X-ray image background suppression method for foreign particle detection of GIS equipment

Abstract

This invention discloses an X-ray image background suppression method for foreign object particle detection in GIS equipment, relating to the field of GIS equipment detection. The invention includes performing preliminary frequency domain filtering on the image to be tested, extracting candidate particles, filtering out high frequencies from the image, strongly correlating the remaining structural contours with a pre-built background template library, performing matching at the feature layer to output the best-matching background image, subtracting the best-matching background from the original image, performing secondary frequency domain filtering, taking the intersection with the candidate particles extracted after the initial filtering, and outputting the final candidate particle detection result. The X-ray image background suppression method for foreign object particle detection in GIS equipment provided by this invention improves the structural background suppression ratio and enhances the sensitivity to small foreign objects. By using intersection and dual-path fusion, the false alarm rate is reduced, improving the algorithm's robustness.

Description

Technical Field

[0001] This invention relates to the field of GIS equipment inspection, and more specifically to a method for suppressing background in X-ray images for detecting foreign particles in GIS equipment. Background Technology

[0002] Background suppression in X-ray images is a key technology for improving the accuracy of foreign particle detection inside GIS (Gas Insulated Switchgear). The core challenge lies in the fact that the complex internal structure of the equipment (such as conductors, bolts, and basin insulators) forms a complex background with high intensity and high texture, while the signal of tiny foreign particles is often weak and has low contrast, making them easily "submerged" by the background.

[0003] External physical apertures can address the problem at its source, significantly improving the signal-to-noise ratio (SNR), but they represent a system-level modification, resulting in poor flexibility and high cost. Industrial applications prefer software-based solutions. Image differencing, which subtracts a background image (a template) free of foreign objects from the image under test, highlights the differences. In industrial inspection, pure background images are often obtained through methods such as time-series integration, but this relies heavily on accurate background templates, requires high consistency in imaging positions, and presents difficulties in modeling complex and irregular background structures.

[0004] Therefore, there is an urgent need for a background suppression method for X-ray images used in the detection of foreign particles in GIS equipment. Summary of the Invention

[0005] This invention provides a method for suppressing background in X-ray images for detecting foreign particles in GIS equipment, thus solving the problems of existing technologies.

[0006] In a first aspect, the present invention provides a method for suppressing background in X-ray images for foreign object particle detection in GIS equipment, comprising the following steps: The image to be tested is initially filtered in the frequency domain to extract candidate particles. After filtering out high frequencies, the remaining structural contours are strongly correlated with the pre-built background template library. Matching is performed at the feature layer to output the best matching background image. The best matching background is subtracted from the original image, and a second frequency domain filter is performed. The intersection of the filter with the candidate particles extracted after the first filter is taken to output the final candidate particle detection result.

[0007] Furthermore, the constructed background template library includes: Multiple sets of GIS images without foreign objects were collected, features were extracted, and a background feature library was constructed after grayscale statistical modeling. Details are as follows: For the i-th image b_i(x,y) in the background library, extract its feature vector mathbf{v}_i, and the gray-level histogram statistics of the image are: ; in, The grayscale mean is... Standard deviation To measure the skewness of the distribution asymmetry, Kurtosis is used to measure the steepness of a distribution.

[0008] Furthermore, the filtering of high frequencies from the image under test specifically includes: The internal conductors and shell edges of GIS appear as smooth curves in the spatial domain. High-frequency filtering removes abrupt changes and sharp edges corresponding to foreign particles. That is, the energy distribution of abrupt changes and sharp edges in the image is in the high frequency range. The low-frequency background structure of the test image after high-frequency filtering is obtained and matched with the background library image to find the best matching background with high confidence.

[0009] Furthermore, the process of filtering out high frequencies from the image under test specifically includes: Construct a mathematical model, F(u,v) = mathcal{F}{f(x,y)}, and design a Gaussian high-pass filter after Fourier transform: H_{hp}(u,v): ; Where D(u,v) is the distance from the frequency domain point (u,v) to the center, and D_0 is the cutoff frequency; Frequency domain filtering is performed as follows: F_h(u,v) = F(u,v) \cdot H_{hp}(u,v); Then, the inverse transformation is performed back to the spatial domain, f_h(x,y) = \mathcal{F}^{-1}{F_h(u,v)}, where f_h(x,y) is the high-frequency enhanced image.

