Underwater fish detection in sonar image based on an improved Faster RCNN

Abstract

For the efficient detection of underwater fish, this paper proposes a target detection algorithm based on the improved Faster region-based convolutional neural network (iFaster RCNN). On one hand, the proposed algorithm combines feature pyramid network (FPN) with the original Faster RCNN for solving the multi-scale problem in target detection. On the other hand, in order to further enhance the detection accuracy and increase detection speed, Distance-Intersection-over-Union (DIoU) is used to replace Intersection-over-Union (IoU). Experimental results show that, with FPN and DIoU, iFaster RCNN has higher detection accuracy for underwater fish. For comparison purposes, VGG16, MobileNetV2, and ResNet50 netwoks are used as the backbone feature extraction networks of iFaster RCNN. Comparative results prove that ResNet50 performs better than the other two netwoks.