An optimization-based guided filter image fusion change detection method and device

Abstract

The application discloses a kind of based on optimization's guiding filter image fusion change detection method and device, method includes: by extreme minimum scale difference operator, pixel value difference value is to the picture of same scene different phase, obtains difference map;Using the way of guiding filter, difference map is fused on global scale, to obtain the optimal difference map;Mean filter is used to phase map I1 and I2, and the base layer difference map corresponding to each difference map is obtained, and the optimized brightness saliency map F1 and F2 are obtained by pca fusion as the image of generating saliency map, according to the principle of maximum saliency to determine weight map;Weight map is regularized for double-scale difference map reconstruction, and the final difference map is obtained;In clustering segmentation stage, by introducing soft threshold function, the final difference map is further processed, to suppress existing abnormal noise.The device includes: processor and memory.

Description

Technical Field

[0001] This invention relates to the field of change detection, and in particular to a method and apparatus for change detection based on optimized guided filtering image fusion. Background Technology

[0002] Change detection is a technique in the field of computer vision used to detect changes or motion in images or video sequences, or to perform change detection on satellite remote sensing images of the same area at different times. [1,2,3] It can also be applied to many fields, such as security monitoring, traffic monitoring, and environmental monitoring. In remote sensing change detection, by performing multidimensional analysis on multispectral images, its difference information is obtained, and finally segmented into binary images of the changed area and the sub-changed area, that is, the changed area is white and the unchanged area is black.

[0003] Change detection in video images involves analyzing two frames of video captured by a camera to determine if any anomalies exist in the monitored area. [4] A common approach is pixel-based change detection. This method identifies regions of change by comparing the value of each pixel in an image or video with the value of background pixels. For example, if the difference between two pixels exceeds a certain set threshold... [5,6] If a change is detected, it is considered to have occurred. Another approach is feature-based change detection. This method extracts features from an image or video, such as edges, textures, and colors, and uses clustering. [7,8] This method detects changes by comparing differences between features. Because of the richness of these differences, the detection results obtained in this way are more accurate. In addition, there are object-level change detection methods. [9] This will further reduce testing costs and shorten operation time.

[0004] In previous studies, change detection methods have shown limitations when processing video images in low-light, wide-field-of-view environments.

[10] The detection performance is unsatisfactory, as changed areas are greatly affected by random noise, leading to detection anomalies. Furthermore, since changed areas in low-light environments lack distinct features, it is difficult to distinguish the boundary between changed and unchanged areas. Moreover, previous change detection methods have not explored methods more suitable for change detection in low-light environments. Therefore, researching change detection in low-light, wide-field-of-view environments has significant practical implications and application value.

[0005] References

[0006] [1]Zhi Li et al.“A method to improve the accuracy of SAR image changedetection by using an image enhancement method”.In:ISPRS Journal ofPhotogrammetry and Remote Sensing 163(2020),pp.137–151.

[0007] [2]Kaiyu Zhang et al.“Unsupervised SAR image change detection for fewchanged area based on histogram fitting error minimization”.In:IEEETransactions on Geoscience and Remote Sensing 60(2022),pp.1–19.

[0008] [3]Jun Wang et al.“Unsupervised change detection between SAR imagesbased on hypergraphs”.In:ISPRS Journal of Photogrammetry and Remte Sensing164(2020),pp.61–72.

[0009] [4]Yong Zhu et al.“Change detection in multitemporal monitoringimages under low illumination”.In:Ieee Access 8(2020),pp.126700–126712.

[0010] [5]Patra S,Ghosh S,Ghosh A.Histogram thresholding for unsupervisedchange detection of remote sensing images[J].International journal of remotesensing,2011,32(21):6071-6089.

[0011] [6] Vázquez-Jiménez R, Ramos-Bernal R N, Romero-Calcerrada R, et al. Thresholding algorithm optimization for change detection to satellite imagery[J]. Color.Image process.’(Ed. CTravieso-Gonzalez) pp, 2018:163-182.

