Ships Target Detection on Water Based on Improved Faster RCNN

Abstract

AbstractThe intelligent research of watercraft detection is of great significance to the construction of marine transportation industry. Based on the Faster RCNN algorithm, this paper proposes a watercraft target detection algorithm based on the improved Faster RCNN. Due to the serious loss of low-level feature information during feature extraction. The fusion path between high-level and low-level features is longer, which increases the difficulty of information positioning. Specially increase the side fusion path network to strengthen the fusion of low-level features and high-level information from bottom to top. In view of the situation where the gradient of the Intersection over Union (IoU) bounding box loss is 0 and the intersection method not judged, the Distance-IoU loss function is introduced to improve the process of NMS filtering out repeated target boxes, improve the accuracy of position regression and improve the missed and false detection situations. Experiments show that the improved Faster RCNN algorithm in different weather environments has mean Average Precision (mAP) of 86.12% on ship images, which is an increase of 13.24% compared with the Faster RCNN algorithm, and the average IoU has increased by 6.94%. The overall detection accuracy is rate is improved.KeywordsShips target detectionFaster RCNNSide fusion path networkDistance-IoU