Transformer substation fire detection method and system based on feature recognition

Abstract

The invention provides a transformer substation fire detection method and system based on feature recognition, and belongs to the technical field of transformer substation safety. The method comprises the following steps: acquiring an infrared thermal image of a transformer substation; preprocessing the infrared thermal image, and identifying a thermal region from the preprocessed image; judging whether the heating area of the infrared thermal image has a fault or not by using a relative temperature difference method, and if so, processing the infrared thermal image to obtain HSI color features; segmenting a fire area image subset from the infrared thermal image according to HSI color features; in combination with the flame texture and the edge features, a sample entropy signal processing algorithm is adopted to obtain a sample entropy value of the fire area image subset; and determining whether a fire occurs according to the sample entropy. The relative temperature difference of the transformer substation is obtained according to the infrared thermal image, whether the image is detected in the next step or not is determined according to the relative temperature difference, and the system does not need to run completely; according to the invention, the processing amount of data can be reduced, and the fire identification precision is high.

Description

Technical Field

[0001] The present invention belongs to the technical field of substation safety, and in particular relates to a substation fire detection method and system based on feature recognition. Background Art

[0002] At present, substations are developing towards automation and unmanned operation. Fire monitoring is an effective way to manage fire safety in substations. In addition to temperature, light and smoke sensing functions, most substation fire protection equipment also has a "remote viewing" function. By processing remotely transmitted images, it is possible to detect whether there is a fire in a timely manner. It is a technology that can identify the category of the target in the image through remote analysis, induction and summary.

[0003] In the prior art, fire identification is generally performed directly on the collected images through algorithms. However, under normal circumstances, the probability of fire occurring is relatively small. If the system detects the collected images from beginning to end each time, the method needs to process a large amount of data, especially when temperature, light, smoke, etc. are present at the same time, the data processing amount is even greater. Summary of the invention

[0004] In view of the deficiencies of the prior art, the present invention provides a substation fire detection method and system based on feature recognition.

[0005] In order to solve one, part or all of the above technical problems, the technical solution adopted by the present invention is: A substation fire detection method based on feature recognition comprises: collecting infrared thermal images of the substation; preprocessing the infrared thermal images and identifying heating areas from the preprocessed images; judging whether there is a fault in the heating area of ​​the infrared thermal images by using a relative temperature difference method, and if there is a fault, processing the infrared thermal images to obtain HSI color features; segmenting a fire area image subset from the infrared thermal images according to the HSI color features; combining flame texture and edge features, using a sample entropy signal processing algorithm to obtain a sample entropy value of the fire area image subset; judging whether a fire occurs according to the sample entropy value.

[0006] Furthermore, the method for preprocessing the infrared thermal image includes: gray-scaling and denoising the infrared thermal image, and then performing histogram equalization.

[0007] Furthermore, the method for identifying the heating area from the preprocessed image includes: using the region growing method, extracting the middle point of the area with the maximum gray value from the preprocessed image as a seed point, performing growth and corrosion, and then performing edge extraction to obtain the heating area.

[0008] Furthermore, the method of using the relative temperature difference method to determine whether there is a fault in the heating area of ​​the infrared thermal image includes: obtaining the relative temperature difference of the heating area, and if the relative temperature difference is not less than 50°C, it is determined that there is a fault in the heating area; the relative temperature difference , where T1 is the temperature of the heating area, T0 is the ambient temperature, and T2 is the temperature of the heating area under normal conditions.

[0009] Furthermore, the method of processing the infrared thermal image to obtain the HSI color feature includes: obtaining the RGB color feature of the infrared thermal image, normalizing the RGB color feature and converting it into the HSI color feature.

[0010] Furthermore, the method for segmenting the fire area image subset from the infrared thermal image includes: using the K-means clustering algorithm, according to the HSI color features of the infrared thermal image, the infrared thermal image is segmented into different types of image subsets according to the types of substation facilities, sky, site, flame and flame irradiation area, wherein the image subset corresponding to the flame is the fire area image subset.