[0010] Furthermore, the step of structurally strongly correlating the remaining structural contours with the pre-built background template library specifically includes: calculating the normalized cross-correlation coefficient between f_h(x,y) and each template b_i(x,y) in the background library as follows: ; Select the background template that maximizes the normalized cross-correlation coefficient as the best matching background b*(x,y).

[0011] Furthermore, subtracting the best-matching background from the original image specifically includes: The model after subtracting the best-matching background is: ; The absolute value is used to preserve the amplitude information of grayscale changes. The resulting difference image d(x,y) contains non-zero values ​​for foreign particles and random noise.

[0012] Furthermore, the second frequency domain filtering also specifically includes: For the differential image, the isolated bright / dark spots corresponding to high frequencies are foreign particles, and the residual background generated based on matching error or ray scattering interference corresponds to low-frequency cloud-like shadows. Then, through frequency domain filtering, the particle signal is separated from d(x,y). Fourier transform: D(u,v) = \mathcal{F}{d(x,y)}; Bandpass or highpass filter: D_f(u,v) = D(u,v) \cdot H_{bp}(u,v), where H_{bp} is designed as a Gaussian bandpass filter with its center frequency aligned to the frequency band corresponding to the expected particle size; Inverse transform: d_f(x,y) = \mathcal{F}^{-1}{D_f(u,v)}, where d_f(x,y) is the denoised particle-enhanced image.

[0013] Furthermore, the intersection of the candidate particles extracted after the initial filtering with the candidate particles is used to output the final candidate particle detection result, specifically including: The intersection of the initial filtered high-frequency detection result f_h(x,y) and the differential high-frequency detection result d_f(x,y) is used to obtain the pixel that points to the same location. The specific operation is as follows: Binarization and Logic Operations: ; Where Thresh represents threshold segmentation, and T is a threshold constant. This indicates that the input image is binarized using a threshold T. right For each connected region k in the array, the following calculation is performed: Area: A_k = \sum_{(x,y) \in R_k} 1; Average contrast: C_k = \frac{1}{A_k} \sum_{(x,y) \in R_k} d(x,y).

[0014] The present invention provides an X-ray image background suppression method for foreign object particle detection in GIS equipment, which improves the structure-background suppression ratio and enhances the sensitivity to small foreign objects. By fusing the intersection of two paths, the false alarm rate is reduced and the robustness of the algorithm is improved. Attached Figure Description

[0015] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and constitute a part of this invention, are not intended to limit the scope of the invention. In the drawings: Figure 1A flowchart of an X-ray image background suppression method for foreign particle detection in GIS equipment, provided as an exemplary embodiment of the present invention. Detailed Implementation

[0016] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention.

[0017] Technical concept of the present invention: Considering frequency domain filtering—the Curvelet transform—the principle is to transform an image into the frequency domain for processing by utilizing the differences in signals at different scales. Among them, the Curvelet transform is particularly suitable for enhancing and separating grain edges due to its ability to sparsely represent curve edges. It can effectively enhance contrast and suppress background noise, and its effect is better than traditional histogram equalization and homomorphic filtering. Combining image difference method and background modeling method, a background library without foreign objects is pre-constructed (background modeling). The image to be tested is first initially extracted by frequency domain filtering to extract candidate regions of particles, and then converted back to the spatial domain and matched with the background library image (template matching). The most similar background is selected for difference, and the difference result is then finely filtered by frequency domain filtering. Finally, the intersection with the frequency domain result of the first step is taken, and the different influence levels are superimposed and statistically analyzed.

[0018] The X-ray image background suppression method for foreign particle detection in GIS equipment provided by this invention aims to solve the above-mentioned technical problems in the prior art.

[0019] The technical solution of the present invention and how the technical solution of the present invention solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of the present invention will now be described with reference to the accompanying drawings.

[0020] Example 1: This invention comprises two stages: offline background feature library construction and online foreign object detection and quantification, such as... Figure 1 As shown; Phase 1: Construction of Offline Background Feature Library Step 1: Acquisition of foreign object-free template image and extraction of grayscale features; Because GIS equipment imaging is affected by factors such as the X-ray source location, tube voltage, and detector angle, there are subtle differences in the background grayscale distribution. In order to accurately match the background during subsequent online inspection, it is necessary to pre-collect multiple sets of GIS equipment X-ray images under different typical operating conditions, which have been manually confirmed to be free of any foreign particles, and construct a background template library.

