Damage detection method based on image change analysis
Through image preprocessing, feature extraction and offset comparison methods, the problems of slow detection speed, low accuracy and poor versatility of existing damage detection algorithms are solved, and high-precision damage detection is achieved, which is suitable for a variety of imaging media and damage types.
Patent Information
- Application Number
- CN202510903788.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-01
- Publication Date
- 2025-10-10
AI Technical Summary
Existing damage detection algorithms have problems such as large data requirements, slow detection speed, strong subjectivity, difficulty in image registration, and difficulty in feature extraction, resulting in insufficient detection accuracy and versatility.
The image preprocessing, LBP and Haar feature extraction, multi-group offset comparison and feature fusion methods are adopted. Through spatial filtering denoising, morphological processing, local texture and regional change feature extraction, combined with offset comparison and fusion algorithms, accurate positioning of the damaged area is achieved.
It improves the accuracy and versatility of damage detection, reduces the false alarm rate, is applicable to a variety of imaging media and damage types, reduces the dependence on image registration accuracy, and is suitable for military and natural disaster detection.
Smart Images
Figure CN120765602A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of image analysis, and particularly discloses a damage detection method based on image change analysis. Background Art
[0002] Currently, existing damage detection algorithms have many problems. Some soft change detection methods, such as those based on Bayesian network inference models, are image-independent and output mathematical expectations, but they require a large amount of prior data. Most hard change detection and combined hard and soft detection methods rely on images before and after the damage. Although data acquisition technology has developed rapidly, traditional manual interpretation methods are slow and highly subjective when faced with massive amounts of data. Among the hard change damage detection methods based on image analysis, the difference method and ratio method have high requirements for image registration, making actual accurate registration difficult; the method based on feature point matching has strict requirements for image features, making it difficult to extract smooth image features; and the detection methods for specific targets have poor versatility.
[0003] Therefore, it is necessary to develop a damage detection method based on image change analysis to solve the above problems. Summary of the Invention
[0004] The object of the present invention is to provide a damage detection method based on image change analysis.
[0005] In order to achieve the above object, the present invention provides the following technical solutions:
[0006] A damage detection method based on image change analysis comprises the following steps:
[0007] S1. Image preprocessing: Acquire two images before and after the damage. The images are acquired based on the actual time span and the main area to be detected is the main area of the image. The imaging medium can be a variety of heterogeneous images such as optical, SAR, and multispectral images, and the two images are of the same type. Spatial filtering is performed on the acquired images to remove noise, and morphological processing is used to improve image quality and highlight target features.
[0008] S2. Feature extraction: Using LBP and Haar features that are insensitive to illumination and color, the feature fusion algorithm is used to extract the local texture features and regional change features of the damaged area, and the location of the damaged area is preliminarily obtained;
[0009] S3. Damage detection: Using the offset comparison method, multiple groups of offset comparisons are selected based on different offset values, and the final accurate location of the damaged area is obtained by fusing the comparison results. This method performs damage detection under the condition of coarse image registration, that is, when a registration error of 2-4 pixels can be accepted.
[0010] Specifically, the spatial domain filtering adopts Gaussian filtering, and removes Gaussian noise in the image by setting the filter kernel size and standard deviation.
[0011] Specifically, the morphological processing adopts erosion and dilation operations, uses a circular structure element, first performs erosion to remove small noise points and burrs, and then performs dilation to restore the size of the target object.
[0012] Specifically, the LBP feature extraction is to calculate the LBP value of each pixel in the image to construct an LBP feature map, the Haar feature extraction is used to detect the features of edges and corners in the image, and the feature fusion adopts a weighted summation method to set weights according to the importance of different features to damage detection.
[0013] Specifically, in the offset comparison, offsets of ±1 pixel, ±2 pixels, and ±3 pixels are set in the horizontal and vertical directions respectively to perform multiple comparisons, and the difference values of the corresponding areas of the images before and after the damage are calculated. The difference value calculation adopts the pixel difference or similarity measurement method.
[0014] Specifically, the fusion of the multiple groups of offset comparison results adopts a mean fusion or a confidence-based fusion algorithm.
[0015] The beneficial effects of the present invention are:
[0016] The damage detection method presented in this paper is highly versatile, performing detection based on pixel-level changes in images, independent of the specific target being damaged. It is applicable not only to detecting damage from artillery fire, but also to detecting changes in natural disasters, urban expansion, and other scenarios. By fusing multiple sets of offset comparison results and processing feature images, the false alarm rate is reduced and detection accuracy is improved. Compared to traditional methods, this method requires less precision in the registration of pre- and post-damage images, resulting in fewer restrictions and superior detection results. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] Figure 1 This is an image processing flow chart of a damage detection method based on image change analysis according to an embodiment of the present invention;
[0018] Figure 2 Calculation diagram of LBP eigenvalues shown in an embodiment of the present invention;
[0019] Figure 3 A schematic diagram of a Haar feature matrix template and corresponding pixel values shown in an embodiment of the present invention;
[0020] Figure 4 Schematic diagram of a damage area based on LBP feature extraction according to an embodiment of the present invention;
[0021] Figure 5 This is a damage area effect diagram obtained by comparing different offset fusions shown in an embodiment of the present invention. DETAILED DESCRIPTION
[0022] The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.
