Smoke Detection in Complex Scenery Using Improved Faster-RCNN

Aiming at the complexity of smoke detection scenarios, high target smoke undetect rate, poor detection effect, and so on. This paper propose an object detection algorithm based on the improved Faster-RCNN; firstly, in order to improve the accuracy of the model smoke detection accuracy, we use the Resnet50 feature extraction network to replace the VGG-16 network in original model. The experimental results show that after replacing the network with Resnet50, the model Map50 is improved by 2.4%; secondly, we conduct an experiment using the swin-transformer as the backbone, and the experimental results on the same smoke dataset show that the detection accuracy of Map50 using the swin-transformer is improved by 3.0% compared with the baseline. It can be conclude that the improved Faster-RCNN detection algorithm obtain a better performance and a good detection results.