The detection and recognition of lightweight underwater fish schools are accomplished using an improved version of YOLOv5s
In order to enhance the precision and effectiveness of detecting and identifying fish schools in underwater environ-ment,this paper proposes an improved DCG-YOLOv5s lightweight underwater fish school detection algorithm.Firstly,to enhance the feature extraction capability and recognition accuracy of the network model,we incorporated deformable convolution into the convolution layer of the Backbone network.Secondly,in order to further improve the recognition performance of the model for underwater fish while simultaneously increasing its receptive field,we employed the lightweight upsampling operator CARAFE.Finally,GhostBottleneck was utilized to replace a portion of C3 structure in the original architecture,achieving a lightweight design without compromising accuracy.The results of the experiment indicate a noteworthy enhancement in the overall precision of detec-tion and computational efficiency of the enhanced model,thereby achieving a lightweight effect.
fish school target detectionYOLOv5s neural networklightweight algorithm
NETL
NSTL
万方数据
