1 includes: S11, uniformly distributing image control points at preset intervals in the target coastal area, and measuring and obtaining the spatial coordinates of each image control point using real-time dynamic positioning technology of the Global Navigation Satellite System; S12, setting the UAV's flight altitude and heading / lateral overlap, and generating an aerial survey line file covering the target area; S13, performing the aerial survey during periods of non-direct sunlight and checking the integrity of the aerial survey data. In step S11, the image control point markers have significant spectral differences from surrounding ground features.
[0034] In this embodiment, the UAV is equipped with a high-resolution optical camera to conduct aerial photography of the coastal area and acquire high-resolution remote sensing images of the coastal area.
[0035] To improve the spatial positioning accuracy of high-resolution remote sensing imagery, image control points (ADCs) are pre-deployed in the target coastal area. The spacing between ADCs is generally between 100m and 200m, and the points are evenly distributed. The ADC markers should be easily identifiable and their precise locations extracted. After deployment, the spatial positions of the ADCs are measured using Global Navigation Satellite System Real-Time Kinematic (GNSS RTK) technology.
[0036] The quality of data collected by the drone directly determines the quality of the subsequent orthophoto. It is recommended to choose the right time to take photos with the drone. Avoid collecting data between 12 noon and 2 pm, as direct sunlight can easily cause the photos to be overexposed.
[0037] Based on the target coastal zone area (2km vertically to the coastline for both sea and land areas), ground resolution, and other requirements, the UAV's flight altitude and overlap are set to generate UAV aerial survey line files. Before takeoff, the aircraft and camera are carefully checked for proper functioning. The UAV then conducts aerial surveys according to the set flight path. Different aircraft models generate other related files besides photos, such as Position and Orientation System Data (POS) files. After the flight, the files are first checked for missing information and missed shots, and then the image quality of each photo is checked.
[0038] S2. Perform aerial triangulation on the aerial images, import the spatial coordinates of pre-deployed image control points for point optimization, and generate a two-dimensional orthophoto.
[0039] Specifically, in step S2, the aerial triangulation optimization process includes iterative correction of the control point residuals after the puncture point.
[0040] In this embodiment, the aerial images and related files are imported into professional UAV aerial photography processing software. Then, the corresponding coordinate system is set, and the ground control points are imported into the processing software. Aerial triangulation is performed first, followed by ground control point identification, and then aerial triangulation optimization. The 2D reconstruction function is selected to generate a 2D orthophoto image, which is then exported. Finally, supervised classification and visual interpretation are used to interpret the ground features in the image.
[0041] S3. Based on orthophotos, automatically interpret coastal land features using supervised classification methods. Specifically, this includes: S31. Selecting training samples of various land feature types evenly from the orthophotos, calculating the Jeffries-Matusta distance and transform divergence values between samples, and selecting a sample set with a dispersiveness value greater than 1.8; S32. Classifying the orthophotos using the maximum likelihood method and outputting the initial classification results. In step S31, the selection of training samples meets the following conditions: S311. The number of samples for each land feature type is evenly distributed; S312. The sample area avoids extreme spectral response regions of the coastal transition zone. In step S32, the maximum likelihood method is run with default parameters for classification.
[0042] In this embodiment, the land features within the flight area are divided into multiple categories according to the sea and land use attributes. Then, the maximum likelihood method of supervised classification in ENVI 5.3 is used to extract different land features. This is mainly accomplished by the following steps.
[0043] Define training samples. Classify the land cover types and select training samples for each type from the aerial imagery using a visual selection method. Export and save the samples. The selection principle is that the training samples for each land cover type are relatively balanced, small, numerous and detailed, and extreme areas should be avoided as much as possible.
[0044] Evaluate the training samples. Place all ground feature training samples in ENVI 5.3 and use the Separability Calculation function (Option / Compute ROI Separability) to calculate the separability between each ground feature sample category. The degree of separability is represented by the Jeffries-Matusita distance and the Transformed Divergence parameter. When the separability value between all ground feature samples is greater than 1.8, it indicates good separability between categories, and the selected training samples can be used for supervised classification.
[0045] Supervised classification. In ENVI 5.3, using the maximum likelihood module, select training samples with good separability and aerial images to be classified, run the classification with default settings, and output the classification results.
[0046] S4. Perform clustering on the initial classification results, merge classification patches with poor spatial continuity, and generate an optimized classification map.
[0047] Specifically, step S4 includes: S41, merging classification patches using a clustering tool; S42, statistically analyzing the number of pixels and standard deviation of each land cover category after merging, and generating a classification statistics report. In step S41, the clustering tool merges small patches from adjacent similar classification regions.
[0048] In this embodiment, the classified aerial imagery is loaded into ENVI 5.3, and a clustering tool is applied to merge small patches with poor spatial continuity and similar neighboring classification regions. Then, a data statistical model is used to calculate the number of pixels, minimum, maximum, and average values, as well as the standard deviation of each band in each category. A classification overlay tool is then used to overlay the various categories of the classification results onto an RGB color composite image or grayscale image, thereby generating an RGB image. Finally, the classification results are converted into vector data, and local classification results are modified.
