Explainable Analysis of Deep Learning Methods for SAR Image Classification

TL;DR

This paper applies explainable AI techniques to deep learning models for SAR image classification, providing insights into model decisions and evaluating explanation methods' effectiveness.

Contribution

It introduces the use of multiple XAI methods to interpret CNN-based SAR image classifiers, highlighting Occlusion's effectiveness and limitations.

Findings

Occlusion provides the most reliable interpretation in terms of Max-Sensitivity.

Explanation heatmaps vary in resolution and reliability.

Insights into CNN decision mechanisms for SAR images.

Abstract

Deep learning methods exhibit outstanding performance in synthetic aperture radar (SAR) image interpretation tasks. However, these are black box models that limit the comprehension of their predictions. Therefore, to meet this challenge, we have utilized explainable artificial intelligence (XAI) methods for the SAR image classification task. Specifically, we trained state-of-the-art convolutional neural networks for each polarization format on OpenSARUrban dataset and then investigate eight explanation methods to analyze the predictions of the CNN classifiers of SAR images. These XAI methods are also evaluated qualitatively and quantitatively which shows that Occlusion achieves the most reliable interpretation performance in terms of Max-Sensitivity but with a low-resolution explanation heatmap. The explanation results provide some insights into the internal mechanism of black-box decisions for SAR image classification.