基于改进YOLOv4的红外行人车辆检测算法

Detection algorithm for infrared pedestrian and vehiclebased on the improved YOLOv4

智能设备对行人和车辆的目标检测对于建设智慧城市有着重要的意义。随着红外技术的发展和普及,红外成像科技具有强抗干扰和全天候的特性,被越来越多地用于解决可见光受限环境带来的问题。论文提出了一种改进YOLOv4深度学习算法对红外图像下的行人车辆进行检测。改进的YOLOv4算法加入了CA注意力机制模块,将位置信息嵌入到通道注意中,增强了对感兴趣区域的表示。此外还设计了CSP2-DBL模块,替换了原本简单的卷积模块叠加,对高分辨率特征性信息的做出了弥补。为了进一步提高网络计算速度,减少计算量,针对红外图像特性,对Head部分进行了裁剪。实验结果表明改进后的模型在FLIR红外数据集上较YOLOv4模型在mAP上提高了0.85个百分点,检测速度提升了2 f/s。

The target detection of pedestrians and vehicles by intelligent devices is of great significance for building a smart city.With the development and popularization of infrared technology,infrared imaging technology with strong anti-interference and all-weather characteristics,is increasingly used to solve the problems caused by the visible light limited environment.In this paper,an improved YOLOv4 deep learning algorithm to detect pedestrians and vehicles in infrared images is proposed.The improved YOLOv4 algorithm incorporates the CA attention mechanism module to embed the location information into the channel attention to enhance the representation of the regions of interest.In addition,the CSP2-DBL module is designed to replace the original simple convolution module superposition and make up for the high-resolution characteristic information.In order to further improve the network computing speed and reduce the amount of computing,the Head part is cut for the characteristics of infrared images.The experimental results show that the improved model improves the mAP by 0.85%and the detection speed by 2 f/s over the YOLOv4 model on the FLIR infrared dataset.