[0011] Furthermore, the K-means algorithm uses Euclidean distance to measure the similarity between samples.

[0012] Furthermore, the method of obtaining the sample entropy value of the fire area image subset by using the sample entropy signal processing algorithm includes: Reduce the fire area image subset to a time series of length N , 1≤n≤N; Construct a set of m-dimensional space vectors Y(i) according to the comparison dimension, , i≤N-m+1; Define the maximum difference between vectors Y(i) and Y(j) , 1≤j≤N-m+1, 0≤k≤m-1; Set the tolerance r and calculate the value for each i that meets The number of , calculate the ratio of this number to the total number of distances , ; Calculate the average ratio of all i , ; Increase the dimension, let m become m+1, repeat the above steps and finally get the sample entropy .

[0013] Furthermore, the method for determining whether a fire has occurred based on a sample entropy value includes: if the sample entropy value is greater than a preset value, determining that a fire has occurred.

[0014] A substation fire detection system based on feature recognition comprises an image processing module, a fault discrimination module and a fire judgment module; the image processing module is used to collect infrared thermal images of the substation, to preprocess the infrared thermal images, and to identify the heating area from the preprocessed images; the fault discrimination module is used to judge whether there is a fault in the heating area by using a relative temperature difference method, and to process the infrared thermal image to obtain HSI color features when there is a fault; the fire judgment module is used to segment a fire area image subset from the infrared thermal image according to the HSI color features and using a K-means algorithm, to combine flame texture and edge features, and to obtain a sample entropy value of the fire area image subset by using a sample entropy signal processing algorithm, and to judge whether a fire occurs according to the sample entropy value.

[0015] Compared with the prior art, the present invention has the following beneficial effects: The present invention first obtains the relative temperature difference of the substation according to the infrared thermal image, and determines whether to perform the next step of detection on the image according to the relative temperature difference, so that the system does not have to run completely, thereby reducing the amount of data processing. At the same time, according to the color difference in the infrared thermal image, the characteristics of the flame are characterized according to the HSI of the substation and clustering is performed using the Euclidean distance, avoiding the influence of local noise, reducing the missed diagnosis rate and the misdiagnosis rate while shortening the recognition time.

[0016] The present invention performs grayscale processing and denoising on the infrared thermal image, and then performs histogram equalization to enhance the image contrast. Subsequently, the region growing method is used to extract the middle point of the area with the maximum grayscale value from the preprocessed image as a seed point, and then edge extraction is performed after growth and corrosion to obtain the heating area, thereby ensuring the accuracy of the extracted target, facilitating the acquisition of the relative temperature difference, and expressing the temperature through the relative temperature difference, with less influence from the ambient temperature.

[0017] The present invention can determine whether all pixel points of the original infrared thermal image are within the characteristic range of the flame area by extracting the HSI features of the original infrared thermal image. If so, corresponding markings are performed. This takes into account more features of the pixel points and can avoid the pixel points in the image from falling into local segmentation, and has strong robustness.

[0018] The present invention adopts sample entropy value to distinguish fire, so as to prevent the problem that objects that may contain hue, saturation and brightness similar to flame, such as fire cabinets, fire pipes, red signs, etc., are marked as fire areas in the substation, thereby improving the detection accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The present invention is further described in detail below in conjunction with the accompanying drawings.

[0020] Figure 1 : Flow chart of embodiment 1 of the present invention; Figure 2: Schematic diagram of embodiment 2 of the present invention. DETAILED DESCRIPTION

[0021] In order to better understand the present invention, the content of the present invention is further clearly described below in conjunction with the embodiments and the accompanying drawings, but the protection content of the present invention is not limited to the following embodiments. In the following description, a large number of specific details are provided to provide a more thorough understanding of the present invention. However, it is obvious to those skilled in the art that the present invention can be implemented without one or more of these details.

