An island reef change detection method based on remote sensing images

Abstract

The present invention relates to a method for detecting reef changes based on remote sensing images, belonging to the technical field of image processing, and solves the problems of inaccurate and large workload in existing reef change detection. The method includes: collecting remote sensing images of the same area at different time phases, performing adaptive superpixel segmentation on the preprocessed remote sensing images, and clustering to obtain multiple clustering regions; extracting feature points according to the number of superpixels in each clustering region of the reference remote sensing image and the remote sensing image to be detected; matching the feature points to obtain matching point pairs, and calculating a homography change matrix based on the matching point pairs; using the homography change matrix to infer the new coordinates of the feature points in the reference remote sensing image in the matching point pairs, and comparing them with the coordinates of the feature points in the matching remote sensing image to be detected to obtain the reef change detection result. Accurate and efficient reef change detection is achieved.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular, to a method for detecting reef changes based on remote sensing images. Background Art

[0002] With the rapid development of earth observation technology, it has become easier and easier to obtain a large number of remote sensing images. Change detection in remote sensing images is a process of identifying scene changes from a pair of images taken at different times in the same geographical area. However, the paired images are usually captured under different conditions, such as different angles, lighting, and even in different seasons, resulting in diverse and unknown changes in the scene.

[0003] Currently, various machine learning models are usually used to classify image images to obtain results of changed or unchanged, including support vector machine (SVM), random forest, decision tree, Markov random field (MRF), and conditional random field (CRF).

[0004] Using machine learning models requires manual annotation of data to create a data set, network model training, which takes time and effort. Moreover, machine learning models are black box models, requiring a large amount of sample data for training and being very sensitive to the quality of training samples. When the model performs poorly, it is difficult to locate the problem due to the unclear internal structure and decision logic of the model. Summary of the Invention

[0005] In view of the above analysis, an embodiment of the present invention aims to provide a method for detecting reef changes based on remote sensing images to solve the problems of inaccurate existing reef change detection and large workload.

[0006] An embodiment of the present invention provides a method for detecting reef changes based on remote sensing images, including the following steps:

[0007] Collect remote sensing images of the same area at different times. After adaptive superpixel segmentation of the preprocessed remote sensing images, multiple clustering regions are obtained by clustering;

[0008] According to the number of superpixels in each clustering region of the reference remote sensing image and the remote sensing image to be detected, feature points are respectively extracted; the feature points are matched to obtain matching point pairs, and a homography change matrix is calculated according to the matching point pairs;

[0009] The new coordinates of the feature points in the reference remote sensing image in the matching point pairs are inferred by using the homography change matrix, and compared with the coordinates of the feature points of the matching remote sensing image to be detected to obtain the reef change detection result.

[0010] Based on a further improvement of the above method, the adaptive superpixel segmentation of the preprocessed remote sensing image includes:

[0011] Calculate the gray values of each pixel in the preprocessed remote sensing image, and calculate the number of superpixels and the compactness according to the gray values of each pixel; obtain the initial seed points according to the number of superpixels and the gradient in a uniform distribution manner;

[0012] Calculate the distance metric values between each pixel and each initial seed point according to the HSV color space values and coordinate values of each pixel, and assign each pixel in the remote sensing image to the seed point with the minimum distance metric value to form initial superpixels; after iteratively updating the seed points in the initial superpixels, form new superpixels until the seed points no longer change, and obtain the segmented superpixels.

[0013] Based on the further improvement of the above method, calculate the number of superpixels according to the gray values of each pixel through the following formula:

[0014]

[0015] where Seg represents the number of superpixels, l max and l min represent the maximum gray value and the minimum gray value in the gray values of each pixel respectively, and h and w represent the height and width of the preprocessed remote sensing image respectively.

[0016] Based on the further improvement of the above method, calculate the compactness according to the gray values of each pixel through the following formula:

[0017]

[0018] where C represents the compactness, p(k) represents the probability that the gray level k appears in the remote sensing image, represents rounding up.

