Ship-Detection-in-Optical-Satellite-Imagery

This project proposes a processing pipeline for automated detection of ships in optical satellite imagery. Dataset from Kaggle’s Airbus Ship Detection Challenge is used for the project. This dataset has 190,000, 768 x 768 pixel images with complex backgrounds of clouds, shore lines, waves and ship-wakes. The large field of view images in the dataset makes saliency detection a necessary first step. The saliency map provides several candidates that can be tested with a classifier. Several feature descriptors are explored, compared and combined to determine the best method for ship detection. Principle Component Analysis (PCA) is employed to reduce the feature size and its results are also investigated. This project achieves a 82% classification score with classical pattern recognition methods.

Contributors: Gaurav Shalin, Saad Rasheed Abbasi and Souvik Roy