Optimization of Ship Target Detection Algorithm Based on Random Forest and Regional Convolutional Network

摘要

Target detection can assist in detecting the position of the target ship, which is an important part of the intelligent ship visual aid system. With the development and perfection of deep learning, the convolutional neural network technology has been continuously optimized. And it can automatically learn and extract features of objects in images, providing stronger distinguishing power and representation ability. In this paper, various optimization algorithms of convolutional neural networks are compared. Aiming at the problem of unbalanced ship targets in remote sensing images of near-port areas, a ship target detection algorithm based on random forest and Faster-RCNN is proposed. The random forest algorithm is used for model optimization due to its insensitivity to multi-collinearity. The positive effect of the optimized algorithm on accuracy is verified through experiments.