[0019] Based on the further improvement of the above method, obtain the initial seed points according to the number of superpixels and the gradient in a uniform distribution manner, including:

[0020] Take the square root of the ratio of the number of pixels in the remote sensing image to the number of superpixels to obtain the distance between adjacent seed points;

[0021] Obtain the coordinates of multiple first seed points according to the distance between adjacent seed points;

[0022] In the n×n neighborhood centered on each first seed point, take the pixel point with the minimum gradient as the initial seed point.

[0023] Based on the further improvement of the above method, calculate the distance metric values between each pixel and each initial seed point according to the HSV color space values and coordinate values of each pixel, and the formula is expressed as follows:

[0024]

[0025] Among them, SD j,i represents the distance metric value between the j-th pixel point and the i-th seed point, represents the HSV color distance between the j-th pixel point and the i-th seed point, represents the spatial distance between the j-th pixel point and the i-th seed point; (H j , S j , V j ) and (H i , S i , V i ) respectively represent the HSV color space values of the j-th pixel point and the i-th seed point; (x j , y j ) and (x i , y i ) respectively represent the coordinate values of the j-th pixel point and the i-th seed point; C represents the compactness; D represents the distance between adjacent seed points.

[0026] Based on the further improvement of the above method, according to the number of superpixels in each clustering region of the reference remote sensing image and the remote sensing image to be detected, feature points are respectively extracted, including:

[0027] Based on the FAST algorithm, for each clustering region in turn, taking 20% of the gray value of the determined central pixel point as the initial threshold, the initial feature points of this clustering region are extracted. If the number of initial feature points is less than the number of superpixels in this clustering region, the initial threshold is iteratively reduced according to a preset ratio, and new feature points are extracted until the number of new feature points is not less than the number of superpixels in this clustering region, and the feature points of this clustering region are extracted.

[0028] Based on the further improvement of the above method, matching point pairs are obtained by matching the feature points, including:

[0029] Based on the HSV color space value, obtain the BRIEF descriptor of each feature point;

[0030] According to the BRIEF descriptor of each feature point, calculate the first Hamming distance between the feature points of each remote sensing image to be detected and the feature points of each reference remote sensing image respectively. If the shortest first Hamming distance is greater than the distance threshold, the first feature point pairs matching the feature points of each remote sensing image to be detected are obtained according to the shortest first Hamming distance; calculate the second Hamming distance between the feature points of each reference remote sensing image and the feature points of each remote sensing image to be detected respectively. If the shortest second Hamming distance is greater than the distance threshold, the second feature point pairs matching the feature points of each reference remote sensing image are obtained according to the shortest second Hamming distance;

[0031] Retain the feature point pairs in which both feature points in the first feature point pair and the second feature point pair are the same to obtain matching point pairs.

[0032] Based on a further improvement of the above method, before calculating the first Hamming distance or the second Hamming distance, identify whether the number of positions with the same numerical values in the BRIEF descriptors of the two feature points exceeds the minimum matching threshold. If it exceeds, calculate the Hamming distance according to the BRIEF descriptors of the two feature points; otherwise, set the Hamming distance corresponding to the two feature points to a preset maximum value.

[0033] Based on a further improvement of the above method, according to the matching point pairs, calculate the homography transformation matrix by the least squares method using the following formula:

[0034]

[0035] where, (x, y) represents the feature point of the reference remote sensing image in the matching point pair, (x', y') represents the feature point of the remote sensing image to be detected in the matching point pair, and (h 11 , h 12 , h 13 , h 21 , h 22 , h 23 , h 31 , h 32 ) represents the element to be solved in the homography transformation matrix.

[0036] Compared with the prior art, the present invention can at least achieve one of the following beneficial effects:

[0037] 1. By adaptively setting the number and tightness of superpixels, more robustly process different lighting conditions, noise levels, and image details, and generate more uniform superpixels; at the same time, consider the HSV color space and distance space to improve the distinguishability of island regions and the quality of the segmentation results; finally, perform clustering based on the segmented image, which not only reduces the amount of data to be processed and improves the clustering efficiency, but also makes it easier for the clustering algorithm to identify and distinguish different regions because the pixels within the superpixels have similar characteristics, thus improving the accuracy of the clustering results.

