Object-oriented remote sensing image change detection method and related device

Abstract

The application discloses an object-oriented remote sensing image change detection method and related equipment, and the method comprises the following steps: obtaining two remote sensing images to form a target remote sensing image through band combination; performing segmentation on the target remote sensing image based on multi-scale segmentation to obtain a multi-scale segmentation result, and determining the best segmentation scale and the optimal object in the multi-scale segmentation result according to the overall goodness F value; extracting object spectral features, constructing a color co-occurrence matrix based on the object, extracting texture feature statistics of the multi-band image through the color co-occurrence matrix, and obtaining texture information containing inter-band correlation; performing feature contribution degree evaluation on the multi-feature difference image by using a random forest model, and screening the optimal feature combination by using a cumulative backward search method; and performing object-oriented random forest change detection based on the optimal feature combination to obtain a final remote sensing image change detection result. The application improves the object-level remote sensing image change detection precision.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to an object-oriented remote sensing image change detection method, system, terminal, and computer-readable storage medium. Background Technology

[0002] With social and economic development, the depth and breadth of human utilization of land resources have increased significantly, leading to a marked increase in the frequency of changes in land resources. Accurately acquiring information on changes in land resources is crucial for understanding the relationship between humans and nature and promoting sustainable socio-economic and ecological development. Remote sensing technology, with its ability to observe large areas in a short time, has rapidly become one of the important means of effectively extracting change information. As remote sensing technology continues to develop, the spatial resolution of remote sensing images is gradually improving, highlighting geometric and textural information. While providing rich details for change detection, this also exacerbates the phenomena of "same object, different spectra" and "same spectra, different objects," bringing new challenges to change detection in remote sensing images.

[0003] Remote sensing image change detection methods can be categorized into pixel-based and object-based methods based on their operational units. Traditional pixel-based change detection methods use individual pixels as the analysis unit, easily ignoring the contextual information of the image and leading to salt-and-pepper noise. Object-oriented techniques group pixels with similar characteristics into spatially continuous sub-regions, effectively avoiding salt-and-pepper noise and integrating spatial neighborhood features to further improve change detection accuracy. Objects represent a higher level of image description; relying solely on single spectral information easily overlooks object details and overall features, leading to missed or false detections. Therefore, integrating multi-feature information has become an inevitable trend. Texture is one of the effective tools for analyzing the overall features of objects, and accurately characterizing texture features is key to improving the accuracy of remote sensing image change detection. Commonly used texture feature extraction techniques include Local Binary Patterns (LBP) operators and Gabor filter gray-level co-occurrence matrix (GLCM). However, most of these texture features are extracted based on single-band information of the image, ignoring the correlation between different bands in multi-band images, resulting in insufficient description of image texture information.

[0004] Furthermore, change detection accuracy does not necessarily increase with the number of features; therefore, it is generally necessary to screen the numerous extracted features. Feature screening mainly includes unsupervised and supervised methods. Unsupervised methods typically determine the optimal feature combination using metrics such as JM distance. Supervised methods estimate the contribution of each feature through model training to determine the optimal feature combination. Supervised methods are widely studied because they are less susceptible to outliers. However, current feature contribution rankings often use the results of a single training iteration or a particular iteration of multiple training iterations as a benchmark. The ranking results often depend on the model training results, and the optimal number of features usually needs to be determined manually based on experience.

[0005] Therefore, existing technologies still need to be improved and developed. Summary of the Invention

[0006] The main objective of this invention is to provide an object-oriented remote sensing image change detection method, system, terminal, and computer-readable storage medium. It aims to solve the problems in the prior art where object-oriented change detection ignores the correlation between bands when extracting texture features, and the optimal number of features needs to be given manually based on experience, resulting in low change detection accuracy.

[0007] To achieve the above objectives, the present invention provides an object-oriented remote sensing image change detection method, which includes the following steps:

[0008] Acquire the first phase of remote sensing images and the second phase of remote sensing images, and combine the first phase of remote sensing images and the second phase of remote sensing images by band to form a target remote sensing image;

[0009] The target remote sensing image is segmented based on multi-scale segmentation to obtain multi-scale segmentation results. The optimal segmentation scale and the optimal object in the multi-scale segmentation results are determined based on the overall goodness-of-facts (F-value).

[0010] Extract the spectral features of the object, construct a color co-occurrence matrix based on the object, and extract the texture feature statistics of the multi-band image through the color co-occurrence matrix to obtain texture information containing inter-band correlation.

