ONDL: An optimized Neutrosophic Deep Learning model for classifying waste for sustainability

Area of Science:

• Environmental Science
• Computer Science
• Artificial Intelligence

Background:

• Sustainability is crucial for a greener planet, with effective waste classification and management playing a vital role.
• Computer algorithms and deep learning offer advanced solutions for waste management challenges.

Purpose of the Study:

• To propose an Optimized Neutrosophic Deep Learning (ONDL) model for accurate waste object classification.
• To evaluate the ONDL model's performance on two distinct waste datasets (DSWM1 and DSWM2).

Main Methods:

• The ONDL model utilizes Deep Transfer Learning (DTL) based on Alexnet, incorporating True (T) neutrosophic domain conversion.
• Grey Wolf Optimization (GWO) is employed for efficient image feature selection within the ONDL architecture.
• Comparative analysis involved testing various DTL models (Alexnet, Googlenet, Resnet18) and neutrosophic domains (T, I, F).

Main Results:

• The ONDL model demonstrated superior efficiency compared to other tested models.
• On DSWM1 (2 classes), ONDL achieved Testing Accuracy (TA) of 0.9189, Precision (P) of 0.9177, Recall (R) of 0.9176, and F1 score of 0.9177.
• On DSWM2 (3 classes), ONDL achieved TA of 0.8532, P of 0.7728, R of 0.7944, and F1 score of 0.7835, showing competitive results.

Conclusions:

• The ONDL model is an effective deep learning approach for waste classification.
• The proposed model achieves competitive performance metrics, contributing to improved waste management sustainability.
• The study highlights the potential of neutrosophic deep learning in addressing environmental challenges.