[0022] Example 1: See Figure 1 ,The purpose of this embodiment is to provide a substation fire detection method based on ,feature recognition.

[0023] S1. Collect infrared thermal images of substations.

[0024] Obtain infrared thermal images of the substation collected by thermal imaging equipment.

[0025] S2. Preprocess the infrared thermal image and identify the heating area from the preprocessed image.

[0026] The method for preprocessing the infrared thermal image of the substation includes: graying and denoising the infrared thermal image, and then performing histogram equalization to enhance the image contrast.

[0027] The method for identifying the heating area from the preprocessed image includes: using the region growing method, extracting the middle point of the area with the maximum gray value from the preprocessed image as a seed point, performing growth and corrosion, and then performing edge extraction to obtain the heating area.

[0028] S3. Use the relative temperature difference method to determine whether there is a fault in the heating area. If there is a fault, process the infrared thermal image to obtain the HSI color feature.

[0029] The method of using the relative temperature difference method to determine whether there is a fault in the heating area includes: obtaining the relative temperature difference of the heating area, and if the relative temperature difference is not less than 50°C, it is determined that there is a fault in the heating area. , where T1 is the temperature of the heating area, T0 is the ambient temperature, and T2 is the temperature of the heating area under normal conditions.

[0030] The method for processing an infrared thermal image to obtain an HSI color feature includes: obtaining the RGB color feature of the infrared thermal image, and converting the RGB color feature into an HSI color feature after normalization processing.

[0031] The formula for converting the normalized RGB color features to HSI color features is: Where H, S and I are the hue, saturation and brightness in the HSI color mode, R, G and B are the normalized values ​​of the three primary colors of red, green and blue in the RGB color, and θ is the hue angle. .

[0032] When it is determined that there is no fault in the heating area of ​​the infrared thermal image, subsequent steps are stopped.

[0033] S4. Based on the HSI color features and the K-means algorithm, the fire area image subset is segmented from the infrared thermal image.

[0034] This step uses the K-means clustering algorithm to divide the infrared thermal image into different types of image subsets according to the types of substation facilities, sky, site, flame and flame irradiation area based on the HSI color features of the infrared thermal image, and marks the types corresponding to the image subsets. The image subset corresponding to the flame is the fire area image subset.

[0035] The K-means clustering algorithm divides the feature matrix (HSI color feature) composed of certain data samples into non-intersecting clusters (substation facilities, sky, site, flame and flame irradiation area) to characterize the clustering performance results. K-means is a distance-based clustering algorithm. This step uses Euclidean distance to measure the similarity between samples. The formula for calculating the Euclidean distance method in K-means is: , where x is the pixel position in the cluster, α is the centroid position in the cluster, n is the number of pixel features, and i is the pixel feature value.

[0036] S5. Combining flame texture and edge features, a sample entropy signal processing algorithm is used to obtain the sample entropy value of the fire area image subset, and whether a fire occurs is determined based on the sample entropy value.

[0037] There may be objects with colors or features similar to flames in the infrared thermal image of the substation. These similar objects may be segmented into the fire area image subset in step S4. For example, fire cabinets, fire pipes, red signs, etc. may be segmented and marked as flame types. Therefore, the accuracy of flame recognition directly using the fire area image subset is relatively low.

[0038] By comparing and analyzing the edge, texture, grayscale and other features of flames and objects similar to flames, such as fire cabinets and fire pipes, it is found that the distribution of flame signals is relatively messy and irregular, and is much more complex than fire cabinets and fire pipes. Therefore, this step calculates the sample entropy value of the fire area image subset before making a fire determination. If the calculated sample entropy value is greater than the preset value, it is determined that a fire has occurred.

