EcoNet-Multi: optimising deep learning frameworks for robust waste detection in diverse global environments

In the current era of technological advancement, the increasing demand for automated environmental monitoring and smart recycling systems plays a crucial role in maintaining a healthy environment for both humans and nature. Deep learning and artificial intelligence-based waste classification have attracted significant attention for effective waste segregation. In the existing literature, several deep learning algorithms are available for image classification in waste management systems. However, hyperparameter optimisation remains a dual challenge: maximising classification accuracy whilst preserving balanced class detection across highly imbalanced waste categories. Addressing these challenges, this paper proposes a Surrogate-Assisted Hybrid Multi-Objective Binary Bat with Cuckoo Search algorithm, termed SMOBBCS (Multi-Objective Binary Bat-Cuckoo Search), which simultaneously minimises classification error and the complement of the macro-F1 score for waste category detection. The algorithm fuses the inertia-weighted frequency-modulated velocity update of the Binary Bat Algorithm with the Lévy-flight perturbation mechanism of Cuckoo Search, governed by a smooth adaptive transition gate that facilitates population evolution from global exploration to local exploitation. A lightweight k-NN surrogate model reduces the number of true fitness evaluations without sacrificing solution quality. Experimental results validate the proposed approach against five state-of-the-art surrogate-assisted multi-objective optimisers (SMEPO, SMWOA, SMBCS, SMPSO, and SMBAT) across four real-world waste detection benchmarks: HGCD, TrashNet, WasteNet, and Realwaste. Four deep learning classifiers (CNN, DNN, GRU, and CNN-DNN) are employed for comprehensive evaluation. The proposed SMOBBCS method achieves mean accuracy of 92.71%, mean Macro-F1 score of 0.9209, mean precision of 0.9225, mean recall of 0.9205, and mean Matthews Correlation Coefficient (MCC) of 0.9052, demonstrating superior performance in automated waste classification tasks.