[0038] 2. Based on the superpixel segmentation and clustering results, adaptively adjust the feature point extraction threshold, automatically find the best balance point, ensure the accurate and robust detection of feature points; obtain the descriptors of feature points based on the HSV color space, use the position information of the descriptors to improve the matching efficiency, and use the cross-matching method to improve the accuracy of matching point pairs, thereby obtaining an accurate homography transformation matrix and further improving the accuracy and reliability of island reef change detection.

[0039] 3. There is no need to spend a lot of time and energy on manual annotation of data. By using geometric ideas to achieve reef change detection, the entire solution has strong interpretability and is easy to operate.

[0040] In the present invention, the above technical solutions can also be combined with each other to achieve more preferred combination solutions. Other features and advantages of the present invention will be described in the subsequent specification. Moreover, some advantages can be made obvious from the specification or understood by implementing the present invention. The objectives and other advantages of the present invention can be realized and obtained from the content specifically pointed out in the specification and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The drawings are only for the purpose of showing specific embodiments and are not considered as a limitation to the present invention. Throughout the drawings, the same reference signs represent the same components.

[0042] Figure 1 It is a flowchart of a reef change detection method based on remote sensing images in an embodiment of the present invention;

[0043] Figure 2 It is an example diagram of the preprocessed remote sensing image in an embodiment of the present invention;

[0044] Figure 3 In an embodiment of the present invention Figure 2 It is an example diagram obtained after adaptive superpixel segmentation;

[0045] Figure 4 In an embodiment of the present invention Figure 3 It is an example diagram obtained after superpixel aggregation;

[0046] Figure 5 In an embodiment of the present invention Figure 3 It is an example diagram obtained after clustering processing. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0047] The following will specifically describe the preferred embodiments of the present invention in conjunction with the drawings. The drawings form a part of this application and are used together with the embodiments of the present invention to explain the principles of the present invention, rather than to limit the scope of the present invention.

[0048] A specific embodiment of the present invention discloses a reef change detection method based on remote sensing images, as Figure 1 shown, including the following steps:

[0049] S1. Collect remote sensing images of the same area at different times. After performing adaptive superpixel segmentation on the preprocessed remote sensing images, perform clustering to obtain multiple clustering regions.

[0050] It should be noted that in this embodiment, satellite remote sensing technology is used to periodically photograph the same area to obtain remote sensing images of the same area at different time phases. Usually, the width of one scene of the image is 30×30 kilometers.

[0051] Since the remote sensing image parameters are different when different satellites take pictures, and the shooting effects are different, preprocessing is performed on the collected remote sensing images, such as orthorectification, image registration, and image fusion. Finally, according to the coordinate information of the reef to be detected, the preprocessed remote sensing images are cropped to obtain remote sensing images of the same reef area at different time phases with consistent sizes, and the remote sensing images are default in the RGB color space.

[0052] Furthermore, in order to accurately outline the reef contour, the preprocessed remote sensing images are subjected to adaptive superpixel segmentation and then clustered again to obtain multiple clustering regions. Among them, in order to better divide the superpixel segmentation regions, the number of superpixels and the compactness are adaptively set. The compactness is used to represent the similarity or proximity between the pixel points within the superpixel.

[0053] Specifically, performing adaptive superpixel segmentation on the preprocessed remote sensing images includes:

[0054] ① Calculating the gray value of each pixel point in the remote sensing image in the RGB color space through the following formula:

[0055] l j =0.229R j +0.587G j +0.114B j Formula (1)

[0056] Wherein, l j represents the gray value of the jth pixel point, and (R j , G j , B j ) represents the RGB color space value of the jth pixel point.

[0057] Calculating the number of superpixels through the following formula according to the gray values of each pixel point:

[0058]

[0059] Wherein, Seg represents the number of superpixels, l max and l min respectively represent the maximum gray value and the minimum gray value among the gray values of each pixel point, and h and w respectively represent the height and width of the preprocessed remote sensing image.