[0049] S5. Verify the accuracy of the optimized classification map through manual visual interpretation. If the difference between the classification result and the manual interpretation exceeds the threshold, readjust the training samples and iterate through steps S3 to S4 until the accuracy requirements are met.
[0050] Specifically, in step S5, manual visual interpretation is performed using quadrats of a preset size, with 3 to 5 quadrats per land cover type. Each quadrat is a square area with side lengths matching the spatial distribution characteristics of the land cover type.
[0051] In this embodiment, the confusion matrix is selected in the classification to evaluate the accuracy of the merged result. Multiple indices such as overall accuracy, Kappa coefficient, and misclassification error will be displayed. When the overall accuracy and Kappa coefficient are large, it can be regarded as a good classification effect, which can be regarded as the final supervised classification interpretation result.
[0052] Secondly, for each land cover type in the aerial image, select 3-5 10*10m quadrats for manual visual interpretation. Compare the differences between the manual visual interpretation results and the supervised classification results. If the difference between the two is large, adjust the selection of training samples and reclassify. If the difference between the two is very small, save it as the final result to ensure that the final accuracy of the aerial image interpretation results is maximized.
[0053] This invention also provides a storage medium storing a computer program. When executed by a processor, the computer program implements some or all of the steps in the various embodiments of the land-sea integrated coastline classification method provided by this invention. The storage medium may be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.
[0054] Those skilled in the art will clearly understand that the techniques in the embodiments of the present invention can be implemented using software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solutions in the embodiments of the present invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or certain parts of the embodiments of the present invention.
[0055] The embodiments of the present invention described above do not constitute a limitation on the scope of protection of the present invention.
Claims
1. A coastline classification method based on land-sea integration, characterized in that, Includes the following steps: S1. Aerial photography of the target coastal area is carried out by an UAV equipped with an optical camera to obtain remote sensing images of a preset range in the direction of the vertical coastline of the sea and land, and simultaneously record positioning and attitude data. S2. Perform aerial triangulation on the aerial images, import the spatial coordinates of pre-deployed image control points for point optimization, and generate a two-dimensional orthophoto. S3. Based on orthophotos, automatically interpret coastal land features using supervised classification methods, specifically including: S31. In orthophotos, select training samples of various land cover types in a balanced manner, calculate the Jeffries-Matusta distance and transform divergence between samples, and filter the sample set with a separation value greater than 1.
8. S32. Classify the orthophoto using the maximum likelihood method and output the initial classification results; S4. Perform clustering on the initial classification results, merge classification patches with poor spatial continuity, and generate an optimized classification map; S5. Verify the accuracy of the optimized classification map through manual visual interpretation. If the difference between the classification result and the manual interpretation exceeds the threshold, readjust the training samples and iterate through steps S3 to S4 until the accuracy requirements are met.
2. The coastline classification method based on land-sea integration according to claim 1, characterized in that, Step S1 includes: S11. Evenly deploy image control points at preset intervals in the target coastal zone area, and measure and obtain the spatial coordinates of each image control point through real-time dynamic positioning technology of the global navigation satellite system. S12. Set the UAV's flight altitude and heading / lateral overlap, and generate an aerial survey line file covering the target area; S13. Perform aerial photography missions during periods of non-direct sunlight and check the integrity of the aerial photography data.
3. The coastline classification method based on land-sea integration according to claim 2, characterized in that, In step S11, the control point markers have significant spectral differences from the surrounding ground features.
4. The coastline classification method based on land-sea integration according to claim 1, characterized in that, In step S31, the selection of training samples satisfies the following conditions: S311. The sample size for each land cover type is evenly distributed; S312. The sample area avoids the extreme spectral response regions of the coastal transition zone.
5. The coastline classification method based on land-sea integration according to claim 1, characterized in that, Step S4 includes: S41. Merge classified patches using clustering tools; S42. Calculate the number of pixels and standard deviation of each object category after merging, and generate a classification statistical report.
6. The coastline classification method based on land-sea integration according to claim 5, characterized in that, In step S41, the clustering tool merges small patches from neighboring similar classification regions.
7. The coastline classification method based on land-sea integration according to claim 1, characterized in that, In step S5, manual visual interpretation is performed using quadrats of a preset size, with 3 to 5 quadrats per land cover type.
8. The coastline classification method based on land-sea integration according to claim 7, characterized in that, The sample plot is a square area with a side length that matches the spatial distribution characteristics of land cover types.
9. The coastline classification method based on land-sea integration according to claim 1, characterized in that, In step S32, the maximum likelihood method is used to perform classification with default parameters.
10. The coastline classification method based on land-sea integration according to claim 1, characterized in that, In step S2, the aerial triangulation optimization process includes iterative correction of the control point residuals after the puncture point.