Our multispectral satellite comprehension technology pioneers a quantum deep image prior framework and a quantum prism light-splitting mechanism, successfully overcoming the underdetermined blind source separation dilemma in multispectral images. Furthermore, it can accurately reconstruct the spectral signatures and abundances of materials in an unsupervised manner. Published in IEEE TIP/TGRS, our technology provides Taiwan＇s remote sensing technique with an intelligent analytical ability.

We pioneer a quantum deep image prior framework and quantum prism light-splitting mechanism, achieving unsupervised learning and transforming an underdetermined system into a solvable overdetermined system. Validated with ESA Sentinel-2 image, our technology achieves high-precision abundance estimation for highly mixed waterbodies. Experimental performance is about 50% better than the current best method, establishing Taiwan＇s global leadership in next-generation satellite intelligent analysis.

To address prohibitive data costs and hardware requirements in remote sensing industries, our team develops an unsupervised QDIP framework. The quantum prism light-splitting mechanism empowers multispectral sensors with hyperspectral-level identifiability, significantly reducing hardware expenses. Validated globally, our technology provides high-precision analysis for climate, agriculture, and mining, establishing a leading position in the practical application of quantum AI.