Flame and Smoke Detection Algorithm for UAV Based on Improved YOLOv4-Tiny

Aiming at the current YOLOv4-tiny network’s insufficient feature fusion capability and low utilization of feature extraction in flame and smoke detection tasks, a flame and smoke detection algorithm based on improved YOLOv4-tiny is proposed. Firstly, a new effective feature layer is added to obtain more detailed feature information and improve the accuracy of small target detection of flame and smoke. Then, the DWCSP feature fusion structure is proposed to improve the network’s ability to integrate and utilize multi-scale feature information on the basis of minimizing the increment of parameters. Finally, the CBAM attention mechanism is embedded to improve the network’s channel and spatial feature expression ability, and enhance the ability to perceive the target. The algorithm is embedded in the UAV equipment. In the detection task of self built flame and smoke data set, the mAP@0.5 reaches 71.11%, which is 6.48% higher than the original algorithm, and meets the needs of FPS and lightweight.