[0060] Traverse the gray values of each pixel in the remote sensing image, count the number of occurrences of each gray level (from 0 to 255), and then divide by the number of pixels N in the remote sensing image to obtain the probability of each gray level appearing in the remote sensing image. Thus, the compactness is calculated through the following formula:

[0061]

[0062] Among them, C represents the compactness, and p(k) represents the probability of the gray level k appearing in the remote sensing image. denotes rounding up.

[0063] Furthermore, according to the way of uniform distribution, the initial seed points are obtained based on the number of superpixels and gradients, including:

[0064] According to the following formula, take the square root of the ratio of the number of pixels N in the remote sensing image to the number of superpixels Seg to obtain the distance D between adjacent seed points:

[0065]

[0066] Obtain the coordinates of multiple first seed points according to the distance between adjacent seed points; within the n×n neighborhood centered on each first seed point, take the pixel point with the minimum gradient as the initial seed point. Preferably, n = 3.

[0067] ② Calculate the distance metric value between each pixel point and each initial seed point according to the HSV color space value and coordinate value of each pixel point, and assign each pixel point in the remote sensing image to the seed point when the distance metric value is the smallest to form an initial superpixel; after iteratively updating the seed points in the initial superpixel, form a new superpixel until the seed points no longer change, and obtain the segmented superpixels.

[0068] It should be noted that the HSV color space is considered in this embodiment, which can better distinguish different colors and has higher discrimination for seawater, vegetation, sand, etc.

[0069] Convert the remote sensing image from the RGB color space to the HSV color space to obtain the HSV color space values of each pixel point: hue (Hue, abbreviated as H), saturation (Saturation, abbreviated as S), and value (Value, abbreviated as V).

[0070] According to the HSV color space value and coordinate value of each pixel point, calculate the distance metric value between each pixel point and each initial seed point through the following formula:

[0071]

[0072] Among them, SD j,i represents the distance metric value between the j-th pixel point and the i-th seed point. represents the HSV color distance between the j-th pixel point and the i-th seed point, and represents the spatial distance between the j-th pixel point and the i-th seed point; (H j , S j , V j ) and (H i , S i , V i ) respectively represent the HSV color space values of the j-th pixel point and the i-th seed point; (x j , y j ) and (x i , y i ) respectively represent the coordinate values of the j-th pixel point and the i-th seed point.

[0073] Assign each pixel point in the remote sensing image to the seed point with the minimum distance metric value to form initial superpixels; calculate the centroid or average point of each initial superpixel as the new seed point, then calculate the distance metric values between each pixel point and each new seed point, and assign each pixel point to the seed point with the minimum distance metric value to form new superpixels. Iterate this update process until the seed points no longer change, and the segmented superpixels are obtained.

[0074] Finally, perform clustering processing on the remote sensing image after superpixel segmentation to obtain multiple clustering regions, improving the recognition accuracy of reef areas.

[0075] Specifically, when performing clustering processing, set the number of categories to 2 to C for multiple clustering respectively, and select the best result among them through visual interpretation to obtain multiple clustering regions.

[0076] Exemplarily, the preprocessed remote sensing image is as shown in Figure 2 , after adaptive superpixel segmentation is as shown in Figure 3 . After aggregating each superpixel, Figure 4 is obtained. Based on the result of segmentation in Figure 3 , after clustering processing, Figure 5 is obtained. It can be seen from Figure 4 and Figure 5 that after performing clustering processing on the remote sensing image after superpixel segmentation again, the outlines of each clustering region are clear and the clustering effect is relatively good.

[0077] Preferably, according to the size and resolution of the remote sensing image, obtain the area size represented by one pixel point, count the number of pixel points included in the reef area in the clustering region, and thus calculate the area of the reef area. Record the areas of the reef areas in the remote sensing images of different time phases, and visually display the change situation of the reef area through a chart.

[0078] Compared with the prior art, in this embodiment, by adaptively setting the number of superpixels and the tightness, it can process different illumination conditions, noise levels, and image details more robustly, generating more uniform superpixels. At the same time, considering the HSV color space and the distance space, it improves the distinguishability of island regions and the quality of the segmentation result. Finally, clustering is performed based on the segmented image, which not only reduces the amount of data to be processed and improves the clustering efficiency, but also makes it easier for the clustering algorithm to identify and distinguish different regions because the pixels within the superpixels have similar characteristics, thus improving the accuracy of the clustering result.

