DEEP LEARNING ALGORITHM IMPLEMENTATION FOR SHIP DETECTION IN SPOT SATELLITE IMAGES

• Main Author: HANIZAM, MOHD HAZIQ NAZMI
• Format: Final Year Project
• Language: English
• Published: IRC 2019
• Online Access: http://utpedia.utp.edu.my/20131/1/Dissertation.pdf; http://utpedia.utp.edu.my/20131/

Marine industry is a large industry especially in the economy sector. Not limited to commercial, this industry also includes naval sector and the small and medium industry of fisheries all over the world. The huge development throughout the industry has also develop many kinds of unlawful act such as piracy and illegal cargo transportation. This has become the call for action for the officials of the sovereignty area to monitor the activities to control the situation and prevent them from become an epidemic that effects the whole industry. In this study, we propose to implement a deep-learning approach for detection of ships on satellite images in various conditions. The deep-learning algorithm to be deployed is Faster R-CNN and to be implemented using MATLAB. The project is carried out with the objective to implement the algorithm on SPOT satellite images that can accurately localize the region of interest (ROI) of the ship. The implementation of the algorithm consists of three stages which are pre-processing, network training and accuracy evaluation. The output of this project will be the localization of ships within the image with confidence scores of the prediction. Based on the results obtained, the deployment of the Faster R-CNN algorithm on ship class objects from SPOT satellite images has achieved a noteworthy performance despite the limitations in the amount of training dataset and specifications of the machines used. We can conclude that the project was able to achieve its objectives within the stipulated timeframe.