Hybrid Squeeze-and-Excitation Convolutional Neural Network with Elastic Weight Consolidation for Longitudinal Learning in High-Accuracy Waste Classification

Abstrak

Waste management has become a global issue. Increased urbanization and per capita consumption have caused unprecedented garbage growth. Sustainability has always been about proper waste management within the ecological framework. Recently, numerous studies have been conducted on automating the identification of waste items. In this study, a Convolutional Neural Network (CNN) model equipped with Squeeze and Excitation (SE) module is proposed based on hybrid squeezing methods for waste item classification. The core aim of this research is to improve the accuracy of classification by highlighting intricate relations between various features encoded within the dataset. Based on extensive tests on a waste dataset, the CNN model with the SE module using hybrid squeezing outperforms all other models. The suggested method's 99.63% accuracy proves its efficacy and robustness. Furthermore, we incorporate Elastic Weight Consolidation (EWC) to enable longitudinal learning, allowing the model to adapt to emerging waste types (e.g., e-waste, biodegradable materials) while retaining prior knowledge with minimal forgetting (<1%). Ablation studies validate the critical role of hybrid squeezing, showing a 1.5% accuracy drop when spatial-wise components are omitted. This revelation affects automated recycling, waste sorting, and intelligent waste management. The proposed technology's accuracy shows its applicability and dependability, advancing sustainable waste management. By automating waste classification with unprecedented precision, the proposed framework can reduce landfill reliance, enhance recycling rates, and inform policy decisions for sustainable urban planning.