A Convolutional Neural Network for Heterogeneous Ship Images Classification

Abstract

In this work, heterogeneous ship images classification has been implemented by using Deep Learning techniques. The main goal of our study is to build a classification model that performs well on data collected from two different sensors i.e. optical sensor and radar sensor. We have proposed a binary classification solution. Our goal was to separate the ship images from the others. We selected a convolutional neural network (CNN) as our classification method and SoftMax as the prediction method. A CNN model was trained from scratch during which we obtained an accuracy score of 97.16%. We also used techniques such as transfer learning and fine-tuning to improve the previous accuracy. We finally obtained 99.30% as training accuracy. During the testing phase, 93.06% was recorded as the best performance.