[0011] Based on the multi-feature difference image, the feature contribution is evaluated using a random forest model, and the optimal feature combination is obtained by cumulative backward search.

[0012] Based on the optimal feature combination, object-oriented random forest change detection is performed to obtain the final remote sensing image change detection result.

[0013] Optionally, the object-oriented remote sensing image change detection method, wherein acquiring a first-phase remote sensing image and a second-phase remote sensing image, and combining the first-phase remote sensing image and the second-phase remote sensing image by bands to form a target remote sensing image, specifically includes:

[0014] Acquiring the first phase of remote sensing images I (1) Second Phase Remote Sensing Image I (2) ,and:

[0015] I (1) ={I i (1) (x i y i ):(x i y i )∈Ω, i=1,...,n;

[0016] I (2) ={I i (2) (x i y i ):(x i y i )∈Ω, i=1,...,n;

[0017] Where i is the pixel index, (x i y i Let I be the grid point coordinates of pixel i in the image spatial domain. i (1) and I i (2) The first phase of remote sensing images I (1) Second Phase Remote Sensing Image I (2) The spectral metric vector of inner pixel i, I i (1) =(I is (1) : s = 1, ..., h), I i (2) =(I is (2) : s = 1, ..., h), where s is the band index, h is the number of bands, n is the total number of pixels, Ω is the image domain, Ω = {(x i y i ): i = 1, ..., n}, I is (1) and I is (2) The first phase of remote sensing images I (1) Second Phase Remote Sensing Image I (2) The spectral value of inner pixel i in the s band;

[0018] The first phase of remote sensing images I (1) Second Phase Remote Sensing Image I (2) A remote sensing image of the target is formed by combining bands;

[0019] I = {I i (x i y i ):(x i y i )∈Ω, i=1,...,n},I i =(I is : s = 1, ...,

[0020] 2h).

[0021] Optionally, the object-oriented remote sensing image change detection method, wherein the step of segmenting the target remote sensing image based on multi-scale segmentation to obtain multi-scale segmentation results, and determining the optimal segmentation scale and optimal object in the multi-scale segmentation results based on the overall goodness-of-facts (F-value), specifically includes:

[0022] Based on a multi-scale segmentation method, the target remote sensing image I is divided into a group of objects O. (c) ={O j (c) :j = 1, ..., m}, where c is the scale index, j is the object index, and m is the number of objects;

[0023] Specifically, the multi-scale segmentation method involves: initially dividing the target remote sensing image I based on graph theory to obtain over-segmented superpixel blocks, and then merging the over-segmented superpixel blocks based on a fractal network evolution algorithm to obtain the multi-scale segmentation result.

[0024] Determining the optimal segmentation scale based on OGF:

[0025]

[0026] Where OGF represents the overall goodness-of-fact (F-value), and MI... normal and LV normal These are the standardized results of the MI index and the spectral local variance LV, respectively, with α being the weighting parameter;

[0027] The optimal segmentation scale is selected as the segmentation scale corresponding to the maximum value of the overall goodness F, and the object under the optimal segmentation scale is the optimal object.

[0028] Optionally, the object-oriented remote sensing image change detection method, wherein extracting object spectral features, constructing a color co-occurrence matrix based on the object, and extracting texture feature statistics of multi-band images through the color co-occurrence matrix to obtain texture information containing inter-band correlations, specifically includes:

[0029] Extract the spectral features of the object, using the mean μ of the spectral features of all pixels within the optimal object. j and variance σ j Spectral characteristics of the optimal object;

[0030] The target remote sensing image I is divided into N gray levels: I si * ∈{1, ..., N}, I si * This represents the gray level of pixel i in the s band;

[0031] Based on the correlation between band s and band s', a color co-occurrence matrix (CCM) is constructed. ss’ (e, r, d, θ); where e and r are gray level indices, e, r ∈ {1, ..., N}, d is the spatial distance, and θ represents the direction;

[0032] CCM ss’ (e, r, d, θ) represents the gray levels I of pixel pair i and i' that are d apart in the θ direction. si * =e,I s’i’ * = The probability of r occurring is given by the following formula:

[0033]

[0034] Based on Color Co-occurrence Matrix (CCM) ss’ (e, r, d, θ) extracts the texture feature statistics of multi-band images to obtain texture information containing inter-band correlation.

[0035] Optionally, in the object-oriented remote sensing image change detection method, the texture feature statistics include contrast, uniformity, second moment of angle, and correlation.

