Lightweight anti-interference flame detection method based on improved YOLOv4-tiny

Abstract

Flame detection serves as an important link of fire prevention and control, and has high requirements for real-time performance, anti-interference performance and accuracy. The flame target detection method at the present stage lacks comprehensive research on the three indexes, and in order to solve the problem, the invention provides the light-weight anti-interference flame detection method based on the improved YOLOv4-tiny. A flame detection model of a double-flow structure is designed by utilizing the dynamic characteristic that flame changes along with time. The method comprises the following steps: firstly, carrying out lightweight improvement on a backbone network of YOLOv4-tiny by adopting depth separable convolution; secondly, in the feature extraction stage, the learning ability of the network for shallow features is improved by further fusing multi-scale features, and meanwhile, an ECA channel attention module is introduced into the FPN, so that the precision is further improved; and finally, an IOU (Intersection over Union) post-processing algorithm is adopted to effectively shield the interference of the fire-like target. In a dataset aspect, an own flame detection dataset is created. Experiments prove that the accuracy, the anti-interference performance and the detection time of the method are comprehensively improved.

Description

technical field [0001] The invention belongs to the field of flame detection, in particular to a lightweight anti-interference flame detection method based on improved YOLOv4-tiny. Background technique [0002] Fire has always been a major disaster that needs people's attention and vigilance. Due to its characteristics of various inducing factors, rapid spread and high outbreak, it will endanger human life and cause huge economic losses. Therefore, fire prevention has become a multidisciplinary research. important subject. As an important part of fire prevention and control, flame detection has always been valued by researchers. Traditional sensor-based flame detection technologies, such as temperature sensors, smoke sensors, and photosensitive sensors, all need to respond to parameters such as temperature, smoke particles, and light intensity during the combustion process of the flame through sensitive components. At the same time, the physical signal is converted into an...

AI Technical Summary

Problems solved by technology

[0029] Aiming at the problems of external interference, rapid detection and positioning in flame detection, the present invention proposes a lightweight and anti-interference flame detection method based on improved YOLOv4-tiny

Images