A Hybrid Model for Dry Waste Classification using Transfer Learning and Dimensionality Reduction

Abstract

The categorization of waste plays a crucial role in efficient waste management, facilitating the recognition and segregation of various waste types to ensure appropriate disposal, recycling, or repurposing. With the growing concern for environmental sustainability, accurate waste classification systems are in high demand. Traditional waste classification methods often rely on manual sorting, which is time-consuming, labor-intensive, and prone to errors. Hence, there is a need for automated and efficient waste classification systems that can accurately categorize waste materials. This research introduces an innovative waste classification system that merges feature extraction from a pre-trained EfficientNet model with Principal Component Analysis (PCA) to reduce dimensionality. The methodology involves two main stages: (1) transfer learning using the EfficientNet-CNN architecture for feature extraction and (2) dimensionality reduction using PCA to reduce the feature vector dimensionality. The features extracted from both the average pooling and convolutional layers are combined by concatenation, and subsequently, classification is performed using a fully connected layer. Extensive experiments were conducted on a waste dataset, and the proposed system achieved a remarkable accuracy of 99.07%. This outperformed state-of-the-art waste classification systems, demonstrating the effectiveness of the combined approach. Further research can explore the application of the proposed waste classification system on more extensive and more diverse datasets, optimize the dimensionality reduction technique, consider real-time implementation, investigate advanced techniques like ensemble learning and deep learning, and assess its effectiveness in industrial waste management systems.