GreenSnap AI: An AI-enabled geo-verified waste reporting and verification system

Le résumé fourni par la source

Urban waste mismanagement remains a serious environmental and public-health challenge, particularly in rapidly growing urban and semi-urban regions. This paper presents GreenSnap-AI, an open-source mobile and web-based software platform for AI-assisted, geo-verified waste reporting and cleanup verification. The system enables citizens to submit geo-tagged waste images through a React Native mobile application, while supervisors and administrators manage verification, cleanup, and monitoring through role-specific mobile and web interfaces. The backend, implemented using Node.js/Express and MongoDB, coordinates user authentication, report management, AI-based image validation, and Haversine-based geospatial verification. Image validation is performed through decoupled AI inference services using MobileNetV3-Large and YOLOv11 models, allowing model replacement without major changes to the core application logic. A 10-meter geo-fence is used to verify that cleanup actions are performed at the reported location. Evaluation on a custom binary waste-presence verification dataset containing 17,556 labeled images using a held-out test subset demonstrated strong performance, with YOLOv11 achieving approximately 99.5% accuracy under the defined evaluation protocol. This result reflects the effectiveness of the model for binary waste validation within the evaluated deployment scenario rather than fine-grained waste recognition. Field testing in Sagri Village, Rawalpindi, demonstrated the practical feasibility of the platform for citizen-led reporting and municipal supervision. The main contribution of GreenSnap-AI is not limited to image classification; rather, it provides a reusable software workflow that integrates visual validation, spatial accountability, role-based access control, and administrative monitoring for transparent waste-management operations.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.