A substation site operation violation behavior detection method based on a multi-channel CNN

Abstract

The application discloses a substation field operation violation behavior detection method based on a multi-channel CNN, belongs to the field of artificial intelligence digital image processing, and comprises the following steps: S1, establishing a substation field operation violation behavior database; S2, calculating the area of a violation behavior region intercepted in S1, dividing the area of the violation behavior region intercepted into N classes according to the area size and counting the area quantity of each class, and then calculating an area distribution probability alpha i ; S3, for the database established in S1, fusing the area distribution probability alpha i , and constructing a multi-channel CNN model; S4, adopting the multi-channel CNN model to train the database established in S1, inputting the image preprocessed in S1.2 into the multi-channel CNN model of S3, and outputting a substation field operation behavior detection result. According to the existing method, the influence of image features of different scales on detection performance is neglected, the detection method can effectively detect targets of different scales by optimizing channel weights through the area distribution of a target region, and has strong robustness to complex substation field operation images.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence digital image processing, and more specifically, relates to a method for detecting violations of substation operation rules based on multi-channel CNN. Background Technology

[0002] In recent years, with the rapid development of the power industry and the continuous increase in power grid equipment, the workload and intensity of substation workers have significantly increased. Affected by factors such as skill level, physical condition, and adverse weather conditions, the risks involved in substation operations have become increasingly complex. In particular, the installation, commissioning, and maintenance of some electrical equipment require workers to operate at heights, posing significant safety hazards. Statistics show that most substation system accidents are closely related to workers' violations of operating procedures. Examples include working at substation sites without wearing safety helmets, climbing at heights without safety belts, smoking in work areas, and operating outdoor high-voltage equipment without wearing insulated boots. Therefore, detecting violations at substation sites is of great importance.

[0003] To reduce substation operation accidents, traditional methods rely on manual monitoring or remote video surveillance of the work site to promptly stop irregular or illegal activities. This method is time-consuming, labor-intensive, and tedious. Existing detection methods mainly extract image features from a single channel or perform channel fusion with fixed weights, ignoring the impact of image features at different scales on detection performance.

[0004] For example, the invention patent with publication number CN108597172A, entitled "A Method and System for Compressed Domain Video Action Recognition Based on New Spatiotemporal Feature Streams," uses preprocessed dual-channel motion vector data and preprocessed residual single-channel data as R / G / B channels respectively to fuse and construct a new spatiotemporal feature image. This image is then input into a CNN convolutional neural network model for training and testing to obtain action recognition category scores. Similarly, the extracted I-frame data and preprocessed motion vector data are input into the CNN convolutional neural network model for training and testing to obtain action recognition category scores. Finally, the action recognition category scores of these three types of data are fused in a 2:1:1 ratio to obtain the final action recognition result. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention proposes a method for detecting violations in substation operations based on multi-channel CNN. By optimizing channel weights through statistical analysis of the area distribution of the target region, this method can effectively detect targets of different scales and exhibits strong robustness to complex substation operation images.

[0006] The present invention adopts the following specific technical solution:

[0007] A method for detecting violations in substation field operations based on multi-channel CNN includes the following steps:

[0008] S1: Establish a database of violations of regulations during substation operations;

[0009] S1.1: Collect images of violations at the substation site;

[0010] S1.2: Image preprocessing;

[0011] S1.3: Extract the area of ​​violation and mark the violation belonging to that area;

[0012] S2: Calculate the area of ​​the intercepted violation area in S1, divide it into N categories according to the size of the intercepted violation area, and count the number of areas in each category, N≥2. Then calculate the area distribution probability α. i , where i represents the i-th type of region, 1≤i≤N;

[0013] S3: For the database established in S1, the probability α of the fused area distribution. i Construct a multi-channel CNN model to improve the detection capability of targets of different sizes;

[0014] S4: A multi-channel CNN model is used to train the database established in S1. The preprocessed images in S1.2 are input into the multi-channel CNN model in S3, and the results of substation site operation behavior detection are output.

[0015] Furthermore, the image preprocessing method in S1.2 is as follows: convert the image into a grayscale image and normalize the image size to M×M pixels.

[0016] Furthermore, in the M×M pixels, M is preferably 2. k , k≥7.

[0017] Furthermore, the categories of violations in S1.3 include not wearing work clothes, not wearing a safety helmet, crossing temporary barriers, not using a safety belt when working at height, not wearing insulated boots when operating outdoor high-voltage equipment, and not having anyone hold the ladder during the operation.

[0018] Furthermore, if there is no such category of violation, then it is considered standard operating procedure.

[0019] Furthermore, the area distribution probability in S2 Among them, A i Let A be the number of regions belonging to the i-th type of region. all This represents the total number of areas where violations occurred.

[0020] Furthermore, the multi-channel CNN model contains N channels, and each channel uses convolutional kernels of different scales to extract image features.

[0021] Furthermore, the multi-channel CNN model consists of convolutional layers, ReLU layers, 2×2 pooling and concatenation layers, fully connected (FC) layers, and softmax classification layers.

[0022] Furthermore, to fully utilize channel information at different scales, the concatenating layer employs the following formula for information fusion:

[0023]

[0024] Where, p i Let P be the output of channel i, and P be the fusion result of N channels. The weight coefficients of the channels adopt the area distribution probability α. i Optimize, α i satisfy

[0025] Furthermore, to improve the efficiency of the detection method, the training and classification of the multi-channel CNN model are both performed on the GPU.

