Detection of the Floating Objects on the Water Surface Based on Improved YOLOv5
Xiaoqian He, Jingcheng Wang, Chaobo Chen, Xueqin Yang
Abstract
An improved YOLOv5 surface floating object detection algorithm is proposed to solve the problems of low detection accuracy and slow speed in the detection of floating objects on the water surface. Firstly, the smooth label is introduced to suppress the over-fitting phenomenon in the training process of the network model. Secondly, the inception topology is used to enhance the feature extraction of floating objects, under the premise of ensuring network accuracy, the low number of parameters and the amount of calculation are maintained. Finally, the loss function of YOLOv5 is optimized to improve the speed and accuracy of prediction frame regression. Experimental results show that the improved mean Average Precision (mAP) of YOLOv5 reaches 96.48%, which is 16.9% higher than that of YOLOv5, and the detection rate up to 30fps. This detection strategy is effective and feasible.