[0012] [7] Ghosh A, Mishra N S, Ghosh S. Fuzzy clustering algorithms for unsupervised change detection in remote sensing images[J]. Information Sciences, 2011, 181(4):699-715.

[0013] [8] Zheng Y, Zhang X, Hou B, et al. Using combined difference image and $k$-means clustering for SAR image change detection[J]. IEEE Geoscience and Remote Sensing Letters, 2013, 11(3):691-695.

[0014] [9] Chen G, Hay G J, Carvalho L M T, et al. Object-based change detection[J]. International Journal of Remote Sensing, 2012, 33(14):4434-4457.

[0015]

[10] Baoqiang Shi et al. “Unsupervised Change Detection in Wide-Field Video Images Under Low Illumination”. In: IEEE Transactions on Circuits and Systems for Video Technology 33.4(2022), pp.1564–1576. Summary of the Invention

[0016] This invention provides an optimized guided filter-based image fusion change detection method and apparatus. It proposes an extreme minimum scale difference operator to generate a difference map under low-light conditions; it proposes a PCA fusion method to generate a saliency map by fusing the global brightness map and the V component, and uses the maximum saliency principle to generate a weight map to guide the final difference map fusion; it proposes a method combining threshold segmentation and clustering algorithms, first using a soft thresholding method to remove pixels prone to incorrect clustering and then normalizing the difference map before final clustering and segmentation. Details are described below.

[0017] Firstly, an optimized guided filter-based image fusion change detection method, the method comprising:

[0018] The difference map is obtained by processing images of the same scene at different times using the extreme minimum scale difference operator and pixel value difference.

[0019] A guided filtering approach is used to fuse the difference maps at a global scale to obtain an optimized difference map.

[0020] The mean filter is applied to the phase maps I1 and I2 to obtain the base layer difference map corresponding to each difference map. The optimized brightness saliency maps F1 and F2 are obtained by PCA fusion and used as the images for generating saliency maps. The weight map is determined according to the principle of maximum saliency.

[0021] The weight map is regularized for dual-scale difference map reconstruction to obtain the final difference map;

[0022] In the clustering and segmentation stage, a soft threshold function is introduced to further process the final difference map to suppress any abnormal noise.

[0023] The extreme minimum scale difference operator is:

[0024]

[0025] Where α is used to ensure that the numerator and denominator are not zero, F is a control parameter, and I1(i,j) is the video temporal phase. Figure 1 I2(i,j) represents the video temporal phase. Figure 1 DI2(i,j) is the resulting difference map.

[0026] Specifically, the method of guided filtering is used to fuse the difference maps at a global scale to obtain the optimized difference map:

[0027] The multi-temporal phase maps I1 and I2 are converted from the RGB color space to the HSV color space, and the luminance component V is extracted and fused with the adaptive global luminance map using PCA.

[0028] The V component in the phase diagram and the calculated adaptive global luminance map are expanded into one-dimensional column vectors V and L. The mean value of each component is calculated. W and H represent the width and height of the input phase diagram, respectively. The output is the mean value μ of the luminance component V. V and adaptive global luminance component μ L ;

[0029]

[0030] Calculate the covariance matrix C of the luminance component V and the global luminance component L. V,L The covariance matrix is ​​decomposed into eigenvalues ​​to obtain the eigenvalue matrix, and the fusion weights are determined by the eigenvalue matrix.

[0031] The two input data sets are the globally adaptive luminance component L and the luminance component V extracted from the temporal phase map. V and L are fused, and the final weight value is calculated.

[0032]

[0033] F = a1*V + a2*L

[0034] Where a1 represents the fusion weight corresponding to the first feature value, i.e., the luminance component V in the phase diagram, a2 represents the fusion weight corresponding to the second feature value, i.e., the weight of the global adaptive luminance component, and the final saliency map of the fusion is F.

