Recognitions of collision avoidance for unmanned surface vessels with visible light camera using broad learning system
Xiaofeng Shan, Qihe Shan, Ruofei Man, Yi Zuo · 2022
Unmanned surface vessels(USVS) are equipped with various types of sensors responsible for the function of the surface vessel to sense the environment. Vision sensors have the advantages of direct recognition effects, low costs and wide recognition contents compared with obstacle avoidance sensors. Traditional recognition algorithms based on vision sensors often use deep learningapproaches, such as YOLO,Fast-RCNN,and SSD etc. Although these algorithms have high accuracy, they also have disadvantages such as complex network structure, high-performance computing equipment, and time-consuming training. In this paper, we propose a fast way to optimize the broad learning system(BLS) structure using the integration algorithm, and use the BLS as an individual learner,The k-fold cross validation approach is taken to avoid model overfitting. And Improv integration learning so as to vote integrate the recognition results of individual learners and output the final results. The experimental reaults that our method achieved a high accuracy on the dataset. compared with both of the single BLS and the traditional recognition algorithm.