嵌入CBAM结构的改进YOLOV3超宽带雷达生命信号检测算法

Improved YOLOV3 UWB radar life signal detection algorithm embedded in CBAM structure

为了提高超宽带雷达生命信号检测准确率,提出了一种改进型YOLOV3检测算法。考虑到超宽带雷达探测的人体呼吸信号成像后宽高相差较大,采用多尺度训练时网络对宽高的敏感程度不一样,因此在设计损失函数时,对宽高损失给予不同的权重系数。为了提高网络通道以及空间特征表达能力,使得网络能够扩大对目标特征区域的感知范围,因此将通道注意力与空间注意力结构(CBAM)嵌入到YOLOV3基础网络Darknet53的每个shortcut层,得到CBAM-YOLOV3网络结构。对比改进前后的算法。实验结果表明,改进后的算法在IoU阈值为0.5时,mAP为98.09%,提升3.90%,在阈值为0.75时,mAP为76.49%,提升16.48%。

In order to improve the accuracy of UWB radar life signal detection,an improved YOLOV3 detection algorithm is proposed in this paper.Considering that the width and height of human respiratory signals detected by UWB radar differ greatly after imaging.The sensitivity of the network to the width and height is different when using multi-scale training.Therefore,when designing the loss function,different weight coefficients are given for the width and height loss.In order to improve the ability of network channel and spatial feature expression,the network can expand the perception range of target feature area.Therefore,channel attention and spatial attention structure(CBAM)is embedded in each short cut layer of YOLOV3 basic network Darknet53,and CBAM-YOLOV3 network structure is obtained.Compare the algorithm before and after the improvement.The experimental results show that the improved algorithm has a mAP of 98.09%and an increase of 3.90%when the IoU threshold is 0.5.At a threshold of 0.75,mAP is 76.49%,an increase of 16.48%.