Abstract
This research examines the utilization of sophisticated convolutional neural networks (CNN) for the automated categorization of plastic waste in order to augment recycling effectiveness. A comparative analysis of three CNN architectures - MobileNet, ResNet-34, and EfficientNet-B1 - was conducted, alongside the introduction of an innovative model, EfficientNet-SED, aimed at enhancing both accuracy and computational efficiency. Empirical evaluations utilizing a dataset comprised of plastic bottles revealed that EfficientNet-SED attained a classification accuracy of 99.1%, exceeding the performance metrics of the other models. These results underscore the potential of modern machine learning methodologies to refine waste sorting systems, thereby facilitating more efficient recycling practices and resource preservation.