[0035] The weighted graph is as follows:

[0036]

[0037] in, This represents the saliency value of the k-th pixel in the n-th image.

[0038] The final difference diagram is as follows:

[0039]

[0040] Among them, B n and D n These represent the base layer and detail layer of the difference diagram, respectively. and These represent the weight graphs of the base layer and detail layer obtained through guided filtering, respectively.

[0041] The soft threshold function introduced in the clustering and segmentation stage is as follows:

[0042] Soft(I (x,y)|T )≡sign(I (x,y) )max{|I (x,y) |T,0}+T

[0043] Where T is the threshold of the soft thresholding function, and I (x,y) DI final The pixel value in the sign(I) (x,y) ) represents a symbolic function.

[0044] A second aspect is an optimized guided filter image fusion change detection device, characterized in that the device comprises: a processor and a memory, wherein the memory stores program instructions, and the processor calls the program instructions stored in the memory to cause the device to perform the method described in any one of the first aspects.

[0045] Third aspect, a computer-readable storage medium storing a computer program, the computer program including program instructions that, when executed by a processor, cause the processor to perform the method described in any one of the first aspects.

[0046] The beneficial effects of the technical solution provided by this invention are:

[0047] 1. This invention proposes an extreme minimum scale difference operator and controls its robustness in complex environments by using a power exponent. This method can more effectively preserve the edge of the changing region while suppressing abnormal noise.

[0048] 2. This invention proposes a PCA fusion method to optimize the saliency map of guided filter image fusion, which integrates the global brightness map and the V component, resulting in a more detailed and clearer contour in the final difference map in the change region.

[0049] 3. In the final detection stage, this invention proposes a method that combines threshold segmentation and clustering algorithm. First, a soft thresholding method is used to remove pixels that are prone to causing incorrect clustering, and the difference map is normalized. Then, the k-medoids algorithm is used to finally cluster and obtain the change detection segmentation map. Attached Figure Description

[0050] Figure 1 The flowchart shows an optimized guided filter-based image fusion change detection method.

[0051] Figure 2 This is a schematic diagram comparing the difference chart and the heatmap.

[0052] Figure 3 This is a schematic diagram of the change detection results under a multi-light source environment;

[0053] Among them, (a) time phase Figure 1 (b) Phase Figure 2 (c) Truth graph; (d) Algorithm results.

[0054] Figure 4This is a schematic diagram of the change detection results in a dark area environment;

[0055] Among them, (a) time phase Figure 1 (b) Phase Figure 2 (c) Truth graph; (d) Algorithm results.

[0056] Figure 5 A schematic diagram for detecting spurious changes after adding noise.

[0057] Among them, (a) the phase with added noise Figure 1 (b) Adding noise during the phase Figure 2 (c) Truth graph; (d) Algorithm results. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below.

[0059] I. Generation of Difference Charts

[0060] In change detection, the quality of the generated difference map is fundamental to achieving the best detection results. In low-light environments, images of the same scene at different times are easily affected by noise from light sources and flat dark areas. Logarithmic ratio operations can convert multiplicative noise in the environment into additive noise, but they still cannot effectively eliminate the influence of abnormal noise. This invention proposes an extreme minimum scale difference operator, which can improve robustness to abnormal noise, such as in formula (1). Here, α is used to ensure that the numerator and denominator are not zero, and F is the control parameter of the extreme minimum operator. Another difference operator is generated using pixel value difference values, as shown in formula (2).

[0061]

[0062]

[0063] By using the extreme value method in formula (1), the local energy of the obtained change region value can be maximized, while suppressing the noise of the difference map to the minimum, reducing the impact of random noise on the detection results and highlighting the change region more clearly. Formula (1) can obtain a clearer outline of the change region while reducing the noise in the non-change region.

