Data Augmentation Method of Object Detection Using Region-Based Convolutional Neural Network in Maritime Imges

Abstract

The fundamental task of maritime transportation surveillance and autonomous ship navigation is to construct a reachable visual perception system that requires high efficiency and high accuracy of marine object detection. Therefore, high-performance deep learning-based algorithms on marine-related datasets were needed. Deep learning methods, however, require a lot of data, but lack a publicly available dataset for object detection in the maritime domain. In this paper, data augmentation method of object detection using region-based convolutional neural network (R-CNN) in maritime image is presented. The R-CNN is introduced for object detection that uses the SVM (Support Vector Machine) algorithm for classification. Firstly, images acquired and captured from the maritime dataset. A data augmentation method is used in this that can automatically extend the object detection dataset in maritime image. Extract the mask of the foreground object and combine it with the new background to automatically generate the location information and data of the object. Next, from acquired images, 70% are used as training data for the basic R-CNN model and the remainder as testing. Through the presenting method, high quality data can be learned by configuring various limited data features and experimental results show that the promising accuracy can be acquired in object detection.