[0079] S2. According to the number of superpixels in each clustering region of the reference remote sensing image and the remote sensing image to be detected, feature points are respectively extracted; the feature points are matched to obtain matching point pairs, and the homography transformation matrix is calculated based on the matching point pairs.

[0080] It should be noted that the prior art usually uses the FAST algorithm (Features from Accelerated Segment Test, an algorithm for quickly detecting corner points or significant points) to extract feature points. The size of the threshold in the FAST algorithm affects the number of detected feature points. A larger threshold will reduce the number of detected feature points, while a smaller threshold will increase the number of feature points. Currently, it is usually through experiments to adjust the threshold to find the best balance point.

[0081] This step is based on the clustering result of step S1, and feature points are sequentially extracted according to the clustering regions. Moreover, when extracting feature points, the FAST algorithm (Features from Accelerated Segment Test, an algorithm for quickly detecting corner points or significant points) is improved, and the threshold is adaptively adjusted in combination with the superpixels in step S1 to ensure the accurate and robust detection of feature points.

[0082] In this embodiment, two remote sensing images for detecting island reef changes are determined from remote sensing images of the same area at different times, where one is used as the reference remote sensing image and the other is used as the remote sensing image to be detected. That is to say, finally, the changes of the remote sensing image to be detected relative to the reference remote sensing image need to be detected.

[0083] It should be noted that extracting feature points respectively according to the number of superpixels in each clustering region of the reference remote sensing image and the remote sensing image to be detected includes:

[0084] Based on the FAST algorithm, for each clustering region in turn, 20% of the gray value of the determined central pixel is used as the initial threshold to extract the initial feature points of the clustering region. If the number of initial feature points is less than the number of superpixels in the clustering region, the initial threshold is iteratively reduced according to a preset ratio, and new feature points are extracted until the number of new feature points is not less than the number of superpixels in the clustering region, and the feature points of the clustering region are extracted.

[0085] Specifically, a circular neighborhood of a fixed size (for example, a Bresenham circle with a radius of 3 pixels) is used as the detection window, and the detection window includes the central pixel and a circle of pixels around it. The feature points are extracted from each clustering region through the following steps:

[0086] Take out a pixel point in turn as the central pixel point, and use 20% of the gray value of the central pixel point as the initial threshold; perform the following detection on the central pixel point: compare the gray value of the central pixel point with the gray values of the pixels in a circle around it in the detection window respectively. If there are t pixel points whose gray values are all greater than the gray value of the central pixel point plus the initial threshold, or all less than the gray value of the central pixel point minus the initial threshold, and t is at least half of the number of pixels in a circle around the detection window, then the central pixel point is an initial feature point; when all pixel points in the clustering region have been taken out and detected, all the initial feature points of the clustering region are obtained.

[0087] Furthermore, if the number of initial feature points is greater than or equal to the number of superpixels in the clustering region, the initial feature points are used as the feature points finally extracted from the clustering region. Otherwise, take out a pixel point in turn as the central pixel point again, and use the value reduced by a preset ratio based on 20% of the gray value of the central pixel point as the new initial threshold, and extract new feature points again until the number of new feature points is not less than the number of superpixels in the clustering region.

[0088] In this embodiment, by comparing the number of feature points extracted from the same clustering region with the number of superpixels, the threshold is adaptively reduced, and the number of points with the absolute value of the difference in gray values greater than the threshold is increased, so as to extract more feature points and ensure that there are at least as many feature points as the number of superpixels in each clustering region.

[0089] Next, the feature points of the reference remote sensing image and the remote sensing image to be detected are matched to obtain matching point pairs, including:

[0090] ① Based on the HSV color space value, obtain the BRIEF descriptor of each feature point.

[0091] It should be noted that the BRIEF (Binary Robust Independent Elementary Features) descriptor is a binary descriptor for feature points, which is efficient and robust.

