HAT: Achieving Comprehensive Inpainting of Lunar Spectral Imagery
Dingruibo Miao, Depei Gu, Jianguo Yan, Zhigang Tu, Jean‐Pierre Barriot · IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing · 2025
Lunar spectral imagery plays an indispensable role in the analysis of geomorphological features and the study of mineral distribution. However, although the Selenological and Engineering Explorer (SELENE) mission of the Japan Aerospace Exploration Agency (JAXA) provides high-resolution lunar spectral imagery, the limited scan width of its imaging instrument leads to substantial data voids, significantly constraining the scientific utility of these data.To address this challenge, we propose an innovative image inpainting model, Hole-aware Transformer (HAT), which is designed to effectively capture long-range feature dependencies and deeply understand image contextual information. Moreover, with its unique Holeaware Multi-Head Attention mechanism (HAMA), the model can adaptively handle data voids of various sizes and shapes, avoid interactions with invalid feature(i.e., those within data missing regions), and hence generate completed images that are more visually and semantically realistic. Through comparative experiments with existing mainstream methods, our HAT model has demonstrated outstanding repair performance. Finally, based on the repair results of the HAT model, we have created a complete lunar spectral image for the first time, providing a new perspective and tool for lunar science research.