Developed a deep learning model to detect ships in satellite images using Mask R-CNN for maritime surveillance.

In this project, I developed a computer vision model to automatically detect ships in satellite imagery. The goal was to support maritime surveillance and monitoring efforts by leveraging deep learning techniques for object detection. 

I used the Airbus Ship Detection Challenge dataset, which consists of satellite images with labeled ship masks. The pipeline included extensive preprocessing such as image resizing, data augmentation, and handling class imbalance due to a large number of images without ships. 

For the model, I implemented Mask R-CNN to perform instance segmentation, enabling precise localization and boundary prediction for ships. The model was trained using TensorFlow/Keras (or PyTorch, depending on what you used), with custom evaluation metrics like Intersection over Union (IoU) and precision recall to assess performance. 

Post-training, I tested the model on unseen satellite images and achieved accurate ship detection, even in challenging conditions like cloud cover, varying ship sizes, and overlapping objects. 

This project demonstrates the application of deep learning in remote sensing, with potential use cases in defense, cargo tracking, illegal fishing detection, and maritime traffic management.