[0036] Optionally, the object-oriented remote sensing image change detection method, wherein the step of evaluating feature contribution using a random forest model based on multi-feature difference images and selecting the optimal feature combination using a cumulative backward search method specifically includes:

[0037] After determining the changed and unchanged object samples, a preset proportion of samples are selected as training samples for model training, and the remaining samples are used as the test set to evaluate the model training accuracy.

[0038] The random forest model is trained T times. Based on the cumulative feature contribution value of the T training times, the features are re-sorted, and the features with the least importance are eliminated in turn to form a new feature combination. The new feature combination is then input into the random forest model for T training times until the number of features reaches the minimum and training stops.

[0039] Based on the fluctuation of test accuracy under multiple feature combinations, the feature combination that meets the preset requirement in terms of the number of features is selected as the optimal feature combination.

[0040] Optionally, the object-oriented remote sensing image change detection method, wherein the step of performing object-oriented random forest change detection based on the optimal feature combination to obtain the final remote sensing image change detection result specifically includes:

[0041] Based on the optimal feature combination, the feature vectors of the sample are cleaned, and only the vector information of the optimal feature combination is retained;

[0042] The classifier is reconstructed to obtain the classification result, which is the object-oriented remote sensing image change detection result.

[0043] Furthermore, to achieve the above objectives, the present invention also provides an object-oriented remote sensing image change detection system, wherein the object-oriented remote sensing image change detection system includes:

[0044] The image acquisition and combination module is used to acquire the first phase of remote sensing images and the second phase of remote sensing images, and to combine the first phase of remote sensing images and the second phase of remote sensing images by band to form a target remote sensing image.

[0045] The multi-scale segmentation module is used to segment the target remote sensing image based on multi-scale segmentation to obtain multi-scale segmentation results, and to determine the optimal segmentation scale and optimal object in the multi-scale segmentation results based on the overall goodness-of-fairness F value.

[0046] The texture feature extraction module is used to extract the spectral features of the object, construct a color co-occurrence matrix based on the object, and extract the texture feature statistics of multi-band images through the color co-occurrence matrix to obtain texture information containing inter-band correlation.

[0047] The feature selection module is used to evaluate the feature contribution based on multi-feature difference images using a random forest model and to select the optimal feature combination using a cumulative backward search method.

[0048] The image change detection module is used to perform object-oriented random forest change detection based on the optimal feature combination to obtain the final remote sensing image change detection result.

[0049] In addition, to achieve the above objectives, the present invention also provides a terminal, wherein the terminal includes: a memory, a processor, and an object-oriented remote sensing image change detection program stored in the memory and executable on the processor, wherein when the object-oriented remote sensing image change detection program is executed by the processor, it implements the steps of the object-oriented remote sensing image change detection method as described above.

[0050] Furthermore, to achieve the above objectives, the present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores an object-oriented remote sensing image change detection program, which, when executed by a processor, implements the steps of the object-oriented remote sensing image change detection method as described above.

[0051] In this invention, a first-phase remote sensing image and a second-phase remote sensing image are acquired. These two images are then combined by band analysis to form a target remote sensing image. The target remote sensing image is segmented using multi-scale segmentation to obtain multi-scale segmentation results. The optimal segmentation scale and optimal object in the multi-scale segmentation results are determined based on the overall goodness-of-facts (F-value). Spectral features of the object are extracted, and a color co-occurrence matrix is ​​constructed based on the object. Texture feature statistics of the multi-band image are extracted using the color co-occurrence matrix to obtain texture information containing inter-band correlations. Based on the multi-feature difference image, a random forest model is used to evaluate feature contribution, and a cumulative backward search method is used to select the optimal feature combination. Based on the optimal feature combination, object-oriented random forest change detection is performed to obtain the final remote sensing image change detection result. This invention uses the object as the basic operation unit. Based on the extraction of object spectral features, it further extracts multi-channel texture features of each object based on the color co-occurrence matrix, introduces inter-band correlations, and combines the cumulative backward search method to determine the optimal feature combination, thereby improving the accuracy of object remote sensing image change detection. Attached Figure Description

[0052] Figure 1 This is a flowchart of a preferred embodiment of the object-oriented remote sensing image change detection method of the present invention;

[0053] Figure 2 This is a flowchart of the entire remote sensing image detection process in a preferred embodiment of the object-oriented remote sensing image change detection method of the present invention.

[0054] Figure 3 This is a schematic diagram of the first phase of remote sensing imagery in a preferred embodiment of the object-oriented remote sensing imagery change detection method of the present invention.

