Enhanced receptive field smoke detection model embedded with attention mechanism

Smoke is an important feature of forest fire early detection. The accuracy of the existing smoke detection model is easily affected by the surrounding environment, and the sampling times of the image center and edge are uneven, which eventually leads to the unsatisfactory effect of feature extraction and low accuracy. In order to solve these problems, a SE_RFB_YOLO fire smoke target detection model is proposed which embeds channel attention mechanism based on YOLOv3 model to improve the anti-interference ability and feature extraction ability of the network; Clustering is used to obtain the a priori frame size that meets the smoke target; The feature map is connected to the RFB-s module to expand the receptive field, and the extracted features are more obvious. Compared with YOLOv3, target detection accuracy of RFBNet model is 86.4%, the recall rate is 89.7%, and the detection speed is 26 fps.The proposed target detection model can better meet the needs of the task and show better detection effect.