[0064] II. PCA luminance saliency diagram

[0065] The difference map generated by the operator will still retain noise information to varying degrees, and the characteristics and advantages of different difference maps are different. This embodiment of the invention proposes to use guided filtering to fuse two difference maps at a global scale to obtain the optimal difference map.

[0066] First, the RGB color space is converted to the HSV color space. Based on the Weber-Fechner theorem, a logarithmic compression strategy is adopted to dynamically and globally compress the phase diagram and obtain a globally adaptive global brightness map.

[0067]

[0068] To address the issue of low pixel gradient in low illumination, the global adaptive brightness L is calculated using equation (3). g (i,j), where L w L represents the brightness value in the phase diagram. wmax This represents the maximum brightness value in the input phase diagram. The logarithmic mean luminance value is expressed as in formula (4):

[0069]

[0070] Here, N represents the number of pixels in the phase image, and β represents a small value to avoid odd values ​​caused by pixels having a value of 0 in dark areas. In low-light environments, a single luminance component cannot fully represent the global brightness reflected by objects. Therefore, this embodiment of the invention fuses the obtained multi-scale luminance features using Principal Component Analysis (PCA). The main luminance information is used as the first component, minimizing the luminance information with low information content in the transform domain, thus preserving the information of the original data to the maximum extent. The multi-phase images I1 and I2 are converted from the RGB color space to the HSV color space. The V component in the HSV color space reflects the brightness of objects; higher brightness is closer to white, and lower brightness is closer to black. The luminance component V is extracted and fused with the adaptive global luminance map using PCA, maximizing the luminance features in all dimensions while suppressing the influence of light source noise and dark area smoothing noise on the luminance component.

[0071]

[0072] Here, the V component in the phase diagram and the calculated adaptive global luminance map are expanded into one-dimensional column vectors V and L, as shown in formula (5). Then, the mean values ​​of these two components are calculated, as shown in formula (6). Here, W and H represent the width and height of the input phase diagram, respectively, and the calculated output is the mean value μ of the luminance component V. V And adaptive global luminance component μx.

[0073]

[0074] Calculate the covariance matrix C of the luminance component V and the global luminance component L. V,LThen, the covariance matrix is ​​decomposed into eigenvalues ​​to obtain the eigenvalue matrix, and the fusion weights are finally determined by the eigenvalue matrix. The calculation method is shown in formula (7). After obtaining the covariance matrix, the eigenvalue matrix and eigenvectors are calculated as shown in formula (8), where Q is the matrix composed of eigenvectors, D is a diagonal matrix, and the elements on its diagonal are C. V,L Eigenvalues ​​of a matrix.

[0075]

[0076] C V,L =QDQ -1 (8)

[0077] The two sets of input data here are the globally adaptive luminance component L and the luminance component V extracted from the phase map, so the obtained feature values ​​are a1 and a2. Next, in order to effectively fuse V and L, the final weight value is calculated. The calculation method is shown in formula (9). In low-light environments, since the gradient change between the target and the surrounding environment is not obvious in the dark area, the final saliency map is obtained by fusing the extracted luminance component V and the globally adaptive luminance component. This map is used to guide the generation of the weight map in the difference map fusion stage. The feature value matrix is ​​obtained according to formula (8):

[0078]

[0079] F=a1*V+a2*L (10)

[0080] Here, a1 represents the fusion weight corresponding to the first eigenvalue, i.e., the luminance component V in the phase diagram, and a2 represents the fusion weight corresponding to the second eigenvalue, i.e., the weight of the globally adaptive luminance component. The final saliency map of the fusion is F, and the calculation method is shown in formula (10).

[0081] III. Guided Filter Difference Map Fusion

[0082] To improve the quality of difference maps generated in low-light environments, a saliency map fused with brightness is proposed as a guide map for the output image in guided filtering, ensuring that the gradient of the output image remains consistent with the guide image. Furthermore, brightness fusion optimizes the final saliency map, thus ensuring that the characteristics of each difference map generation operator are fully utilized during the fusion stage.

