Maritime Ship Detection using Convolutional Neural Networks from Satellite Images

The significance of efficient monitoring and control of marine traffic for the purpose of safety and security of the ships cannot be overemphasized in the current scenario where global trade and commerce is at its pinnacle. Various stakeholders are concerned with serious maritime issues related to hijacking of ships, illegal fishing, encroachments of sea borders, and illicit exchange of sea cargo, accidents, and military attacks. This requires an automated, accurate, fast, and robust sea monitoring system which can avoid or mitigate the negative effects of such issues. This paper proposes, implements, and evaluates a CNN based deep learning model which can accurately identify ships from the images captured from satellite images. Two models CNN Model 1 and CNN Model 2 having different architectures are trained, validated, and tested on the Airbus satellite images dataset. Both classification accuracy and loss functions are measured by varying the number of the epochs. Also, the complexity comparison of the two models is performed by measuring the training time. The paper concludes that the proposed models are automatic, fast and accurate in terms of their performance on the Airbus dataset by achieving a maximum accuracy of 89.7%.