[0055] Figure 4 This is a schematic diagram of the second phase of remote sensing imagery in a preferred embodiment of the object-oriented remote sensing imagery change detection method of the present invention;

[0056] Figure 5 This is a schematic diagram of the standard results in the change detection results;

[0057] Figure 6 This is a schematic diagram of the detection results using Comparison Algorithm 1;

[0058] Figure 7This is a schematic diagram of the detection results using Comparison Algorithm 2;

[0059] Figure 8 This is a schematic diagram of the detection results of the method of the present invention;

[0060] Figure 9 This is a schematic diagram of a preferred embodiment of the object-oriented remote sensing image change detection system of the present invention;

[0061] Figure 10 This is a schematic diagram of the operating environment of a preferred embodiment of the terminal of the present invention. Detailed Implementation

[0062] To make the objectives, technical solutions, and advantages of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0063] Current object-oriented change detection methods typically extract texture features based on single-band images, neglecting inter-band correlation and resulting in insufficient characterization of image texture features. This invention, building upon the extraction of object spectral features, further extracts multi-channel texture features for each object based on the color co-occurrence matrix, introducing inter-band correlation to improve change detection accuracy.

[0064] In feature selection, current methods often rank features based on the results of a single training iteration or a single iteration across multiple training iterations. The ranking often depends on the model's training outcome, and the optimal number of features is typically determined manually based on experience. This invention proposes a cumulative backward search method that ranks features based on the sum of their contributions over multiple iterations. This method ensures that, within a certain range of fluctuations, it automatically selects relatively stable and representative feature combinations.

[0065] The preferred embodiment of the object-oriented remote sensing image change detection method of the present invention, such as... Figure 1 and Figure 2 As shown, the object-oriented remote sensing image change detection method includes the following steps:

[0066] Step S10: Acquire the first phase remote sensing image and the second phase remote sensing image, and combine the first phase remote sensing image and the second phase remote sensing image by band to form a target remote sensing image.

[0067] Specifically, acquiring the first phase of remote sensing images I (1) Second Phase Remote Sensing Image I (2) (For example Figure 3 and Figure 4 (These represent Sentinel-2 images of the Shenzhen area in 2021 and 2022, respectively), and:

[0068] I (1) ={I i (1) (x i y i ):(x i y i )∈Ω, i=1,...,n;

[0069] I (2) ={I i (2) (x i y i ):(x i y i )∈Ω, i=1,...,n;

[0070] Where i is the pixel index, (x i y i Let I be the grid point coordinates of pixel i in the image spatial domain. i (1) and I i (2) The first phase of remote sensing images I (1) Second Phase Remote Sensing Image I (2) The spectral metric vector of inner pixel i, I i (1) =(I is (1) : s = 1, ..., h), I i (2) =(I is (2) : s = 1, ..., h), where s is the band index, h is the number of bands, n is the total number of pixels, Ω is the image domain, Ω = {(x i y i ): i = 1, ..., n}, I is (1) and I is (2) The first phase of remote sensing images I (1) Second Phase Remote Sensing Image I (2) The spectral value of inner pixel i in the s band.

[0071] Then the first phase of remote sensing images I (1) Second Phase Remote Sensing Image I (2) A remote sensing image of the target is formed by combining bands:

[0072] I = {I i (x i y i ):(x i y i)∈Ω, i=1,...,n},I i =(I is : s = 1, ...,

[0073] 2h).

[0074] Step S20: Segment the target remote sensing image based on multi-scale segmentation to obtain multi-scale segmentation results, and determine the optimal segmentation scale and optimal object in the multi-scale segmentation results based on the overall goodness-of-facts (F-value).

[0075] Specifically, the target remote sensing image I is divided into a group of objects O based on a multi-scale segmentation method. (c) ={O j (c) :j=1,...,m}, where c is the scale index, j is the object index, and m is the number of objects.

[0076] Specifically, the multi-scale segmentation method involves: initially dividing the target remote sensing image I based on graph theory to obtain over-segmented superpixel blocks, and then merging the over-segmented superpixel blocks based on the Fractal Net Evolution Approach (FNEA) to obtain the multi-scale segmentation result.

[0077] The optimal segmentation scale is determined based on the OGF (Overall Goodness F-measure). The larger the OGF value, the better the segmentation effect. The segmentation scale corresponding to the maximum OGF value is selected as the optimal scale. Once the optimal scale is determined, the objects under the corresponding optimal scale are the optimal objects.