[0083] First, a mean filter is applied to the phase diagrams I1 and I2 to obtain the underlying difference diagram corresponding to each difference diagram. The size of the mean filter is set to 31×31. The calculation method is shown in formula (11):

[0084] B n =DI n *A (11)

[0085] Here, n = 1, 2, is used to calculate the basic features of the two sets of difference maps, and at the same time separate the detail layers of the difference maps. The main purpose here is to ensure the integrity of the change area in the detail layer. The calculation method is shown in formula (12):

[0086] D n =DI n -B n (12)

[0087] Unlike previous guided image fusion methods, this approach omits the Laplacian filter and low-pass filter reconstruction process for the guided image because an optimized saliency map has already been obtained. In the second part, optimized brightness saliency maps F1 and F2 are obtained through PCA fusion and used as the images for generating the saliency map. Next, a weight map is determined based on the principle of maximum saliency. As shown in formula (13):

[0088]

[0089] here This represents the saliency value of the k-th pixel in the n-th image. Thus, the weighted map is obtained based on the principle of maximum pixel saliency. However, such a weight map is susceptible to severe noise interference and is blurry at the boundaries between changing and non-changing regions, leading to excessive noise during the fusion process. This is especially true in dark areas, where pixel values ​​are relatively flat and easily affected by pixel value jump noise. Therefore, based on the principle of spatial consistency, F... n As a guide map, for (Here, n = 1, 2) Guided filtering is performed to obtain a spatially consistent weight map with smooth transition edges. The calculation formula is shown in (14):

[0090]

[0091] Here (Here, n = 1, 2) represents the pilot filter. and These represent the weight maps of the base layer and detail layer obtained through guided filtering, respectively. Finally, the weight maps are regularized for dual-scale difference map reconstruction, and the final difference map fusion calculation is shown in formula (15). Where B... n and D n These represent the base layer and detail layer of the difference map, respectively. final This represents the difference diagram obtained from the final fusion.

[0092]

[0093] IV. Two-stage clustering algorithm based on K-medoids

[0094] To ensure the accuracy of binary classification results during the clustering stage, a combination of threshold segmentation and clustering algorithms is proposed to guarantee the stability of the classification results. In the clustering segmentation stage, a soft threshold function is introduced to adjust the threshold for binary classification (DI). final Further processing is performed to suppress any remaining abnormal noise in the difference map. The calculation method is shown in formula (16):

[0095] Soft(I (x,y)|T )≡sign(I (x,y) )max{|I (x,y) |T,0}+T (16)

[0096] Here, T is the threshold of the soft thresholding function. After extensive experimental testing, T = 0.1 is set here. (x,y) DI final The pixel values ​​in the image are analyzed. Since pixel values ​​vary significantly in changing regions and less so in static regions, k-medoids are used to achieve the final binary image segmentation.

[0097] To verify the effectiveness of the proposed change detection algorithm in low-light wide field-of-view environments, experiments were conducted in different environments.

[0098] V. Quality Analysis of Difference Charts

[0099] from Figure 2 It can be seen that the proposed difference graph operator and difference graph fusion method yielded the optimal difference graph.

[0100] VI. Detection of Environmental Changes with Multiple Light Sources

[0101] 1) Dark Environment Change Detection

[0102] from Figure 3 and Figure 4 As can be seen from the figure, the detection effect of the change detection algorithm proposed in the embodiments of the present invention is very close to the ground truth map, and it can accurately detect the change area in both multi-light source environment and dark environment, while preserving the edge contour of the change area well.

[0103] 2) Change detection under random noise

[0104] To further test the noise resistance of each algorithm in complex environments, noise was artificially added to the video images before detection was performed. Figure 5 As can be seen, the addition of random noise can still effectively suppress and eliminate the influence of noise, and accurately detect the area of ​​change.

