Convolutional neural networks for optical image recognition of ships

Abstract

Object and purpose of research. This paper discusses classification of ships based on their optical, infrared and radar images to estimate the potential of modern computer-based vision algorithms and approaches towards the accomplishment of this task. Materials and methods. The paper analyses the evolution of computer hardware and software that make machine vision and machine learning solutions suitable for image-based recognition of ships. It describes main approaches to shipborne radar developments in various countries, outlining the two ways of making the classification models based on networks “trained” for image classification. Main results. Analysing the progress in convolutional neural networks, this paper outlines the main trends in machine vision developments, points out their most promising applications in ship image classification and suggests two classification models for optical ship images with quantitative estimates of recognition accuracy. Conclusion. The findings of this study outline the trends in ship image classification technology and could be helpful in infrared and radar imagery recognition.

Keywords

• neural network
• machine vision
• image processing
• recognition of objects
• dataset
• computer-based vision