[0092] In this step, a circular area is constructed with the feature point as the center according to a preset radius as the search area. According to the preset number of point pairs, multiple point pairs are randomly selected within the search area, denoted as Q1(q a ,q b ), Q2(q a ,q b ), …, Q r (q a ,q b ), where (q a ,q b ) represents the two pixel points in each point pair, and the abscissa value of the first pixel point q a is greater than the abscissa value of the second pixel point q b .

[0093] Based on the HSV color space values, the HSV fusion values of each pixel point in each point pair are calculated through the following formula:

[0094]

[0095] where SR r,m represents the HSV fusion value of the m-th pixel point in the r-th point pair, and (H r,m , S r,m , V r,m ) represents the HSV color space value of the m-th pixel point in the r-th point pair, and m = 1, 2.

[0096] Furthermore, compare the HSV fusion values of each pixel point in each point pair. If the HSV fusion value of the first pixel point is greater than that of the second pixel point, the comparison result is 1; otherwise, it is 0. Concatenate the comparison results of each point pair to obtain a binary string, which is the BRIEF descriptor.

[0097] ② According to the BRIEF descriptors of each feature point, calculate the first Hamming distance between the feature points of each remotely sensed image to be detected and the feature points of each reference remotely sensed image respectively. If the shortest first Hamming distance is greater than the distance threshold, obtain the first feature point pairs that match the feature points of each remotely sensed image to be detected according to the shortest first Hamming distance; calculate the second Hamming distance between the feature points of each reference remotely sensed image and the feature points of each remotely sensed image to be detected respectively. If the shortest second Hamming distance is greater than the distance threshold, obtain the second feature point pairs that match the feature points of each reference remotely sensed image according to the shortest second Hamming distance.

[0098] It should be noted that in this step, considering the different numbers of feature points extracted from the reference remotely sensed image and the remotely sensed image to be detected, a cross-matching method is adopted. The feature points of the remotely sensed image to be detected and the feature points of the reference remotely sensed image are respectively used as the first feature points in the feature point pairs to determine the second feature points that match them, so as to obtain two sets of feature point pairs: the first feature point pairs and the second feature point pairs.

[0099] ③ Retain the feature point pairs in which the two feature points in the first feature point pairs and the second feature point pairs are the same to obtain the matching point pairs.

[0100] It should be noted that in this step, the co-existing feature point pairs are found from the two sets of feature point pairs as the matching point pairs to improve the accuracy of the matching point pairs.

[0101] Exemplarily, in the first feature point pair, there is (P a , P b ), and in the second feature point pair, there is (P b , P a ), then it is considered that the feature point P a and the feature point P b are a correct match to obtain the matching point pair. Otherwise, it is considered an incorrect match and removed.

[0102] Preferably, in order to improve the matching efficiency, before calculating the first Hamming distance or the second Hamming distance, identify whether the number of values that are the same at the same positions in the BRIEF descriptors of the two feature points exceeds the minimum matching threshold. If it exceeds, calculate the Hamming distance according to the BRIEF descriptors of the two feature points. Otherwise, set the Hamming distance corresponding to the two feature points to the preset maximum value.

[0103] Exemplarily, the descriptor length is 6, and the minimum matching threshold is 2 / 3 of the descriptor length, that is, 4; the descriptor of the feature point P a is 110101, the descriptor of the feature point P b is 110111, the descriptor of the feature point P c is 100001, and the descriptor of the feature point P aand feature point P b If the number of values that are the same at the same position is 5, which exceeds the minimum matching threshold, then calculate feature point P according to the descriptor a with feature point P b calculate the Hamming distance; for feature point P a and feature point P c If the number of values that are the same at the same position is 2, which is less than the minimum matching threshold, do not calculate the Hamming distance between these two feature points and directly set it to a preset maximum value

[0104] Randomly sample at least four groups from the obtained matching point pairs, and use the following formula to calculate the homography transformation matrix by the least squares method, that is, obtain the mapping matrix from the reference remote sensing image to the to-be-detected remote sensing image

[0105]

[0106] where, (x, y) represents the feature point of the reference remote sensing image in the matching point pair, (x', y') represents the feature point of the to-be-detected remote sensing image in the matching point pair, and (h 11 , h 12 , h 13 , h 21 , h 22 , h 23 , h 31 , h 32 ) represents the element to be solved in the homography transformation matrix

