OEVS-fusion: Olfactory-enhanced visual semantic recognition framework for ground stain detection in indoor environments

Abstract

Floor stains in indoor environments, such as pet households, stores, and public areas, often remain undetected, leading to health and hygiene issues. This challenge is exacerbated when conventional cleaning robots attempt to address these stains through repeated cleaning motions, inadvertently spreading contaminants and increasing the stain area. To address this, we propose a ground stain detection framework that integrates olfactory (smell) and visual semantic information by leveraging an electronic nose and a camera module. The approach involves separate networks for extracting image and gas features, followed by feature fusion into a combined representation, which is processed by a decision fusion network for final detection. Experimental results demonstrate a more than 15 % improvement in recognition accuracy over vision-only methods.