딥 러닝을 활용한 해상 선박 및 부유물 탐지 및 식별 기술

Abstract

We focus on detecting and identifying floating matters and vessels using deep learning. To this end, an aircraft equipped with a hyperspectral image sensor is used, and image data of 127 spectra are collected. In this paper, vessels and floating matters are detected and identified by using three spectra among 127 spectra, as a base technology that detects and identifies vessels and floating matters with 127 spectra. For dataset construction, we labeled all the vessels and floating matters in the images collected. As a result, it is difficult to detect all vessels and floating matters because the resolution of the images is not high enough to detect vessels and floating matters. However, most of the vessels andfloating matters are detected by deep learning-based approach. In the future work, more spectra are utilized to improve the accuracy of vessels and floating matters detection.