[0078]

[0079] Where OGF represents the overall goodness-of-fact (F-value), and MI... normal and LV normal These are the standardized results of the MI index (i.e., Moran's I index) and the spectral local variance LV, respectively, with α being the weighting parameter.

[0080]

[0081] Among them, w jj′ w is the spatial adjacency weight if and only if superpixels j and j′ have a common boundary. jj′ =1,I j denoted as the mean of the pixel spectral measure within superpixel j, and μ as the mean of the image spectral measure.

[0082]

[0083] Where, n x and ny These represent the number of rows and columns of the image, |O j | represents the area of ​​superpixel j, σ j Let be the spectral standard deviation of superpixel j.

[0084] Finally, the segmentation scale corresponding to the maximum overall goodness F value is selected as the optimal segmentation scale, and the object under the optimal segmentation scale is the optimal object.

[0085] Step S30: Extract the spectral features of the object, construct a color co-occurrence matrix based on the object, and extract the texture feature statistics of the multi-band image through the color co-occurrence matrix to obtain texture information containing inter-band correlation.

[0086] Specifically, the spectral features of the object are extracted, using the mean μ of the spectral features of all pixels within the optimal object. j and variance σ j The spectral characteristics of the optimal object.

[0087] The target remote sensing image I is divided into N gray levels: I si * ∈{1, ..., N}, I si * This represents the gray level of pixel i in band s; based on the correlation between bands s and s', a color co-occurrence matrix (CCM) is constructed. ss’ (e, r, d, θ); where e and r are grayscale indices, e, r ∈ {1, ..., N}, d is the spatial distance, and θ represents the direction, typically selected from 4 or 8 directions uniformly distributed within 0°-360°; CCM ss’ (e, r, d, θ) represents the gray levels I of pixel pair i and i' that are d apart in the θ direction. si * =e,I s’i’ * = The probability of r appearing (s' represents another band, i' represents another pixel), the formula is as follows:

[0088]

[0089] Where Pr represents probability.

[0090] Based on Color Co-occurrence Matrix (CCM) ss’ (e, r, d, θ) extracts the texture feature statistics of multi-band images. The texture feature statistics include contrast, uniformity, second moment of angle, and correlation, etc., to obtain texture information containing inter-band correlation. That is, the texture feature statistics contain texture information of band correlation.

[0091] Step S40: Based on the multi-feature difference image, use the random forest model to evaluate the feature contribution, and use the cumulative backward search method to select the optimal feature combination.

[0092] Specifically, after determining the changed and unchanged object samples, a preset proportion of samples are selected as training samples for model training, and the remaining samples are used as the test set to evaluate the model training accuracy. To ensure the stability of the selected features, the random forest model is trained T times. Based on the cumulative value of the feature contribution after T training times, the features are re-sorted, and the features with the least importance are eliminated in turn to form new feature combinations, which are then input into the random forest model for T training times. This process is repeated until the number of features reaches the minimum, at which point training stops. Based on the fluctuation of the test accuracy under multiple feature combination conditions, the feature combination that meets the preset requirements for the number of features (e.g., a small number of features with high test accuracy, obtained by comprehensively considering experimental results) is selected as the optimal feature combination.

[0093] Step S50: Perform object-oriented random forest change detection based on the optimal feature combination to obtain the final remote sensing image change detection result.

[0094] Specifically, based on the feature selection results, the feature vectors of the samples are cleaned, and only the vector information of the optimal feature combination is retained. Based on this, the classifier is reconstructed, and the final classification result is the object-oriented remote sensing image change detection result.

[0095] This invention, based on the color co-occurrence matrix, introduces inter-band correlation into traditional texture extraction methods, extracting feature statistics that characterize intra-band and inter-band texture features respectively. Meanwhile, to reduce feature redundancy, this invention proposes a cumulative backward search method for feature selection to determine the optimal feature combination. On the one hand, this can more accurately characterize the texture features of the object, and on the other hand, it ensures the stability of the feature selection results, further improving the accuracy of change detection.

[0096] A comparative experiment was designed to further demonstrate the effectiveness of the method of the present invention:

[0097] The experimental environment was simulated using software programming on a system with a Core™ i5-7300U 2.60GHz CPU, 8GB of memory, and Windows 10 Ultimate.

[0098] Simulation experiments were conducted on changes in remote sensing images using the method of this invention, the traditional object-level gray-level co-occurrence matrix texture feature extraction method (comparison algorithm 1), and the single backward search method based on object GLCM (comparison algorithm 2), and the changes detection results were quantitatively evaluated.