[0105] An optimized guided filter-based image fusion change detection device includes a processor and a memory. The memory stores program instructions, and the processor calls the program instructions stored in the memory to cause the device to execute the following method steps in Embodiment 1:

[0106] The difference map is obtained by processing images of the same scene at different times using the extreme minimum scale difference operator and pixel value difference.

[0107] A guided filtering approach is used to fuse the difference maps at a global scale to obtain an optimized difference map.

[0108] The mean filter is applied to the phase maps I1 and I2 to obtain the base layer difference map corresponding to each difference map. The optimized brightness saliency maps F1 and F2 are obtained by PCA fusion and used as the images for generating saliency maps. The weight map is determined according to the principle of maximum saliency.

[0109] The weight map is regularized for dual-scale difference map reconstruction to obtain the final difference map;

[0110] In the clustering and segmentation stage, a soft threshold function is introduced to further process the final difference map to suppress any abnormal noise.

[0111] The extreme minimum scale difference operator is:

[0112]

[0113] Where α is used to ensure that the numerator and denominator are not zero, F is a control parameter, and I1(i,j) is the video temporal phase. Figure 1 I2(i,j) represents the video temporal phase. Figure 2 DI2(i,j) is the resulting difference map.

[0114] Specifically, guided filtering is used to fuse the difference maps at a global scale to obtain the optimized difference map:

[0115] The multi-temporal phase maps I1 and I2 are converted from the RGB color space to the HSV color space, and the luminance component V is extracted and fused with the adaptive global luminance map using PCA.

[0116] The V component in the phase diagram and the calculated adaptive global luminance map are expanded into one-dimensional column vectors V and L. The mean values ​​of each component are calculated. W and H represent the width and height of the input phase diagram, respectively. The output is the mean value μV of the luminance component V and the mean value μ of the adaptive global luminance component. L ;

[0117]

[0118] Calculate the covariance matrix C of the luminance component V and the global luminance component L.V,L The covariance matrix is ​​decomposed into eigenvalues ​​to obtain the eigenvalue matrix, and the fusion weights are determined by the eigenvalue matrix.

[0119] The two input data sets are the globally adaptive luminance component L and the luminance component V extracted from the temporal phase map. V and L are fused, and the final weight value is calculated.

[0120]

[0121] F = a1*V + a2*L

[0122] Where a1 represents the fusion weight corresponding to the first feature value, i.e., the luminance component V in the phase diagram, a2 represents the fusion weight corresponding to the second feature value, i.e., the weight of the global adaptive luminance component, and the final saliency map of the fusion is F.

[0123] The weighted graph is as follows:

[0124]

[0125] in, This represents the saliency value of the k-th pixel in the n-th image.

[0126] The final difference graph is as follows:

[0127]

[0128] Among them, B n and D n These represent the base layer and detail layer of the difference diagram, respectively. and These represent the weight graphs of the base layer and detail layer obtained through guided filtering, respectively.

[0129] In the clustering and segmentation stage, a soft threshold function is introduced as follows:

[0130] Soft(I (x,y)|T )≡sign(I (x,y) )max{|I (x,y) |T,0}+T

[0131] Where T is the threshold of the soft thresholding function, and I (x,y) DI final The pixel value in the sign(I) (x,y) ) represents a symbolic function.

[0132] It should be noted that the device descriptions in the above embodiments correspond to the method descriptions in the embodiments, and the embodiments of the present invention will not be repeated here.

[0133] The execution entities of the aforementioned processor and memory can be devices with computing functions such as computers, microcontrollers, and single-chip microcomputers. In specific implementations, the embodiments of the present invention do not limit the execution entities and can select them according to the needs of actual applications.

[0134] Data signals are transmitted between the memory and the processor via a bus, which will not be elaborated upon in this embodiment of the invention.