[0021] For the i-th background template image b_i(x,y), its feature vector is defined as mathbf{v}_i, where b_i(x,y): the gray value of the i-th background image without foreign objects at the spatial coordinates (x,y); \mathbf{v}_i: A vector representing the statistical characteristics of the gray-level distribution of the image, defined as: ; Where, \mathbf{v}_i is the background feature vector, which is a four-dimensional real vector; = \frac{1}{N}\sum_{x,y} b_i(x,y) is the mean gray level, reflecting the overall brightness level. = \sqrt{\frac{1}{N}\sum_{x,y}(b_i(x,y) - \mu_i)^2} represents the grayscale standard deviation, reflecting the image contrast. = \frac{\frac{1}{N}\sum(b_i(x,y)-\mu_i)^3}{\sigma_i^3} is the skewness that measures the asymmetry of the distribution, reflecting the asymmetry of the gray-level distribution. \frac{\frac{1}{N}\sum (b_i(x,y)-\mu_i)^4}{\sigma_i^4} represents the kurtosis, which measures the steepness of the distribution and reflects the steepness of the gray-level distribution. N is the total number of pixels in the image. Step 1 establishes a background feature library containing diverse imaging conditions, providing benchmark data for robust matching in the subsequent step 3 and avoiding differential residue caused by minor fluctuations in imaging conditions. Phase Two: Online Foreign Object Detection and Quantification Step 2: High-pass filtering of the image under test in the frequency domain to initially extract candidate high-frequency components; The smooth edges of the conductors inside the GIS correspond to low-frequency components in the frequency domain, while the sharp edges of foreign particles correspond to high-frequency components. By directly suppressing the low-frequency background in the frequency domain using a high-pass filter, the signal-to-noise ratio (SNR) of the foreign particles can be significantly improved. Fourier transform: Transforms the image f(x,y) to be measured from the spatial domain to the frequency domain; Where, \mathcal{F}: two-dimensional discrete Fourier transform operator. (u,v): frequency domain coordinates, in units of period / image width, representing spatial frequency. F(u,v): frequency domain complex matrix, containing amplitude spectrum and phase spectrum.

[0022] Gaussian high-pass filter design: ; Where D(u,v) = \sqrt{(u - u_c)^2 + (v - v_c)^2}: the Euclidean distance from the frequency domain point (u,v) to the center frequency (u_c, v_c). D_0: the cutoff frequency, a key hyperparameter. When D = D_0, the filter gain is approximately 1 - e^{-0.5} \approx 0.39. Frequencies below D_0 are significantly attenuated, while frequencies above D_0 are preserved.

[0023] Frequency domain filtering and inverse transform: , , • cdot: matrix dot product operation; f_h(x,y): high-frequency image enhancement. Smooth conductor backgrounds are suppressed to dark colors, while foreign particles and conductor edges appear as bright outlines. Technical effect: Improved contrast-to-noise ratio (CNR) of foreign particles. At this point, all possible small structures that could be foreign objects in the image are highlighted, but some high-frequency components of background edges are also included, requiring further identification in subsequent steps.

[0024] Step 3: Background template matching based on high-frequency structures; Directly matching the original image f(x,y) with the background library b_i(x,y) is susceptible to interference from overall brightness shifts. However, f_h(x,y) filters out low-frequency brightness information and retains only structural high-frequency edges. By performing similarity matching in this feature space, the background template that best matches the structural contour of the current image under test can be found. The normalized cross-correlation coefficient between f_h(x,y) and each template b_i(x,y) in the background library is: ; Select the background template that maximizes the normalized cross-correlation coefficient as the best matching background b*(x,y); Where, \bar{f}_h: the pixel mean of the high-frequency enhanced image f_h(x,y).

[0025] \bar{b}_i: The average pixel value of the background template b_i(x,y).

[0026] \gamma_i \in [-1, 1]: The closer to 1, the more similar the structures.