[0039] The process of obtaining sample entropy value includes: 1) Reduce the dimension of the fire area image subset to a time series of length N , 1≤n≤N; 2) Construct a set of m-dimensional space vectors Y(i) according to the comparison dimension, , i≤N-m+1; 3) Define the maximum difference distance between vectors Y(i) and Y(j) , 1≤j≤N-m+1, 0≤k≤m-1; 4) Set the allowable deviation r and calculate the The number of , calculate the ratio of this number to the total number of distances , ; 5) Calculate the average ratio of all i , ; 6) Increase the dimension, let m become m+1, repeat the above steps and finally get the sample entropy .

[0040] This embodiment is verified below.

[0041] The safety area of ​​a 110kV substation was selected as the experimental area, and the entire experimental process lasted 100 minutes. The substation professionals ignited the fire at 30 minutes, and the flame was extinguished after 10 minutes. The video was shot by a thermal imaging camera, and an image was extracted from the video every 30 seconds (a total of 200 images, including 20 fire images) and the fire recognition model script was called. The model script was written using the development tool Python 3.7 OpenCV library, and the sample entropy threshold was selected as 0.8. A sample entropy greater than 0.8 was judged as a fire, and a sample entropy less than 0.8 was judged as no fire.

[0042] The traditional watershed algorithm was used as Algorithm 1 and compared with the algorithm 2 of this application. 200 substation fire monitoring images were segmented and compared for fire recognition. The fire recognition results of the two algorithms are as follows: algorithm A B C D E F 1 14 6 37 20 200 21.68 2 20 0 8 20 200 15.21 Where A is the number of fires correctly identified in fire photos, B is the number of fires not correctly identified in fire photos, C is the number of non-fire photos diagnosed as fire photos, D is the total number of fire photos, E is the total number of photos, and F is the judgment time.

[0043] Among the 20 images that actually showed fire, algorithm 1 identified 14 images with a misdiagnosis rate of 30%, while algorithm 2 identified 20 images with no misdiagnosis. Among the 180 images that were actually not fires, algorithm 1 identified 37 images as fire images with a misdiagnosis rate of 20.56%, while algorithm 2 identified 8 images as fire images with a misdiagnosis rate of 4.44%. The average time t for fire recognition and alarm for algorithm 1 and algorithm 2 were 21.68s and 15.21s, respectively.

[0044] Example 2: See Figure 2 The purpose of this embodiment is to provide a substation fire detection system based on feature recognition, and the detection system includes an image processing module, a fault identification module and a fire identification module.

[0045] The image processing module is used to collect infrared thermal images of the substation, to preprocess the infrared thermal images, and to identify the heating area from the preprocessed images. In combination with Embodiment 1, the image processing module is used to perform steps S1-S2.

[0046] The fault identification module is used to determine whether there is a fault in the heating area by using the relative temperature difference method, and is used to process the infrared thermal image to obtain the HSI color feature when there is a fault. In combination with embodiment 1, the fault identification module is used to execute step S3.

[0047] The fire determination module is used to segment the fire area image subset from the infrared thermal image according to the HSI color feature and the K-means algorithm, and is used to combine the flame texture and edge features and use the sample entropy signal processing algorithm to obtain the sample entropy value of the fire area image subset, and is used to determine whether a fire occurs according to the sample entropy value. In combination with Example 1, the fire determination module is used to execute steps S4-S5.

[0048] The specific manner in which each unit or module performs operations in this embodiment has been described in detail in Embodiment 1 and will not be elaborated on here.

[0049] Those skilled in the art should understand that for the sake of convenience and brevity in description, the present application only uses the division of the above-mentioned functional modules or units as examples in the present embodiment. In actual applications, the above-mentioned functions can be assigned to different functional modules or units as needed, that is, the internal structure of the present embodiment can be divided into different functional modules or units to complete all or part of the functions described above.

[0050] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Other modifications or equivalent substitutions made to the technical solution of the present invention by ordinary technicians in the field should be included in the scope of the claims of the present invention as long as they do not depart from the spirit and scope of the technical solution of the present invention.

