基于改进YOLOv4-tiny的舰面多目标检测算法

Shipboard Multi-target Detection Algorithm Based on Improved YOLOv4-tiny

针对舰面多目标的检测问题,提出一种改进YOLOv4-tiny的舰面多目标检测算法。在卷积神经网络中引入卷积注意力模块(convolutional block attention module,CBAM),通过混合通道特征和空间特征来关注舰面目标和抑制背景特征,提高网络的抗背景干扰能力;针对目标尺度变化加入空间金字塔池化结构(spatial pyramid pooling,SPP)以融合不同尺度的特征,提高对不同大小目标的检测能力;使用Mish激活函数替代Leaky ReLU激活函数以获得更好的泛化能力。实验结果表明:5类舰面目标的平均检测精度为92.22%,接近YOLOv4算法的96.48%,而检测速度(frames per second,FPS)达到了42.5帧/s,远高于YOLOv4的18帧/s;该算法能较好地平衡准确率和速度的关系,可以对舰面目标进行实时检测。

An improved YOLOv4-tiny multi-target detection algorithm is proposed to solve the problem of multi-target detection on shipboard.A convolutional block attention module(CBAM)is introduced into the convolutional neural network to focus on shipboard targets and suppress background features by mixing channel features and spatial features,so as to improve the anti-background interference ability of the network;an spatial pyramid pooling(SPP)structure is added according to the change of target scale to fuse features of different scales,so as to improve the detection ability of targe ts of different sizes;Mish activation function is used instead of Leaky ReLU activation function for better generalization ability.The experimental results show that the average detection accuracy of five kinds of shipboard targets is 92.22%,which is close to the 96.48%of YOLOv4 algorithm,and the detection speed frames per second(FPS)reaches 42.5 frame/s,which is much higher than the 18 frame/s of YOLOv4 algorithm.The algorithm balances the relationship between accuracy and speed,and can detect the target on the warship in real time.