[0026] The beneficial effects of this invention are as follows:

[0027] (1) The multi-channel CNN model uses multiple convolutional channels of different scales to extract image features and optimizes the channel weights by statistically analyzing the area distribution of the target region. It can effectively detect targets of different scales and has strong robustness to complex substation operation images.

[0028] (2) It can be extended to behavior detection in other scenarios. Attached Figure Description

[0029] Figure 1 This is a schematic flowchart of the detection method of the present invention;

[0030] Figure 2 This is a schematic diagram of a multi-channel CNN model according to a preferred embodiment of the present invention. Detailed Implementation

[0031] The present invention will be further described below with reference to specific embodiments. Unless otherwise specified, the raw materials and methods used in the embodiments of the present invention are commercially available materials and conventionally used methods in the art.

[0032] GPU stands for Graphics Processing Unit, also known as a display core, visual processor, or display chip, and is a microprocessor used for image processing. CNN stands for Convolutional Neural Networks, a recognition algorithm based on convolutional neural networks.

[0033] Example 1

[0034] like Figure 1 As shown, a method for detecting violations in substation field operations based on multi-channel CNN includes the following steps:

[0035] S1: Establish a database of violations of regulations during substation operations;

[0036] S1.1: Collect images of violations at the substation site;

[0037] S1.2: Image preprocessing: Convert the image to grayscale and normalize the image size to M×M pixels, where M is set to 256;

[0038] S1.3: Capture the area of ​​violation and mark the violation in that area. Violation categories include not wearing work clothes, not wearing a safety helmet, crossing temporary barriers, not using a safety belt when working at height, not wearing insulated boots when operating outdoor high-voltage equipment, and not having anyone hold the ladder during the operation. Otherwise, it is considered a standard operation.

[0039] S2: Calculate the area of ​​the region where the violation was intercepted in S1, divide it into three categories according to the size of the intercepted area, count the number of regions in each category, and then calculate the area distribution probability. Among them, A i Let A be the number of regions belonging to the i-th type of region. all This represents the total number of areas where violations occurred.

[0040] S3: For the database established in S1, the probability α of the fused area distribution. i A multi-channel CNN model is constructed to improve the detection capability of targets of different sizes. A multi-channel CNN model contains N channels, such as... Figure 2 As shown, in this embodiment, N=3, and each channel uses convolutional kernels of different scales to extract image features. The model consists of convolutional layers, ReLU layers, 2×2 Pooling and concatenating layers, FC fully connected layers, and softmax classification layers.

[0041] To fully utilize channel information at different scales, the concatenating layer employs the following formula for information fusion:

[0042]

[0043] In this embodiment, p i Let P be the output of channel i, and P be the fusion result of the three channels. The weight coefficients of the channels adopt the area distribution probability α. i To optimize, α1, α2, and α3 were calculated and set to 0.3, 0.4, and 0.3, respectively.

[0044] S4: A multi-channel CNN model is used for training and classification. To improve the efficiency of the detection method, both training and classification of the multi-channel CNN model are performed on the GPU.

[0045] S4.1: Train the database established in S1;

[0046] S4.2: Input the preprocessed grayscale image with a pixel size of 256×256 from S1.2 into the multi-channel CNN model in S3 to realize the detection of illegal behavior in substation operations based on multi-channel CNN.

[0047] The above description is merely a preferred embodiment of the present invention and is 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 principles of the present invention are included within the scope of protection of the present invention.

Claims

1. A method for detecting substation site operation violation behavior based on a multi-channel CNN, characterized in that, The method comprises the following steps: S1: establishing a substation site operation violation behavior database; S1.1: collecting substation site violation behavior images; S1.2: image preprocessing; S1.3: intercepting a violation behavior region and labeling the violation behavior to which the region belongs; S2: calculate the area of the region of the interception violation behavior in S1, divide into classes according to the area of the region of the interception violation behavior, and count the number of regions of each class, then calculate the area distribution probability : wherein is the number of regions belonging to the class of the class regions, is the total number of regions of violation behavior, denotes the class region, ; S3: fuse area distribution probability for the database established in S1 A multi-channel CNN model is constructed to improve the detection capability of different size targets. S4: training the database established in S1 using a multi-channel CNN model, inputting the preprocessed images in S1.2 into the multi-channel CNN model in S3, and outputting substation site operation behavior detection results; The multi-channel CNN model comprises channels, each channel adopts a convolution kernel of different scales to extract image features, and the multi-channel CNN model is composed of a convolution layer, a ReLu layer, a pooling layer, a concate layer, an FC full connection layer and a softmax classification layer, wherein the concate layer adopts the following formula for information fusion: Wherein, The output of the channel , The fusion result of the channel The weight coefficient of the channel adopts area distribution probability Optimization is carried out, Satisfy . 2.The substation field operation violation behavior detection method based on the multi-channel CNN according to claim 1, wherein, The image preprocessing method of S1.2 is: converting the image into a gray image, normalizing the image size to pixels. 3.The substation field operation violation behavior detection method based on the multi-channel CNN according to claim 2, characterized in that, The In the pixel Preferably , . 4.The substation field operation violation behavior detection method based on the multi-channel CNN according to claim 1, wherein, The violation behavior categories in S1.3 include not wearing a work uniform, not wearing a safety helmet, crossing a temporary fence, not using a safety belt for climbing operation, not wearing insulated boots for operating outdoor high-voltage equipment, and no one in the escalator during operation.

5. The method according to claim 4, wherein, If there is no violation behavior category, it is a standard operation. 6.The substation field operation violation behavior detection method based on the multi-channel CNN according to claim 1, wherein, To improve the operation efficiency of the detection method, the training and classification of the multi-channel CNN model are completed in a GPU.

Images