Enhanced convolutional neural network methodology for solid waste classification utilizing data augmentation techniques
Daniel Hogan Itam, Ekwueme Chimeme Martin, Ibiba Taiwo Horsfall
Abstract
• Cleaning, normalizing, and resizing solid waste image dataset to increase its variability and robustness. • Establishing a robust training that incorporates both traditional preprocessing methods and novel augmentation strategies. • An enhanced convolutional neural network (CNN) methodology that effectively classified solid waste into distinct categories. The increasing volume of solid waste generated globally necessitates efficient classification systems to enhance recycling and waste management processes. Convolutional Neural Networks (CNNs) have emerged as a powerful tool for image classification tasks, including solid waste identification. However, difficult external variables including changes in illumination, occlusion, and background clutter can have a big impact on CNN performance. Furthermore, pooling procedures frequently cause classic CNNs to lose spatial information, which might impair performance on tasks requiring extremely fine sense of place. This paper presents a comprehensive study on the application of an improved CNN-based models for solid waste classification. In the present study we explored image data resizing, augmentation technique and hyperparameter tuning to improve the performance of the proposed model. The results demonstrate that the improved-CNN model achieved high accuracy of 94.40 % compared to the conventional CNN and other deep learning model such as ResNet-50, Inception-V3 and VGG-19 (81.83, 66.67, 52.83 and 56.00 %).