EfficientNet-Based Deep Learning Model for Advanced Waste Classification

One of the main obstacles to recycling systems in developing countries like Bangladesh is waste classification. About 22.4 million tons of garbage are produced in the nation each year, of which 70% is organic waste and 30% is nonorganic. Regarding this unmanaged situation, it is critical to build an autonomous smart garbage sorting system that makes use of intelligent capabilities. Waste classification from waste images is a crucial computer vision problem that must be solved by developing an intelligent system. The accuracy and efficiency of conventional waste classification techniques are not that remarkable. First, a dataset of 5012 waste images is generated from nine distinct classes—including inorganic and organic waste. The goal of the proposed study is to develop an automatic image classifier that can identify objects and determine the kind of waste material in order to increase the efficiency and accuracy of waste categorization processes. Feature extraction from photos is done in this work using pre-trained EfficientNet models (EfficientNet B0, EfficientNet B2, EfficientNet B3, and EfficientNet B4) on ImageNet. This allows the classifier to provide predictions and differentiate the type of garbage from its related category. EfficientNet B0, EfficientNet B2, EfficientNet B3, and EfficientNet B4 models have been employed appropriately for this experimental dataset based on the EfficientNet pre-trained models. Based on the experimental results, it can be seen that EfficientNet B0 and B2 both achieve 93% accuracy through transfer learning. Furthermore, EfficientNet B3 and EfficientNet B4 models surpass earlier image classification methods with 91% classification accuracy.