Power transformation equipment image processing method based on improved dual-light fusion

Abstract

The invention discloses a power transformation equipment image processing method based on improved dual-light fusion. The method comprises the following steps: a, judging whether the environment illumination intensity is a strong light environment or a weak light environment; b, enhancing the contrast ratio of the infrared light image; c, preliminarily obtaining an equipment contour on the two images; d, respectively preprocessing the visible light image and the infrared light image by means of an improved Canny edge detection algorithm to obtain an accurate equipment edge; e, coding and marking, and generating an edge link; f, processing the image size and the rotation angle of the visible light image or the infrared light image; and g, aligning the visible light image and the infrared light image processed in the above steps through the edge link generated in the step e, and combining the visible light image and the infrared light image into an equipment image with a heating condition. According to the method, the two images are fused in the strong light environment, so that the fault defect can be identified more accurately during subsequent equipment inspection and detection.

Description

Technical Field

[0001] The present invention belongs to the technical field of power equipment inspection, and relates to an intelligent image processing technology when detecting thermal defects of power equipment, in particular to an image processing method for substation equipment based on improved dual-light fusion. Background Art

[0002] In the power system, infrared thermal imaging technology has been widely used for live detection of substation equipment. Infrared thermal imagers can scan and detect hot spots of substation equipment without power outages, and find possible defects in the equipment through the distribution of temperature anomalies. However, the effectiveness of infrared thermal imaging is limited in some cases. For example, in environments with complex backgrounds such as high reflectivity or direct sunlight, infrared images may be misjudged. In addition, infrared images usually have low resolution and lack sufficient details to accurately locate equipment defects.

[0003] On the other hand, visible light images provide rich visual information and can clearly show the shape, structure and surface features of the device. However, the effectiveness of visible light images will be significantly reduced when the ambient light conditions are poor, such as at night or under smoke. Detection methods based on a single mode have problems such as insufficient information and high misjudgment rate, where a single mode refers to infrared light or visible light.

[0004] Therefore, fusing visible light images with infrared images and utilizing the complementary advantages of the two images is an important means to improve detection accuracy and robustness. Although the existing dual-light fusion technology can improve the detection effect to a certain extent, it still has some shortcomings. For example, traditional fusion methods usually use simple image superposition or weighted averaging, which makes it difficult to effectively align feature points in images of different modalities, resulting in low quality of fused images. In addition, the existing technology has poor adaptability when dealing with different environmental conditions such as strong light or weak light, which can easily lead to image fusion failure or information loss. Summary of the invention

[0005] In order to improve the accuracy of the fusion alignment of visible light images and infrared images, the present invention designs an image processing method for substation equipment based on improved dual-light fusion, which performs a series of processing on the dual-light images before alignment and fusion, thereby ensuring the accuracy after alignment and fusion.

[0006] The technical solution adopted by the present invention is a method for processing substation images based on improved dual-light fusion, which is a process of synchronously photographing and processing a local area of ​​a substation through an infrared camera and a visible light camera to align the device contours and fuse them into a device image with a heating condition. The steps of the image processing method are as follows:

[0007] a. Determine whether the ambient light intensity is a strong light environment or a weak light environment;

[0008] In strong light conditions,

[0009] b. Enhance the contrast of infrared images;

[0010] c. Use joint bilateral filtering to process the infrared image and the visible light grayscale image processed in step a, retain the parts of the two images with the same weighted grayscale value weights, delete the rest, and preliminarily obtain the device outline on the two images;

[0011] d. Preprocess the visible light image and infrared light image separately with the help of the improved Canny edge detection algorithm, compare the gradient of each pixel with the surrounding pixels in at least 8 directions to obtain the precise device edge;

[0012] e. Encode and mark the points on the device edge of the visible light image and the infrared light image respectively, and generate edge links;

[0013] f. Use SURF algorithm and ORB algorithm to process the image size and rotation angle of visible light image or infrared light image;

[0014] g. Align the visible light image and the infrared light image processed in the above steps through the edge links generated in step e, and merge them into a device image with heating conditions.

[0015] In the step a, the determination of the ambient light intensity is performed by setting different times corresponding to different lighting environments.

[0016] In a strong light environment, the visible light image is processed to decide whether to switch to a low light environment.

[0017] The steps of processing the visible light image are: converting the visible light image into the HSV color space, extracting the brightness value V in the HSV color space and calculating the average value, comparing the average value with the brightness threshold, and then deciding whether to switch to a low-light environment.

[0018] In the step b, the specific method for enhancing the contrast of the infrared light image is: extracting the brightness value in the brightness channel in the HSV color space of the visible light image, and matching the brightness value with the histogram of the infrared image.