[0099] like Figure 3 and Figure 4As shown, Figure 3 and 4 These are Sentinel-2 images of the Shenzhen area from 2021 and 2022, respectively.

[0100] Figures 5-8 For the change detection results, among which Figure 5 For standard results, Figure 6 , Figure 7 and Figure 8 The results are compared with those of Algorithm 1, Algorithm 2, and the method of this invention, respectively. Figures 5-8 It can be seen that the method of the present invention is closer to the standard change detection results.

[0101] To verify the effectiveness of the algorithm of this invention, the false negative rate, false positive rate, overall accuracy, and Kappa coefficient are listed in Table 1. It can be seen that the method of this invention outperforms the comparative algorithms in all indicators, accurately verifying the effectiveness of the method of this invention.

[0102] Experiment Name False negative rate False detection rate Overall accuracy Kappa coefficient Comparison Algorithm 1 18.52％ 4.35％ 88.89％ 0.7778 Comparison Algorithm 2 4.17％ 8.00％ 94.44％ 0.8880 Method of the present invention 1.12％ 0 98.12％ 0.9589

[0103] Furthermore, regarding the multi-scale segmentation method described in this invention, this invention uses graph theory and fractal network evolution algorithms as examples, but does not exclude combinations of other superpixel segmentation and superpixel merging algorithms. For example, Simple Linear Iterative Clustering (SLIC) and Entropy Rate (ERS) superpixel segmentation, etc. Regarding the training model and classifier described in this invention, this invention uses a random forest model as an example, but does not exclude other commonly used machine learning models, such as deep neural networks, Bayesian classifiers, etc.

[0104] Furthermore, such as Figure 9 As shown, based on the above-described object-oriented remote sensing image change detection method, the present invention also provides an object-oriented remote sensing image change detection system, wherein the object-oriented remote sensing image change detection system includes:

[0105] The image acquisition and combination module 51 is used to acquire the first phase remote sensing image and the second phase remote sensing image, and to combine the first phase remote sensing image and the second phase remote sensing image by band to form a target remote sensing image.

[0106] The multi-scale segmentation module 52 is used to segment the target remote sensing image based on multi-scale segmentation to obtain multi-scale segmentation results, and to determine the optimal segmentation scale and optimal object in the multi-scale segmentation results based on the overall goodness-of-fairness F value.

[0107] The texture feature extraction module 53 is used to extract the spectral features of the object, construct a color co-occurrence matrix based on the object, and extract the texture feature statistics of the multi-band image through the color co-occurrence matrix to obtain texture information containing the correlation between bands.

[0108] The feature selection module 54 is used to evaluate the feature contribution based on the multi-feature difference image using a random forest model and to select the optimal feature combination using a cumulative backward search method.

[0109] The image change detection module 55 is used to perform object-oriented random forest change detection based on the optimal feature combination to obtain the final remote sensing image change detection result.

[0110] Furthermore, such as Figure 10 As shown, based on the above-mentioned object-oriented remote sensing image change detection method and system, the present invention also provides a terminal, which includes a processor 10, a memory 20 and a display 30. Figure 10 Only some of the terminal components are shown; however, it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0111] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as a hard disk or memory. In other embodiments, the memory 20 may be an external storage device of the terminal, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc. Further, the memory 20 may include both internal and external storage devices. The memory 20 is used to store application software and various types of data installed on the terminal, such as the program code installed on the terminal. The memory 20 can also be used to temporarily store data that has been output or will be output. In one embodiment, the memory 20 stores an object-oriented remote sensing image change detection program 40, which can be executed by the processor 10 to implement the object-oriented remote sensing image change detection method of this application.

[0112] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor, or other data processing chip, used to run program code stored in the memory 20 or process data, such as executing the object-oriented remote sensing image change detection method.

[0113] In some embodiments, the display 30 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. The display 30 is used to display information on the terminal and to display a visual user interface. The components 10-30 of the terminal communicate with each other via a system bus.

[0114] In one embodiment, when the processor 10 executes the object-oriented remote sensing image change detection program 40 in the memory 20, it implements the steps of the object-oriented remote sensing image change detection method as described above.

[0115] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores an object-oriented remote sensing image change detection program, which, when executed by a processor, implements the steps of the object-oriented remote sensing image change detection method as described above.