[0135] Based on the same inventive concept, embodiments of the present invention also provide a computer-readable storage medium, the storage medium including a stored program, which, when the program is running, controls the device where the storage medium is located to execute the method steps in the above embodiments.

[0136] The computer-readable storage medium includes, but is not limited to, flash memory, hard disk, solid-state drive, etc.

[0137] It should be noted that the description of the readable storage medium in the above embodiments corresponds to the description of the method in the embodiments, and the embodiments of the present invention will not be repeated here.

[0138] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated.

[0139] A computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in or transmitted through a computer-readable storage medium. A computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium can be magnetic or semiconductor, etc.

[0140] Unless otherwise specified, the model numbers of the various devices in this embodiment of the invention are not limited, and any device that can perform the above functions is acceptable.

[0141] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0142] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for detecting image fusion changes based on optimized guided filtering, characterized in that, The method includes: The difference map is obtained by processing images of the same scene at different times using the extreme minimum scale difference operator and pixel value difference. A guided filtering approach is used to fuse the difference maps at a global scale to obtain an optimized difference map. Phase diagram and A mean filter is used to obtain the underlying difference map corresponding to each difference map, and PCA fusion is used to obtain the optimized brightness saliency map. and The weight map is determined based on the principle of maximum saliency for the image used to generate the saliency map; The weighted graph is regularized for dual-scale difference map reconstruction to obtain the final difference map; In the clustering and segmentation stage, a soft threshold function is introduced to further process the final difference map in order to suppress any abnormal noise. The method employs guided filtering to fuse the difference maps at a global scale, resulting in an optimized difference map: Multi-temporal diagram and Convert from RGB color space to HSV color space, extract the luminance component V and perform PCA fusion with the adaptive global luminance map; The V component in the phase diagram and the calculated adaptive global luminance map are expanded into one-dimensional column vectors V and L. The mean value of each component is calculated. W and H represent the width and height of the input phase diagram, respectively. The output is the mean value of the luminance component V. and adaptive global luminance component ; Calculate the covariance matrix of the luminance component V and the global luminance component L. The covariance matrix is ​​decomposed into eigenvalues ​​to obtain the eigenvalue matrix, and the fusion weights are determined by the eigenvalue matrix. The two input data sets are the globally adaptive luminance component L and the luminance component V extracted from the temporal phase map. V and L are fused, and the final weight value is calculated. in, This represents the fusion weight corresponding to the first eigenvalue, i.e., the brightness component V in the phase diagram. The fusion weight corresponding to the second eigenvalue is represented by F, which is the weight of the global adaptive brightness component L. The final saliency map of the fusion is F.

2. The image fusion change detection method based on optimized guided filtering according to claim 1, characterized in that, The extreme minimum scale difference operator is: Here, α is used to ensure that the numerator and denominator are not zero, and F is a control parameter. This is video phase diagram 1. This is video phase diagram 2. The resulting difference graph.

3. The image fusion change detection method based on optimized guided filtering according to claim 1, characterized in that, The weighted graph is as follows: in, This represents the saliency value of the k-th pixel in the n-th image.

4. The image fusion change detection method based on optimized guided filtering according to claim 1, characterized in that, The final difference diagram is as follows: in, and These represent the base layer and detail layer of the difference diagram, respectively. and These represent the weight graphs of the base layer and detail layer obtained through guided filtering, respectively.

5. The image fusion change detection method based on optimized guided filtering according to claim 1, characterized in that, The soft threshold function introduced in the clustering segmentation stage is as follows: Where T is the threshold of the soft thresholding function. express Pixel values ​​in ) represents a symbolic function.

6. A device for detecting image fusion changes based on optimized guided filtering, characterized in that, The device includes a processor and a memory, the memory storing program instructions, the processor invoking the program instructions stored in the memory to cause the device to perform the method according to any one of claims 1-5.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, the computer program including program instructions that, when executed by a processor, cause the processor to perform the method described in any one of claims 1-5.

Images