Claims

1. A substation fire detection method based on feature recognition, characterized in that: include: Collect infrared thermal images of substations; Preprocessing the infrared thermal image and identifying the heating area from the preprocessed image; The relative temperature difference method is used to determine whether there is a fault in the heating area of ​​the infrared thermal image. If there is a fault, the infrared thermal image is processed to obtain the HSI color feature; Segment the fire area image subset from the infrared thermal image based on the HSI color features; Combining flame texture and edge features, a sample entropy signal processing algorithm is used to obtain the sample entropy value of the fire area image subset; Determine whether a fire occurs based on the sample entropy value.

2. The substation fire detection method based on feature recognition according to claim 1 is characterized in that: The method for preprocessing the infrared thermal image includes: graying and denoising the infrared thermal image, and then performing histogram equalization.

3. The substation fire detection method based on feature recognition according to claim 1 is characterized in that: The method for identifying the heating area from the preprocessed image includes: using the region growing method, extracting the middle point of the area with the maximum gray value from the preprocessed image as a seed point, performing growth and corrosion, and then performing edge extraction to obtain the heating area.

4. The method for detecting substation fire based on feature recognition according to claim 1, characterized in that: The method for determining whether there is a fault in the heating area of ​​an infrared thermal image using the relative temperature difference method includes: obtaining the relative temperature difference of the heating area, and if the relative temperature difference is not less than 50°C, determining that there is a fault in the heating area; the relative temperature difference , where T1 is the temperature of the heating area, T0 is the ambient temperature, and T2 is the temperature of the heating area under normal conditions.

5. The substation fire detection method based on feature recognition according to claim 1 is characterized in that: The method for processing an infrared thermal image to obtain an HSI color feature includes: obtaining an RGB color feature of the infrared thermal image, normalizing the RGB color feature, and converting the RGB color feature into an HSI color feature.

6. The method for detecting fire in a substation based on feature recognition according to claim 1, characterized in that: The method for segmenting a fire area image subset from an infrared thermal image includes: using a K-means clustering algorithm, according to the HSI color features of the infrared thermal image, the infrared thermal image is segmented into different types of image subsets according to the types of substation facilities, sky, site, flame and flame irradiation area, wherein the image subset corresponding to the flame is the fire area image subset.

7. The method for detecting fire in a substation based on feature recognition according to claim 6, characterized in that: The K-means algorithm uses Euclidean distance to measure the similarity between samples.

8. The method for detecting fire in a substation based on feature recognition according to claim 1, characterized in that: The method of obtaining the sample entropy value of the fire area image subset by using the sample entropy signal processing algorithm includes: Reduce the fire area image subset to a time series of length N , 1≤n≤N; Construct a set of m-dimensional space vectors Y(i) according to the comparison dimension, , i≤N-m+1; Define the maximum difference between vectors Y(i) and Y(j) , 1≤j≤N-m+1, 0≤k≤m-1; Set the tolerance r and calculate the value for each i that meets The number of , calculate the ratio of this number to the total number of distances , ; Calculate the average ratio of all i , ; Increase the dimension, let m become m+1, repeat the above steps and finally get the sample entropy .

9. The method for detecting fire in a substation based on feature recognition according to claim 1, characterized in that: The method for determining whether a fire occurs according to a sample entropy value includes: if the sample entropy value is greater than a preset value, determining that a fire occurs.

10. A substation fire detection system based on feature recognition, characterized in that: It includes an image processing module, a fault identification module and a fire identification module; The image processing module is used to collect infrared thermal images of the substation, to preprocess the infrared thermal images, and to identify the heating area from the preprocessed images; The fault identification module is used to determine whether there is a fault in the heating area by using the relative temperature difference method, and is used to process the infrared thermal image to obtain the HSI color feature when there is a fault; The fire determination module is used to segment a fire area image subset from an infrared thermal image based on HSI color features and using a K-means algorithm, and is used to combine flame texture and edge features and use a sample entropy signal processing algorithm to obtain a sample entropy value of the fire area image subset, and is used to determine whether a fire has occurred based on the sample entropy value.

Images