[0019] In the step f, the method of performing image size processing on the visible light image or the infrared light image by means of the SURF algorithm is: using box filters of different sizes to perform convolution on the image to achieve image enlargement and reduction.

[0020] The beneficial effects of the present invention are:

[0021] 1. The present invention realizes high-precision alignment and fusion of visible light and infrared images through an improved dual-light fusion algorithm. The improved edge detection and feature matching algorithms greatly improve the clarity and detail retention of the fused image.

[0022] 2. Adaptive edge detection: The adaptive Canny edge detection algorithm introduced automatically adjusts the edge detection threshold according to the local features of the image (such as contrast and brightness), so that the device edge information can be accurately extracted under different lighting and background conditions. This function greatly enhances the robustness of the system and enables it to adapt to various complex environments.

[0023] 3. Multimodal feature matching optimization: By using improved SURF and ORB algorithms, the key feature points in visible light and infrared images are ensured to be highly consistent. This optimized feature matching technology not only improves the accuracy of image fusion, but also provides a stable foundation for subsequent target recognition.

[0024] 4. Versatility and scalability: The image processing method proposed in this invention can be widely used in various detection scenarios of substation equipment, not just in the detection of heating defects. The algorithm framework has strong versatility and scalability, can adapt to different detection requirements, and provides a solid technical foundation for smart grid equipment detection. DETAILED DESCRIPTION

[0025] When the present invention is implemented, when a detection device with an infrared camera and a visible light camera is manually held or a robot is used for directional inspection, the two cameras take pictures of one device at the same time to generate an infrared image and a visible light image with thermal rendering. Finally, the two images are merged into one image to reflect the thermal distribution of the device, which is then used to detect whether the device has faults or defects.

[0026] After obtaining the two images, firstly, the sunrise and sunset times are determined by obtaining the weather information of the day as a reference for switching between strong and weak light environments. If the current time is a weak light environment, the visible light image is not clear due to the dim light, and it cannot be processed and fused for the infrared image. Therefore, in a weak light environment, only the infrared image is used for detection when detecting the device. The focus of the present invention is that the visible light image is very clear in a strong light environment, which provides the basic conditions for alignment and fusion. In a strong light environment, it may be cloudy or the light may suddenly dim, but whether to switch to a weak light environment depends on the real-time acquisition of the light intensity signal by the sensitive sensor, and the decision to switch to a weak light environment is made by processing the visible light image. The software converts the visible light image after the light dims into the HSV color space, extracts the brightness value V in the HSV color space and calculates the average value, compares the average value with the brightness threshold, and then decides whether to switch to a weak light environment. For example, if the average V value is lower than 0.3, the brightness threshold can be adjusted according to the actual situation, and then the weak light environment is judged and switched; if it is higher than 0.7, the strong light environment is maintained.

[0027] Before specific alignment and fusion, the contrast of the infrared image must be adjusted first. Infrared images are usually single-channel grayscale images. When processing or displaying, pseudo-color techniques, such as heat maps, are usually applied to map grayscale values ​​to colors: low grayscale values ​​are mapped to cold colors, such as blue or purple, high grayscale values ​​are mapped to warm colors, such as red or yellow, and intermediate grayscale values ​​are mapped to green, orange, etc. This pseudo-color mapping can intuitively show the difference in temperature or radiation intensity, and finally display it as a three-channel color image. However, infrared images usually have low contrast, especially when the temperature difference is small, which causes their edge details to be less obvious than visible light images. When applying Canny edge detection, the edges of infrared images are often not clear enough and the detection effect is poor. Therefore, we need to enhance the contrast of the infrared image first. The specific method is to extract the brightness channel in the HSV color space of the visible light image taken at this moment, extract the brightness value, and match the infrared image with this brightness value in the form of a histogram. After matching the brightness through the infrared image histogram, the grayscale distribution of the infrared image can be adjusted to make it closer to the visible light image. This adjustment can stretch the brightness dynamic range of the infrared image, making its contrast and brightness characteristics close to those of the visible light image. This can reduce the difference in brightness and contrast between the two images, so that when using Canny edge detection later, a more consistent threshold can be applied, improving the detection effect of weak edges, enhancing the effect of image fusion, and ultimately improving the accuracy and robustness of detection.

[0028] Next, the contour of the infrared image is extracted, and the image is smoothed using joint bilateral filtering to retain edge features. First, the visible light image is converted into a grayscale image, and then the two images are weighted according to the grayscale value similarity and spatial relationship. Although the grayscale values ​​of the weighted images are inconsistent, their relative grayscale order is consistent. Since the heating area in the infrared image usually corresponds to a higher grayscale value, while the non-heating area is relatively low, we can retain the characteristics of the heating area by setting an appropriate threshold and preliminarily obtain the rough outline of the device.