[0107] Preferably, use the RANSAC (Random Sample Consensus) algorithm to obtain the optimal homography transformation matrix from the homography transformation matrix calculated by iterative sampling

[0108] S3. Use the homography transformation matrix to infer the new coordinates of the feature points in the reference remote sensing image in the matching point pair, and compare them with the coordinates of the feature points of the matching to-be-detected remote sensing image to obtain the island reef change detection result

[0109] It should be noted that input the coordinates of the feature points in the reference remote sensing image in the matching point pair into formula (7) to obtain the new coordinates after change, and compare them with the coordinates of the feature points of the matching to-be-detected remote sensing image. If the coordinate difference between the two feature points being compared exceeds the error threshold, it indicates that a change has occurred at this feature point. Record the matching point pair and mark it in the two preprocessed remote sensing images to obtain the island reef change detection result. Exemplarily, the result of whether the matching point pair has changed is marked in different colors in the remote sensing image

[0110] Compared with the prior art, a reef change detection method based on remote sensing images provided by this embodiment can more robustly process different illumination conditions, noise levels, and image details by adaptively setting the number of superpixels and compactness, and generate more uniform superpixels. At the same time, considering the HSV color space and distance space, it improves the distinguishability of island regions and the quality of the segmentation results. Finally, clustering is performed based on the segmented images, which not only reduces the amount of data to be processed and improves the clustering efficiency, but also makes it easier for the clustering algorithm to identify and distinguish different regions because the pixels within the superpixels have similar characteristics, thus improving the accuracy of the clustering results. Based on the superpixel segmentation and clustering results, the feature point extraction threshold is adaptively adjusted to automatically find the best balance point, ensuring that the detection of feature points is both accurate and robust. The descriptor of the feature points is obtained based on the HSV color space, and the matching efficiency is improved by using the position information of the descriptor. The accuracy of the matching point pairs is improved by using the cross-matching method, so as to obtain an accurate homography transformation matrix, further improving the accuracy and reliability of reef change detection. There is no need to spend a lot of time and effort on manual annotation of data, and the reef change detection is realized by using geometric ideas. The whole scheme is highly interpretable and easy to operate.

[0111] Those skilled in the art can understand that all or part of the processes of implementing the methods of the above embodiments can be completed by instructing relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium. Among them, the computer-readable storage medium is a disk, an optical disc, a read-only memory, or a random access memory, etc.

[0112] As mentioned above, the above are only the preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention.

Claims

1. A method for detecting island and reef changes based on remote sensing images, characterized in that: The following steps are involved: Collect remote sensing images of the same area at different times, perform adaptive superpixel segmentation on the pre-processed remote sensing images, and cluster them to obtain multiple clustering areas; According to the number of superpixels in each clustering area of ​​the reference remote sensing image and the remote sensing image to be detected, feature points are extracted respectively; Matching the feature points to obtain matching point pairs, and calculating a homography change matrix according to the matching point pairs; The new coordinates of the feature points in the reference remote sensing image in the matching point pair are inferred by using the homography change matrix, and compared with the coordinates of the feature points in the matched remote sensing image to be detected to obtain the island reef change detection result; The step of adaptively segmenting the pre-processed remote sensing image into superpixels comprises: Calculate the gray value of each pixel in the remote sensing image after preprocessing, and calculate the number and density of superpixels according to the gray value of each pixel; obtain initial seed points according to the number of superpixels and gradients in a uniform distribution manner; calculate the distance measurement value between each pixel and each initial seed point according to the HSV color space value and coordinate value of each pixel, and assign each pixel in the remote sensing image to the seed point with the smallest distance measurement value to form an initial superpixel; after iteratively updating the seed point in the initial superpixel, a new superpixel is formed until the seed point no longer changes, and the segmented superpixel is obtained; The extracting of feature points according to the number of superpixels in each clustering area of ​​the reference remote sensing image and the remote sensing image to be detected comprises: Based on the FAST algorithm, for each cluster area in turn, 20% of the grayscale value of the determined central pixel is used as the initial threshold to extract the initial feature points of the cluster area. If the number of initial feature points is less than the number of superpixels in the cluster area, the initial threshold is iteratively reduced according to a preset ratio, and new feature points are extracted until the number of new feature points is not less than the number of superpixels in the cluster area, and the feature points of the cluster area are extracted.

