ESA-Net: A Network with Efficient Spatial Attention for Smoky Vehicle Detection
Jianan Zhou, Shaowei Qian, Zhongzong Yan, Jingbo Zhao, He Wen
Abstract
As automobile exhaust is one of the main sources of air pollution, the detection of smoky vehicle with high efficient is an important prerequisite of automobile exhaust regulation. The most commonly used method at present is using additional sensors, which requires excessive costs. This paper proposes the Efficient Spatial Attention Net (ESA-Net) for the sensor-free detection of smoky vehicle based on videos collected in road fixed monitoring system. In the proposed method, the YOLOv4 and the ResNet are used for vehicles detection and smoke image classification, respectively. Thus the feature spatial representation of each ResNet block combined by weight combination can be obtained by the ESA-Net. Experiments results show that the ESA-Net outperforms several methods based on convolution neural network.