Entropy-Centric Explainable AI for Remote Sensing Image Segmentation

Explainability is crucial for AI adoption in high-stakes fields: The paper identifies that deep learning models used in remote sensing achieve high performance but lack transparency in how they make decisions. This "black-box" nature creates trust barriers that prevent wider adoption of AI in critical domains where decisions have real-world consequences.

Image segmentation explanation lags behind other AI tasks: While researchers have made progress explaining AI decisions in image classification, the paper notes that "image segmentation still offers considerable room for improvement," suggesting that current explainability methods are not equally effective across different types of AI applications.

New evaluation methods are needed to assess explanation quality: The paper proposes a novel methodology to measure whether the regions highlighted by explainability tools are actually relevant to the AI's decisions. This addresses a practical problem: without proper evaluation methods, we cannot reliably verify that explanations are truthful and meaningful.

Entropy-based approaches show promise for transparency: The proposed entropy-centric method demonstrated superiority over existing explainability approaches for segmentation tasks, suggesting that information theory concepts can help make AI decision-making more interpretable and trustworthy in remote sensing applications.