[0027] Select the best matching background template b^*(x,y): ; This step achieves subpixel-level virtual registration. Even if the device has slight displacement or jitter during actual imaging, NCC matching can select the background template with the closest position and rotation angle, providing a high-quality benchmark for subsequent differential calculations. Step 4: Spatial domain image differencing to eliminate structural background; By subtracting the best-matched background template from the original image, large, regular structures such as GIS equipment casings and conductors can theoretically be eliminated, leaving only localized grayscale variations caused by foreign particles. The absolute value is used to preserve the amplitude information of grayscale changes. That is, foreign objects may appear as spots that are brighter (positive difference) or darker (negative difference) than the background. The absolute value is used to uniformly preserve the amplitude of changes. In the obtained difference image d(x,y), the foreign object particles and random noise contain non-zero values.

[0028] Step 5: Fine-filter the difference image in the frequency domain to separate residual artifacts from real foreign objects; Specifically, residual artifacts in differential images typically exhibit a slowly changing "cloud-like" distribution, representing low-frequency interference; while real foreign particles appear as isolated, extremely small bright / dark spots, representing high-frequency signals. Frequency domain filtering is then used for separation. Fourier transform: D(u,v) = \mathcal{F}{d(x,y)}; Gaussian high-pass filter: D_f(u,v) = D(u,v) \cdot H_{bp}(u,v), where H_{bp} is designed as a Gaussian bandpass filter with its center frequency aligned to the frequency band corresponding to the expected particle size; Alternatively, Gaussian bandpass filtering (selectable depending on the video segment): ; Where D_1: Low-frequency cutoff frequency, used to suppress artifacts (typically D_1) <D_2); D_2: High-frequency cutoff frequency, used to suppress extremely high frequency detector electronic noise; Filtering and inverse transform: , , The resulting d_f(x,y) is called a clean grain-enhanced image. Residual cloud-like artifacts are largely removed, leaving only clean, well-defined specks of foreign particles in the image. Step 6: Dual-path decision-level fusion and quantitative statistics.

[0029] Pathway 1 (f_h) has high sensitivity but is prone to false alarms (misclassifying edges as particles), while Pathway 2 (d_f) has high specificity but may miss detections (if the match is poor). A logical AND (intersection) operation is performed on the results of the two pathways; only areas jointly confirmed by both are identified as foreign objects, which can greatly reduce the false alarm rate. Finally, geometric and grayscale features are extracted from the confirmed areas to quantify the risk. Binarization and Logic Operations: ; Where Thresh represents threshold segmentation, and T is a threshold constant. This function performs binarization of the input image with a threshold T. If the output pixel value is greater than T, it is set to 1; otherwise, it is set to 0. T_1 and T_2 are the detection thresholds for the two channels, respectively. \land is a logical AND operation. Connectivity analysis and feature quantization operations: For the k-th connected region R_k (foreign particle) in M_{final}, calculate: area (measures physical size): A_k = \sum_{(x,y) \in R_k} 1, average contrast (measures material density difference): C_k = \frac{1}{A_k} \sum_{(x,y) \in R_k} d(x,y); Output a structured detection report, including the number of foreign objects (sum_k) and the center coordinates (x_k, y_k) of each foreign object.

[0030] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0031] Furthermore, in the embodiments of the present invention, the functional modules can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or in the form of hardware plus software functional modules.

[0032] Those skilled in the art will understand that embodiments of the present invention can be provided as methods or systems. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects.

[0033] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. 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 apparatus that includes that element.

[0034] The above are merely embodiments of the present invention and are not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the claims of the present invention.

[0035] Other embodiments of the invention will readily occur to those skilled in the art upon consideration of the invention disclosed herein in the specification and examples. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of the invention are indicated by the foregoing claims.

[0036] It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A method for suppressing background in X-ray images for foreign particle detection in GIS equipment, characterized in that, Includes the following steps: The image to be tested is initially filtered in the frequency domain to extract candidate particles. After filtering out high frequencies, the remaining structural contours are strongly correlated with the pre-built background template library. Matching is performed at the feature layer to output the best matching background image. The best matching background is subtracted from the original image, and a second frequency domain filter is performed. The intersection of the filter with the candidate particles extracted after the first filter is taken to output the final candidate particle detection result.