[0023] See also Figure 1 The embodiment of the present invention provides a damage detection method based on image change analysis, which specifically includes the following steps:
[0024] S1. Image preprocessing: Taking a battlefield damage detection task as an example, optical imaging images are acquired before and after damage. The images are then fed into a spatial filtering algorithm, such as a Gaussian filter, with an appropriate filter kernel size and standard deviation, to remove Gaussian noise from the image. Morphological processing then proceeds, using erosion and dilation operations. An appropriate structuring element (such as a circular structuring element) is selected. Erosion is first performed to remove small noise points and glitches in the image, followed by dilation to restore the target object's size and highlight the contours of the damaged area.
[0025] S2. Feature Extraction Stage: For the preprocessed image, the LBP algorithm is used to calculate the LBP value for each pixel in the image, constructing an LBP feature map that reflects the local texture details of the image. Simultaneously, the Haar feature extraction algorithm is used to detect features such as edges and corners in the image. The LBP feature map and Haar features are fused, for example, through a weighted summation method, assigning weights to different features based on their importance to damage detection, to preliminarily determine the possible location of the damage area.
[0026] S3, Damage Detection Stage: Multiple offsets are set on the coarsely registered pre- and post-damage images, such as ±1 pixel, ±2 pixels, and ±3 pixels horizontally and vertically, respectively. For each offset, the difference between the corresponding regions in the pre- and post-damage images is calculated, for example, using pixel difference or similarity metrics. The difference values obtained from these multiple offset comparisons are fused, for example, using mean fusion or a confidence-based fusion algorithm, to determine the precise location of the final damage region, completing the damage detection task.
[0027] Damage area extraction based on LBP features
[0028] Two images taken at different times under the same battlefield environment may be affected by external factors such as lighting due to the different shooting times. Since the LBP feature extraction process quantifies the relationship between the surrounding points and the current point, the quantization can more effectively eliminate the influence of lighting on the image. Therefore, this application uses LBP features to obtain the damaged area in the image. The calculation steps of LBP features are as follows:
[0029] See also Figure 2 , select a 3*3 (pixel) window in the original image for traversal, and set the center point to , the pixel value is , surrounded by eight areas The corresponding pixel value is , the center point and Compare one by one, if ,but ,otherwise .
[0030] Different from the traditional LBP feature calculation, this application uses the comparison result in the form of a vector as the LBP feature of the point, expressed as Since the LBP algorithm is very robust to environmental changes such as illumination, For a non-damaged area, the correlation between this point and the surrounding pixels will not change much due to changes in time or illumination. Therefore, this application uses formula (1) and formula (2) to calculate Whether the point is a damage area:
[0031] (1)
[0032] (2)
[0033] in, express The point is the probability value of the damage area; Expressed as The length of the eigenvector; Indicates the image before and after damage The feature corresponding to the point; is the threshold value used to judge Whether the point is a damage area; The damage results are shown in the figure The pixel value of the point. The damage area extraction results of the features are as follows Figure 3 shown.
[0034] Damage area extraction based on fusion of different features
[0035] In the ever-changing battlefield environment, the damage area cannot be effectively and accurately extracted by extracting a single feature. Therefore, a method of fusion of different features is adopted to comprehensively consider the extraction of different features to improve the detection accuracy. This application selects Haar features that reflect the local pixel changes of the detection target and fuses them with the above-mentioned LBP features.
[0036] First, perform Haar feature extraction: see Figure 3 This application uses three types of Haar feature templates with resolutions of 5*5, 7*7, 9*9, and 11*11 to traverse the image. The corresponding number of eigenvalues is 3*4=12. Define the template Corresponding eigenvalue The sum of the pixel values in the white rectangle in the template minus the sum of the pixel values in the black rectangle. Perform binarization processing, if ,but ,otherwise Get the template The corresponding eigenvalue , and the Haar feature of the point is expressed in the form of a vector. .
[0037] In order to obtain a more accurate damage area, this application uses a feature fusion method to compare the LBP and Haar features of the images before and after damage to obtain the damage area. The judgment formula is as follows:
[0038] (3)
[0039] (4)
[0040] in, express The point is the probability value of the damage area; Expressed as the length of the Haar feature vector; Indicates the image before and after damage The Haar feature corresponding to the point; Respectively represent weights; is the threshold value used to judge Whether the point is a damage area; The damage results are shown in the figure The pixel value of the point. The damage area extraction result based on LBP and feature fusion is as follows Figure 4 shown.