[0116] In summary, this invention provides an object-oriented remote sensing image change detection method and related equipment. The method includes: acquiring a first-phase remote sensing image and a second-phase remote sensing image; combining the first-phase and second-phase remote sensing images by bands to form a target remote sensing image; segmenting the target remote sensing image based on multi-scale segmentation to obtain multi-scale segmentation results; determining the optimal segmentation scale and optimal object in the multi-scale segmentation results based on the overall goodness-of-facts (F-value); extracting the spectral features of the object; constructing a color co-occurrence matrix based on the object; extracting texture feature statistics of the multi-band image through the color co-occurrence matrix to obtain texture information containing inter-band correlations; evaluating the feature contribution using a random forest model based on the multi-feature difference image; and selecting the optimal feature combination using a cumulative backward search method; performing object-oriented random forest change detection based on the optimal feature combination to obtain the final remote sensing image change detection result. This invention uses objects as the basic operating unit. Based on the extraction of object spectral features, it further extracts multi-channel texture features of each object based on the color co-occurrence matrix, introduces inter-band correlation, and combines the cumulative backward search method to determine the optimal feature combination, thereby improving the accuracy of object remote sensing image change detection.

[0117] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal that includes that element.

[0118] Of course, those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.). The program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The computer-readable storage medium can be a memory, magnetic disk, optical disk, etc.

[0119] It should be understood that the application of the present invention is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. An object-oriented remote sensing image change detection method, characterized in that, The object-oriented remote sensing image change detection method includes: Acquire the first phase of remote sensing images and the second phase of remote sensing images, and combine the first phase of remote sensing images and the second phase of remote sensing images by band to form a target remote sensing image; The target remote sensing image is segmented based on multi-scale segmentation to obtain multi-scale segmentation results. The optimal segmentation scale and the optimal object in the multi-scale segmentation results are determined based on the overall goodness-of-facts (F-value). Extract the spectral features of the object, construct a color co-occurrence matrix based on the object, and extract the texture feature statistics of the multi-band image through the color co-occurrence matrix. The texture feature statistics contain texture information related to band correlation. Based on the multi-feature difference image, the feature contribution is evaluated using a random forest model, and the optimal feature combination is obtained by screening the texture feature statistics using a cumulative backward search method. Based on the optimal feature combination, object-oriented random forest change detection is performed to obtain the final remote sensing image change detection result; The acquisition of the first and second phases of remote sensing images, and the band combination of the first and second phases of remote sensing images to form a single target remote sensing image, specifically includes: Acquiring the first phase of remote sensing images I (1) Second phase remote sensing images I (2) ,and: I (1) = { I i (1) ( x i ， y i )：( x i ， y i )∈ Ω ， i = 1，...， n}； I (2) = { I i (2) ( x i ， y i )：( x i ， y i )∈ Ω ， i = 1，...， n}； in, i For pixel index, ( x i , y i (pixels) i The grid point coordinates in the image spatial domain, I i (1) and I i (2) These are the first phase of remote sensing images. I (1) Second phase remote sensing images I (2) Inner pixel i spectral measure vector, I i (1) = ( I is (1) : s =1,..., h ), I i (2) = ( I is (2) : s = 1, ..., h ), s For band indexing, h For the number of bands, n This represents the total number of pixels. Ω For the image domain, Ω = {( x i , y i ): i = 1, ..., n }, I is (1) and I is (2) These are the first phase of remote sensing images. I (1) Second phase remote sensing images I (2) Inner pixel i exist s Spectral values ​​of the band; The first phase of remote sensing images I (1) Second phase remote sensing images I (2) A remote sensing image of the target is formed by combining bands. I ; I = { I i ( x i ， y i )：( x i ， y i )∈ Ω ， i = 1，...， n}， I i = ( I is ： s = 1，...，2 h )； The process of segmenting the target remote sensing image based on multi-scale segmentation to obtain multi-scale segmentation results, and determining the optimal segmentation scale and optimal object in the multi-scale segmentation results based on the overall goodness-of-facts (F-value), specifically includes: The target remote sensing image is segmented based on a multi-scale segmentation method. I Divide into a group of objects O (c) = { O j (c) : j = 1, ..., m },in, c For scale indexing, j For object indexing, m Number of objects; Specifically, the multi-scale segmentation method involves using graph theory to segment the remote sensing image of the target. I An initial segmentation is performed to obtain oversegmented superpixel blocks, and the oversegmented superpixel blocks are merged based on a fractal network evolution algorithm to obtain the multi-scale segmentation result. Determining the optimal segmentation scale based on OGF: ； in, OGF The F-value represents the overall goodness. MI normal and LV normal These are the standardized results of the MI index and the spectral local variance LV, respectively. α These are weight parameters; The optimal segmentation scale is selected as the segmentation scale corresponding to the maximum value of the overall goodness F, and the object under the optimal segmentation scale is the optimal object. The process of evaluating feature contribution using a random forest model based on multi-feature difference images and selecting the optimal feature combination using a cumulative backward search method specifically includes: After determining the changed and unchanged object samples, a preset proportion of samples are selected as training samples for model training, and the remaining samples are used as the test set to evaluate the model training accuracy. The random forest model is trained T times. Based on the cumulative feature contribution value of the T training times, the features are re-sorted, and the features with the least importance are eliminated in turn to form a new feature combination. The new feature combination is then input into the random forest model for T training times until the number of features reaches the minimum and training stops. Based on the fluctuation of test accuracy under multiple feature combinations, the feature combination that meets the preset requirement in terms of the number of features is selected as the optimal feature combination.

