Efficient Remote Sensing under Real-world Constraints

The School of EECS is hosting the following HDR Progress Review 3 Seminar:

Efficient Remote Sensing under Real-world Constraints

Speaker: Yiyun Zhang

Abstract: Remote sensing systems are increasingly expected to support practical decision-making in settings such as urban monitoring, disaster response, satellite data transmission, and visual geo-localisation. Yet real-world deployment is rarely as clean as standard benchmarks assume. Labels can be scarce, multispectral data can be expensive to transmit, and ground-level observations may only provide a narrow field of view rather than a complete panorama.

This talk presents a unified research direction on efficient remote sensing intelligence under practical constraints. It covers foundation-model-guided approaches for reducing annotation requirements in satellite image understanding, bandwidth-aware benchmarking and neural reconstruction for multispectral satellite image delivery, and cross-view geo-localisation reasoning under limited visual observations. A key theme is that remote sensing models should not only perform well under ideal inputs, but should also remain useful when evidence, bandwidth, and supervision are limited. The broader goal is to move from idealised remote sensing benchmarks toward deployable systems that can operate reliably under realistic constraints.