2. The X-ray image background suppression method for foreign particle detection in GIS equipment according to claim 1, characterized in that, in, The background template library mentioned above includes: Multiple sets of GIS images without foreign objects were collected, features were extracted, and a background feature library was constructed after grayscale statistical modeling. Details are as follows: For the i-th image b_i(x,y) in the background library, extract its feature vector mathbf{v}_i, and the gray-level histogram statistics of the image are: ； in, The grayscale mean is... Standard deviation To measure the skewness of the distribution asymmetry, Kurtosis is used to measure the steepness of a distribution.

3. The X-ray image background suppression method for foreign particle detection in GIS equipment according to claim 2, characterized in that, in, The high-frequency filtering of the image under test specifically includes: The internal conductors and shell edges of GIS appear as smooth curves in the spatial domain. High-frequency filtering removes abrupt changes and sharp edges corresponding to foreign particles. That is, the energy distribution of abrupt changes and sharp edges in the image is in the high frequency range. The low-frequency background structure of the test image after high-frequency filtering is obtained and matched with the background library image to find the best matching background with high confidence.

4. The X-ray image background suppression method for foreign particle detection in GIS equipment according to claim 3, characterized in that, The process of filtering out high frequencies from the image under test specifically includes: Construct a mathematical model, F(u,v) = mathcal{F}{f(x,y)}, and design a Gaussian high-pass filter after Fourier transform: H_{hp}(u,v): ; Where D(u,v) is the distance from the frequency domain point (u,v) to the center, and D_0 is the cutoff frequency; Frequency domain filtering is performed as follows: F_h(u,v) = F(u,v) \cdot H_{hp}(u,v); Then, the inverse transformation is performed back to the spatial domain, f_h(x,y) = \mathcal{F}^{-1}{F_h(u,v)}, where f_h(x,y) is the high-frequency enhanced image.

5. The X-ray image background suppression method for foreign particle detection in GIS equipment according to claim 4, characterized in that, The step of structurally strongly correlating the remaining structural contours with the pre-built background template library specifically includes: the normalized cross-correlation coefficient between f_h(x,y) and each template b_i(x,y) in the background library is: ； Select the background template that maximizes the normalized cross-correlation coefficient as the best matching background b*(x,y).

6. The X-ray image background suppression method for foreign particle detection in GIS equipment according to claim 5, characterized in that, The step of subtracting the best-matching background from the original image specifically includes: The model after subtracting the best-matching background is: ； The absolute value is used to preserve the amplitude information of grayscale changes. The resulting difference image d(x,y) contains non-zero values ​​for foreign particles and random noise.

7. The X-ray image background suppression method for foreign particle detection in GIS equipment according to claim 6, characterized in that, The frequency domain secondary filtering also specifically includes: For the differential image, the isolated bright / dark spots corresponding to high frequencies are foreign particles, and the residual background generated based on matching error or ray scattering interference corresponds to low-frequency cloud-like shadows. Then, through frequency domain filtering, the particle signal is separated from d(x,y). Fourier transform: D(u,v) = \mathcal{F}{d(x,y)}; Bandpass or highpass filter: D_f(u,v) = D(u,v) \cdot H_{bp}(u,v), where H_{bp} is designed as a Gaussian bandpass filter with its center frequency aligned to the frequency band corresponding to the expected particle size; Inverse transform: d_f(x,y) = \mathcal{F}^{-1}{D_f(u,v)}, where d_f(x,y) is the denoised particle-enhanced image.

8. The X-ray image background suppression method for foreign particle detection in GIS equipment according to claim 7, characterized in that, The intersection of the candidate particles extracted after the initial filtering with the candidate particles extracted after the initial filtering is used to output the final candidate particle detection result, which specifically includes: The intersection of the initial filtered high-frequency detection result f_h(x,y) and the differential high-frequency detection result d_f(x,y) is used to obtain the pixel that points to the same location. The specific operation is as follows: Binarization and Logical Operations: ； Where Thresh represents threshold segmentation, and T is the threshold constant. This indicates that the input image is binarized using a threshold T. right For each connected region k in the equation, the following calculation is performed: Area: A_k = \sum_{(x,y) \in R_k} 1; Average contrast: C_k = \frac{1}{A_k} \sum_{(x,y) \in R_k} d(x,y).

Images