Architecture Overview

LiteFODNet is designed for small foreign object debris detection in runway surveillance imagery, where targets are visually diverse, sparse, and often difficult to separate from complex backgrounds. The architecture introduces CMSP for compact multiscale context, SCR for efficient downsampling, FFM for channel-level feature calibration, and SPA for spatial attention along separate axes. Together, these modules guide the detector toward small, ambiguous objects while controlling computational burden. The model is positioned for aviation safety monitoring where real-time inference and reliable small-object localization are critical.

Architecture Diagram