改进视觉注意力网络的水下目标检测算法

Improved visual attention network underwater target detection algorithm

针对传统水下图像检测方式易受水中光线、复杂环境的影响,造成水下目标识别精度不高,泛化性较差的问题,提出了一种改进YOLOv5s的水下目标检测算法。首先引入了Double MSRCR算法,解决了水下物体清晰度低,特征模糊的问题。在网络结构方面,主干网络引入C_VAN模块,提升了神经网络特征提取能力;其次在颈部网络中,引入RFB_S感受野,增强神经网络的多尺度适应能力;最后引入NAMAttention空间与通道注意力机制,增强网络上下文特征的表达能力。所提方法相较于Faster-RCNN检测精确度提高了6.5%,相较于YOLOv4检测精确度提高了4.1%,相较于YOLOv5s检测精确度提高2.7%,检测速度提升了56.34 fps,证明了方法适用于实时水下检测任务。

Aiming at the problem that the traditional underwater image detection method is easily affected by the light and complex environment in the water,resulting in low accuracy and poor generalization of underwater target recognition,an improved YOLOv5s underwater target detection algorithm is proposed.Firstly,a bilateral MSRCR algorithm is proposed to solve the problem of low clarity and fuzzy features of underwater objects.In terms of network structure,the C_VAN module is introduced into the backbone network,which improves the feature extraction ability of the neural network.Secondly,RFB_S receptive field is introduced into the neck network to enhance the multi-scale adaptability of the neural network.Finally,NAMAttention space and channel attention mechanism are introduced to enhance the expression ability of network context features.The proposed method improves the detection accuracy by 6.5%compared with Faster-RCNN,4.1%compared with YOLOv4,and 2.7%compared with YOLOv5 s.The detection frame rate is improved by 56.34 fps,which proves that the method is suitable for real-time underwater detection tasks.