A kind of chemical gas leak detection method and system based on spectral video

Abstract

The present application provides a chemical gas leakage detection method based on spectral video, which includes: preprocessing the collected spectral video of chemical gas at the image level; dividing the spectral video into multiple segments in time series, and Sampling a video frame in each segment, and extracting the features of each video frame respectively; by fusing the local motion information of multiple adjacent video frames in the features of each video frame, the short-term timing enhancement of each video frame is obtained. Features; through the channel-by-channel integration method of shifting and weighting, the timing information is fused in the short-term timing enhancement feature of each video frame to obtain the global timing fusion feature of each video frame; the short-term timing enhancement feature and the global timing fusion feature The residual connection is incorporated into the main network, and the detection result of the chemical gas leakage is calculated. The application can efficiently and real-time detect and present information such as gas leakage cloud position, concentration distribution, and diffusion trend in spectral video.

Description

technical field [0001] This application relates to the technical field of gas detection, in particular to a method and system for detecting chemical gas leaks based on spectral video. Background technique [0002] Timely detection and control of chemical gas leakage is the premise and the most important thing for the safe operation of industrial sites. However, accidents such as fire, explosion and poisoning caused by gas leakage are still common all over the world. The gas detection method with spectral video technology as the core can effectively warn chemical gas leakage with its advantages of long distance and large range. [0003] However, there is no mature and targeted classification recognition algorithm applied in the field of spectral video gas detection. On the one hand, the traditional motion detection algorithm cannot efficiently distinguish leaked gas from other moving targets, and in practical applications, the main body of the leaked gas cloud is often mixed...

AI Technical Summary

Problems solved by technology

[0003] However, there is currently no mature and targeted classification recognition algorithm applied in the field of spectral video gas detection.

On the one hand, the traditional motion detection algorithm cannot efficiently distinguish leaked gas from other moving targets, and in practical applications, the main body of the leaked gas cloud is often mixed with various noises, making it difficult to accurately and effectively identify gas leaks

On the other hand, the main body of the gas leakage cloud lacks obvious morphological features. In the case of low image contrast, low gas leakage concentration, and rapid changes in the gas cloud distribution, multiple consecutive frames of images are needed to barely locate the cloud. The target detection task of traditional computer vision relies heavily on artificially designed features. In the process of artificially designing features, some inherent features of gas leak images are often ignored, which will lead to a large number of false positives and negative negatives; at the same time, deep learning It is not well applied in the field of motion detection, and the existing convolutional neural network (CNN) architecture is also difficult to perform the target recognition and segmentation tasks without obvious features such as gas leaks.

Images