Smoke Detection by Combining Dynamic Feature Analysis and Improved YOLOv5
ZHANG Junfei
GONG Faming
WEI Wenjian
Abstract：Aiming at the problems of low recognition accuracy,missing detection of small targets,false detection and poor re-al-time performance in existing dynamic smoke detection algorithms,a detection algorithm combining dynamic feature analysis of smoke and improved YOLOv5 is proposed.DFA_YOLO algorithm includes dynamic feature analysis and target detection.In the part of dynamic feature analysis,the interframe difference method is improved and the motion direction judgment based on pixel block is proposed to analyze the dynamic feature of smoke,so as to filter out the non-smoke moving objects in the image.In the part of target detection,this paper improves YOLOv5,uses K-means++clustering algorithm to obtain anchor more consistent with smoke and flame size,adds convolutional attention module in the backbone feature extraction network,and adds small target detection layer.Experimental results show that compared with YOLOv5,mAP is improved by 2.9%,and the average detection speed is improved by 6fps.
Keywords：interframe difference methodobject detectionYOLOv5K-means++CBAM
Publication Date：2025-10-20
Online Publishing Date：2026-01-16(First online date of this platform, not the publication date of the document)
Pages：7( 2760-2765,2772 )