Segmenting Ships in Satellite Imagery With Squeeze and Excitation U-Net

TL;DR

This paper introduces a modified U-Net architecture with squeeze and excitation modules, optimized for satellite ship segmentation, achieving comparable performance with improved computational efficiency.

Contribution

Proposes a Squeeze and Excitation U-Net with IoU-optimized loss for efficient ship segmentation in satellite imagery.

Findings

Achieves comparable segmentation performance to complex models

Offers a computationally efficient alternative

Effectively optimizes IoU score during training

Abstract

The ship-detection task in satellite imagery presents significant obstacles to even the most state of the art segmentation models due to lack of labelled dataset or approaches which are not able to generalize to unseen images. The most common methods for semantic segmentation involve complex two-stage networks or networks which make use of a multi-scale scene parsing module. In this paper, we propose a modified version of the popular U-Net architecture called Squeeze and Excitation U-Net and train it with a loss that helps in directly optimizing the intersection over union (IoU) score. Our method gives comparable performance to other methods while having the additional benefit of being computationally efficient.