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

Area of Science:

• Artificial Intelligence
• Environmental Science
• Computer Science

Background:

• Automated environmental monitoring and smart recycling are crucial for ecological health.
• Deep learning for waste classification aids effective waste segregation.
• Existing methods struggle with hyperparameter optimization for imbalanced waste data.

Purpose of the Study:

• To propose a novel algorithm, SMOBBCS, for optimizing waste classification.
• To address the dual challenge of maximizing accuracy and maintaining class balance in waste detection.
• To improve hyperparameter optimization for imbalanced waste categories.

Main Methods:

• Developed a Surrogate-Assisted Hybrid Multi-Objective Binary Bat with Cuckoo Search (SMOBBCS) algorithm.
• Fused Binary Bat Algorithm's velocity update with Cuckoo Search's perturbation mechanism.
• Utilized a k-NN surrogate model to reduce fitness evaluations and employed deep learning classifiers (CNN, DNN, GRU, CNN-DNN).

Main Results:

• The SMOBBCS algorithm achieved a mean accuracy of 92.71% and a mean Macro-F1 score of 0.9209.
• Demonstrated superior performance compared to five state-of-the-art surrogate-assisted multi-objective optimizers.
• Achieved high mean precision (0.9225), recall (0.9205), and MCC (0.9052) across four benchmarks.

Conclusions:

• The proposed SMOBBCS method significantly enhances automated waste classification accuracy and class balance.
• This approach offers a robust solution for hyperparameter optimization in imbalanced waste datasets.
• SMOBBCS shows great potential for advancing smart recycling and environmental monitoring systems.