SpectraNet: A unified deep learning framework for infrared spectroscopy-based prediction of plastic recyclability, type classification, and microplastic identification

As global plastic pollution and microplastic contamination intensify, efficient plastic triage for recycling, material identification, and microplastic monitoring is critical for environmental sustainability. To support this effort, we systematically synthesize findings from interdisciplinary studies and establish an open-access infrared spectral database for plastics and microplastics, serving as a foundational resource for the scientific community. Building on this, we present SpectraNet, an innovative deep learning framework that integrates mid-infrared (MIR) spectroscopy with advanced algorithms to support three critical analytical tasks: (1) plastic recyclability assessment; (2) plastic type identification; (3) microplastic type identification. It achieves 92.63 % accuracy for recyclability classification, 95.06 % for microplastic type identification, and 95.86 % for microplastic recognition. On the private test set (Data6), SpectraNet achieved over 98 % accuracy, demonstrating excellent generalization. By precisely identifying high-risk plastic types such as PVC and PS, SpectraNet provides a practical tool for environmental hazard detection, microplastic exposure assessment, and toxicological risk prioritization in contaminated ecosystems, underscoring its robustness under spectral variability and strong potential for scalable deployment in recycling, environmental diagnostics, and real-time Internet of Things (IoT)-based microplastic monitoring. These quantitative results confirm the model's robustness under spectral variability and its strong potential for scalable deployment in industrial recycling, environmental diagnostics, and real-time microplastic monitoring via industrial IoT systems.