Changlin Song, Chuanfu Liu, Tingping Feng, Junmin Li, Yanlin Pan, and Simon X. Yang
Convolutional neural network, faster R-CNN, Infrared thermal imaging video smoke detection
Smoke is a noticeable feature when a fire occurs. Therefore, the detection of smoke becomes an important research direction for fire prevention. At present the research on smoke detection is mainly based on the Convolutional Neural Network (CNN) under the visible light condition, which is not suitable for the changeable outdoor environment. Therefore, a dual-spectral video smoke detection system based on an improved Faster R-CNN is proposed. Firstly, the fusion network of the VGG16 network and the improved Convolutional Block Attention Module (CBAM) network is used to extract features, which is used to solve the problem that it is difficult to extract smoke colour features, shape features and texture features in real-time. Secondly, the non-maximum suppression (NMS) algorithm of Faster R-CNN is improved to solve the precision degradation problem. Finally, the infrared thermal imaging camera is used to identify the heat source after selecting the smoke, which solves the problem that white clouds, fog and other similar objects are misidentified as smoke and improves the identification system’s accuracy. The experimental results show that the improved model is superior to the visible light detection method only and the method can be well used for smoke detection.
Important Links:
Go Back