[0041] Damage area fusion based on offset comparison
[0042] In battlefield environments, image acquisition at different time periods in the same area may be difficult to achieve accurate registration and alignment due to interference from external factors such as war. Therefore, this application uses an offset comparison method to compare the features of the image before damage with the offset image. The damaged images are compared to obtain different damaged areas. Finally, the different damaged areas are fused and compared by the fusion algorithm to obtain the final more accurate damaged area, so that damage detection can be performed under rough image registration, reducing the impact of registration accuracy on the detection process. The offset comparison algorithm is shown in formulas (4)-(6):
[0043] ;(5)
[0044] ; (6)
[0045] in, Indicates a point The probability value of the damaged area after offset comparison; Indicates the offset; Indicates the midpoint of the image before damage The corresponding features are LBP and Haar; Indicates the midpoint of the image after damage After the offset The corresponding LBP and Haar features; Expressed as the length of the LBP feature vector; Expressed as the length of the Haar feature vector; Respectively represent weights; is the threshold value used to judge Whether the point is a damage area; The damage results are shown in the figure The pixel value of the point; It is a feature fusion algorithm, which represents the fusion of effect graphs with different offsets. It can be an "and" operation, an "or" operation, etc. This shows the final damage area effect.
[0046] Since this application uses the offset comparison algorithm to obtain the final damage area effect map, the selection of the offset becomes particularly important. Selecting the appropriate offset has an important impact on the accurate acquisition of the results. This application verifies the impact of the offset on the detection accuracy by selecting different offsets. The selection is shown in Table 1:
[0047] Table 1 Offset directions represented by different offsets
[0048] The offset in the table is expressed as: ,in Represents the direction of the offset, respectively 、 、 、 ; Represents the offset size in pixels.
[0049] In order to verify the influence of offset comparison in different directions on the result characteristics, this application selects three different fusion offsets. The first one is The offset is fused to obtain the result; the second option is The offset is fused to obtain the result; the third option is The following are the damage area effects obtained when three different offsets are fused. By comparison, it can be seen that the selection of different offsets has a great impact on the accuracy of the detection results. Figure 5 shown.
[0050] In summary, this application effectively solves the shortcomings of existing damage detection technology through a unique image change analysis method, and has broad application prospects in change detection in military and other fields.
[0051] The above-described embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the patent for this invention shall be determined by the appended claims.
Claims
1. A damage detection method based on image change analysis, characterized in that: The following steps are involved: S1. Image preprocessing: Acquire two images before and after the damage. The images are acquired based on the actual time span and the main area to be detected is the main area of the image. The imaging medium can be a variety of heterogeneous images such as optical, SAR, and multispectral images, and the two images are of the same type. Spatial filtering is performed on the acquired images to remove noise, and morphological processing is used to improve image quality and highlight target features. S2. Feature extraction: Using LBP and Haar features that are insensitive to illumination and color, the feature fusion algorithm is used to extract the local texture features and regional change features of the damaged area, and the location of the damaged area is preliminarily obtained; S3. Damage detection: Using the offset comparison method, multiple groups of offset comparisons are selected based on different offset values, and the final accurate location of the damaged area is obtained by fusing the comparison results. This method performs damage detection under the condition of coarse image registration, that is, when a registration error of 2-4 pixels can be accepted.
2. The damage detection method based on image change analysis according to claim 1, characterized in that: The spatial domain filtering adopts Gaussian filtering, and removes Gaussian noise in the image by setting the filter kernel size and standard deviation.
3. The damage detection method based on image change analysis according to claim 1, characterized in that: The morphological processing adopts erosion and dilation operations, uses a circular structure element, first performs erosion to remove small noise points and burrs, and then performs dilation to restore the size of the target object.
4. The damage detection method based on image change analysis according to claim 1, characterized in that: The LBP feature extraction is to calculate the LBP value of each pixel in the image to construct an LBP feature map. The Haar feature extraction is used to detect the features of edges and corners in the image. The feature fusion adopts a weighted summation method to set weights according to the importance of different features to damage detection.
5. The damage detection method based on image change analysis according to claim 1, characterized in that: In the offset comparison, offsets of ±1 pixel, ±2 pixels, and ±3 pixels are set in the horizontal and vertical directions respectively to perform multiple comparisons, and the difference values of the corresponding areas of the images before and after the damage are calculated. The difference value calculation adopts the pixel difference or similarity measurement method.
6. The damage detection method based on image change analysis according to claim 1 or 5, characterized in that: The fusion of the multiple groups of offset comparison results adopts a mean fusion or a confidence-based fusion algorithm.