Faster R-CNN for Marine Organism Detection and Recognition Using Data Augmentation

Abstract

Recently, Faster Region-based CNN(Faster R-CNN) has achieved marvelous accomplishment in object detection and recognition. In this paper, Faster R-CNN is applied to marine organism detection and recognition. However, the training of Faster R-CNN requires a mass of labeled samples which are difficult to obtain for marine organism. Therefore, three data augmentation methods are proposed dedicated to underwater-imaging. Specifically, the inverse process of underwater image restoration is used to simulate different marine turbulence environments. Perspective transformation is proposed to simulate different view of camera shooting. Illumination synthesis is used to simulate different marine illuminating environments. The performance of each data augmentation method, together with Faster R-CNN is evaluated by experiments on real world underwater dataset, which validate the effectiveness of the proposed method for marine organism detection and recognition.