2. The island and reef change detection method based on remote sensing images according to claim 1 is characterized in that: The number of superpixels is calculated according to the grayscale value of each pixel using the following formula: Among them, Seg represents the number of superpixels, l max and l min They represent the maximum grayscale value and the minimum grayscale value of each pixel respectively, and h and w represent the height and width of the preprocessed remote sensing image respectively.

3. The island and reef change detection method based on remote sensing images according to claim 1 is characterized in that: The compactness is calculated according to the gray value of each pixel using the following formula: Among them, C represents compactness, p(k) represents the probability of gray level k appearing in the remote sensing image, Indicates rounding up.

4. The island and reef change detection method based on remote sensing images according to claim 2 is characterized in that: The method of obtaining the initial seed point according to the number of superpixels and the gradient in a uniform distribution manner includes: The distance between adjacent seed points is obtained by taking the square root of the ratio of the number of pixels and the number of superpixels in the remote sensing image; Obtaining coordinates of multiple first seed points according to distances between adjacent seed points; In the n×n neighborhood centered on each first seed point, the pixel with the smallest gradient is taken as the initial seed point.

5. The island and reef change detection method based on remote sensing images according to claim 4 is characterized in that: The distance measurement value between each pixel point and each initial seed point is calculated according to the HSV color space value and coordinate value of each pixel point. The formula is as follows: Among them, SD j,i Represents the distance metric between the j-th pixel and the i-th seed point, Represents the HSV color distance between the jth pixel and the i-th seed point, represents the spatial distance between the jth pixel and the i-th seed point; (H j ,S j ,V j ) and (H i ,S i ,V i ) represent the HSV color space values ​​of the j-th pixel and the i-th seed point respectively; (x j ,y j ) and (x i ,y i ) represent the coordinate values ​​of the j-th pixel point and the i-th seed point respectively; C represents the compactness; and D represents the distance between adjacent seed points.

6. The island and reef change detection method based on remote sensing images according to claim 1 is characterized in that: The matching of the feature points to obtain matching point pairs includes: Based on the HSV color space value, obtain the BRIEF descriptor of each feature point; According to the BRIEF descriptor of each feature point, the first Hamming distance between the feature point of each remote sensing image to be detected and the feature point of each reference remote sensing image is calculated respectively. If the shortest first Hamming distance is greater than the distance threshold, the first feature point pair matching the feature point of each remote sensing image to be detected is obtained according to the shortest first Hamming distance; the second Hamming distance between the feature point of each reference remote sensing image and the feature point of each remote sensing image to be detected is calculated respectively. If the shortest second Hamming distance is greater than the distance threshold, the second feature point pair matching the feature point of each reference remote sensing image is obtained according to the shortest second Hamming distance; The feature point pairs in which the two feature points in the first feature point pair and the second feature point pair are the same are retained to obtain matching point pairs.

7. The island and reef change detection method based on remote sensing images according to claim 6 is characterized in that: Before calculating the first Hamming distance or the second Hamming distance, identify whether the number of identical values ​​at the same position in the BRIEF descriptors of the two feature points exceeds the minimum matching threshold. If so, calculate the Hamming distance based on the BRIEF descriptors of the two feature points. Otherwise, set the Hamming distance corresponding to the two feature points to the preset maximum value.

8. The method for detecting island and reef changes based on remote sensing images according to claim 1, characterized in that: According to the matching point pairs, the homography change matrix is ​​calculated using the least squares method using the following formula: Among them, (x, y) represents the feature point of the reference remote sensing image in the matching point pair, (x', y') represents the feature point of the remote sensing image to be detected in the matching point pair, and (h 11 ,h 12 ,h 13 ,h 21 ,h 22 ,h 23 ,h 31 ,h 32 ) represents the element to be solved in the homography transformation matrix.

Images