IoT-Enabled Embedded Sensing for Intelligent E-Waste Management: A Review of Classification, Traceability, and Hazard-Aware Sorting Architectures with a Proposed Edge-Native Integration Framework

Abstract

Global Electronic-Waste (e-waste) generation is rising far faster than the rate of formally documented collection and recycling, and most Electrical and Electronic Equipment (EEE) end-of-life streams are still sorted manually or diverted to informal, hazard-prone processing routes. Embedded sensing and the Internet of Things (IoT) offer a route to unit-level identification, hazard screening, and traceable reverse logistics that could close part of this gap, yet the underlying research is fragmented across waste-informatics, computer-vision, Radio-Frequency Identification (RFID), and battery-safety communities. This review synthesizes literature on (i) IoT-enabled bin- and facility-level e-waste monitoring, (ii) embedded and edge machine-learning classification of e-waste streams, (iii) RFID-and blockchain-based unit-level traceability, and (iv) embedded hazard sensing for battery-bearing e-waste, and proposes a conceptual edge-native architecture that unifies these strands into a single sorting node. A structured narrative review was conducted across IEEE Xplore, ScienceDirect, SpringerLink, Nature/Scientific Reports, and MDPI, screening publications on IoT waste-monitoring systems, embedded/TinyML classification models, RFID-based EEE traceability, and gas/thermal sensing for lithium-ion battery hazards. Peer-reviewed and technical-report sources are synthesized into comparative tables of embedded classification models, IoT-enabled e-waste and WEEE tracking systems, and existing work benchmarked against the proposed framework; accuracy–footprint trade-offs, a duty-cycled energy budget, a sensing-modality capability profile, and the global generation–recycling gap. The synthesis indicates that formally documented collection and recycling remain below approximately 25% of the more than 60 Mt of e-waste generated annually, that compressed on-device classifiers can retain 85%–90% accuracy at sub-300 kB footprints while narrowing the accuracy gap to cloud-class models to under 6% points, and that no reviewed system combines fill/presence sensing, on-device classification, unit-level identification, and hazard screening within a single power- and latency-constrained sorting node. This review identifies integration, together with standardized unit-level identifiers and connectivity-layer security, as the primary direction for future work.