2. The object-oriented remote sensing image change detection method according to claim 1, characterized in that, The extraction of object spectral features, the construction of a color co-occurrence matrix based on the object, and the extraction of texture feature statistics from multi-band images using the color co-occurrence matrix to obtain texture information containing inter-band correlations, specifically includes: Extract the spectral features of the object, using the average spectral features of all pixels within the optimal object. μ j and variance σ j Spectral characteristics of the optimal object; The target remote sensing image I Divided into N Gray levels: I si * ∈{1,..., N }, I si * Represents pixels i exist s Gray levels in the band; According to the band s and band s Inter-correlation, constructing a color co-occurrence matrix: CCM ss’ ( e , r, d, θ );in, e and r These are grayscale indexes, e , r ∈{1,..., N }, d For spatial distance, θ Indicates direction; CCM ss’ ( e , r , d , θ ) indicates in θ The distance in the direction is d pixel pairs i and i ' grayscale level I si * = e , I s’i’ * = r The probability of occurrence is given by the following formula: ； Based on Color Co-occurrence Matrix (CCM) ss’ ( e , r, d, θ Extract texture feature statistics from multi-band images to obtain texture information containing inter-band correlations.

3. The object-oriented remote sensing image change detection method according to claim 2, characterized in that, The texture feature statistics include contrast, uniformity, second moment of angle, and correlation.

4. The object-oriented remote sensing image change detection method according to claim 1, characterized in that, The process of performing object-oriented random forest change detection based on the optimal feature combination to obtain the final remote sensing image change detection result specifically includes: Based on the optimal feature combination, the feature vectors of the sample are cleaned, and only the vector information of the optimal feature combination is retained; The classifier is reconstructed to obtain the classification result, which is the object-oriented remote sensing image change detection result.

5. An object-oriented remote sensing image change detection system, characterized in that, The object-oriented remote sensing image change detection system is used to implement the object-oriented remote sensing image change detection method according to any one of claims 1-4, and the object-oriented remote sensing image change detection system includes: The image acquisition and combination module is used to acquire the first phase of remote sensing images and the second phase of remote sensing images, and to combine the first phase of remote sensing images and the second phase of remote sensing images by band to form a target remote sensing image. The multi-scale segmentation module is used to segment the target remote sensing image based on multi-scale segmentation to obtain multi-scale segmentation results, and to determine the optimal segmentation scale and optimal object in the multi-scale segmentation results based on the overall goodness-of-fairness F value. The texture feature extraction module is used to extract the spectral features of the object, construct a color co-occurrence matrix based on the object, and extract the texture feature statistics of multi-band images through the color co-occurrence matrix to obtain texture information containing inter-band correlation. The feature selection module is used to evaluate the feature contribution based on multi-feature difference images using a random forest model and to select the optimal feature combination using a cumulative backward search method. The image change detection module is used to perform object-oriented random forest change detection based on the optimal feature combination to obtain the final remote sensing image change detection result.

6. A terminal, characterized in that, The terminal includes: a memory, a processor, and an object-oriented remote sensing image change detection program stored in the memory and executable on the processor. When the object-oriented remote sensing image change detection program is executed by the processor, it implements the steps of the object-oriented remote sensing image change detection method as described in any one of claims 1-4.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores an object-oriented remote sensing image change detection program, which, when executed by a processor, implements the steps of the object-oriented remote sensing image change detection method as described in any one of claims 1-4.

Images