[0029] Next, the edge of the device is detected and extracted using the Canny edge detection algorithm. The traditional Canny edge detection algorithm only detects 4 direction points outside a pixel at most, namely 0°, 45°, 90°, and 135°. This method improves the direction to at least 8 directions, that is, the brightness and contrast changes must be compared in at least 8 directions on the circumference of a pixel. This method can improve the accuracy of edge detection, especially when processing infrared images with large noise. The NMS suppression threshold is dynamically adjusted for different edge strengths. Strong edges use stricter suppression standards, while weak edges relax the suppression standards to retain more potential edge information. This step can obtain the precise outline of the device.

[0030] After obtaining the precise outlines of the equipment in the visible light image and infrared image, alignment cannot be performed directly, because there may be multiple identical parts on an electric power equipment. If no processing is done, the software may be broken or misaligned after alignment. For example, a device contains three identical transformer bushings. During alignment, bushing A phase in the infrared image is aligned with bushing C phase in the visible light image. This is a misalignment phenomenon. Therefore, contour edge linking is required. Encode and mark the points on the edge of the equipment in the visible light image and infrared image respectively, and generate edge links.

[0031] Since the focal lengths and angles facing the device of the two cameras are different, the ratios and offset angles of the two images are different, so the image size and angle calibration must be performed before alignment. The present invention uses the SURF algorithm to achieve image zooming and the ORB algorithm to achieve image angle rotation adjustment.

[0032] Image zooming is not a physical scaling of the image itself, but is achieved by changing the filter scale. As the filter scale increases, the algorithm can capture large structural information in the image, similar to the effect of a reduced image. As the filter scale decreases, the algorithm captures details in the image, similar to the effect of a magnified image. SURF uses box filters of different sizes to approximate Gaussian filters and convolve the image. This filter design is more computationally efficient than traditional Gaussian filters, greatly speeding up the operation.

[0033] The effects of zooming in and out are reflected in the scale space constructed by SURF. By applying filters to images at different scales, SURF simulates different scaling ratios. The adaptive Gaussian filter in the algorithm dynamically adjusts the scale of the filter according to the complexity of the local texture of the image, thereby ensuring that feature points can be stably detected in areas of different complexity.

[0034] Finally, during alignment, the infrared image and the visible light image are matched through the same coding mark to achieve alignment of the edge contour.

[0035] By fusing the two images in a strong light environment, fault defects can be identified more accurately during subsequent equipment inspections.

Claims

1. A method for processing substation equipment images based on improved dual-light fusion is a process of synchronously photographing and processing a local area of ​​a substation equipment through an infrared camera and a visible light camera to align the equipment contours and fuse them into an equipment image with heating conditions, characterized in that: The steps of the image processing method are as follows: a. Determine whether the ambient light intensity is a strong light environment or a weak light environment; In strong light conditions, b. Enhance the contrast of infrared images; c. Use joint bilateral filtering to process the infrared image and the visible light grayscale image processed in step a, retain the weighted grayscale value parts of the two images with the same weight, delete the rest, and preliminarily obtain the device outline on the two images; d. Preprocess the visible light image and infrared light image separately with the help of the improved Canny edge detection algorithm, compare the gradient of each pixel with the surrounding pixels in at least 8 directions to obtain the precise device edge; e. Encode and mark the points on the device edge of the visible light image and the infrared light image respectively, and generate edge links; f. Use SURF algorithm and ORB algorithm to process the image size and rotation angle of visible light image or infrared light image; g. Align the visible light image and the infrared light image processed in the above steps through the edge links generated in step e, and merge them into a device image with heating conditions.

2. According to claim 1, a substation equipment image processing method based on improved dual-light fusion is characterized in that: In the step a, the determination of the ambient light intensity is performed by setting different times corresponding to different lighting environments.

3. According to claim 2, a method for processing substation images based on improved dual-light fusion, characterized in that: In a strong light environment, the visible light image is processed to decide whether to switch to a low light environment.

4. According to claim 3, a method for processing substation images based on improved dual-light fusion is characterized in that: The steps of processing the visible light image are: converting the visible light image into the HSV color space, extracting the brightness value V in the HSV color space and calculating the average value, comparing the average value with the brightness threshold, and then deciding whether to switch to a low-light environment.

5. According to claim 1, a method for processing substation images based on improved dual-light fusion, characterized in that: In the step b, the specific method for enhancing the contrast of the infrared light image is: extracting the brightness value in the brightness channel in the HSV color space of the visible light image, and matching the brightness value with the histogram of the infrared image.

6. According to claim 1, a method for processing substation images based on improved dual-light fusion, characterized in that: In the step f, the method of performing image size processing on the visible light image or the infrared light image by means of the SURF algorithm is: using box filters of different sizes to perform convolution on the image to achieve image enlargement and reduction.