Rapid rarefied smoke segmentation method based on multiple attention mechanisms

Abstract

The invention discloses a rarefied smoke rapid segmentation method based on multiple attention mechanisms. The method is characterized in that useless interference information is filtered out for multiple times by utilizing multiple attention mechanisms (time domain, channel and space attention mechanisms), and the algorithm speed is improved by utilizing the sensitivity of the frame difference method to smoke and the subsequent lightweight of the whole network, so that the just appearing semitransparent smoke can be captured. Experiments prove that by using the algorithm on a smoke data set of a national laboratory of Chinese science and technology university fire science, early rarefied smoke can be accurately segmented while a large amount of interference is filtered, and the processing delay of each frame is less than 50ms. The problem that an existing rarefied smoke monitoring algorithm cannot give consideration to accuracy, interference screening capacity and low processing time delay at the same time is solved, and rapid and accurate segmentation of the early-stage rarefied smoke is achieved.

Description

technical field [0001] The invention belongs to the field of image processing, and relates to a method for fast segmentation of thin smoke based on a multi-attention mechanism. Background technique [0002] As a catastrophic problem faced by all countries in the world, fire is sudden and destructive, and can pose a great threat to human property and ecological resources. And if the fire can be detected in time, not only can the casualties be greatly reduced, but also a lot of ecological and social resources can be saved. [0003] By observing the occurrence of fire, we can know that before the fire, there will be a process of diffuse and diffuse smoke: from thin to rich, from translucent to milky white or black. If the thin translucent smoke that has just appeared can be captured, the early warning of fire can be realized with maximum efficiency. The specific characteristics of thin smoke are: 1) thin smoke has no fixed shape, and its edges gradually blend into the backgro...

AI Technical Summary

Problems solved by technology

However, in detecting the thin smoke that has just appeared, the effect is far from the ideal goal.

It mainly includes: 1) The method based on the static feature of the image is less sensitive to the thin smoke, and often produces missed detection; 2) The method based on the dynamic feature is more sensitive to the thin smoke of the movement, but it will bring a lot of non-smog dynamics Information, that is, false alarm; 3) The method based on the simple combination of dynamic and static has improved the detection effect of thin smoke, but due to the lack of pertinence to smoke, there are still some missed detection and false alarm, and the detection efficiency is low. Can't keep up with real-time changes in smoke

[0004] In summary, the accuracy of the current method for thin smoke and a large amount of interference are a pair of difficult contradictions

Images