Vision-Based Waste Classification for Handheld PET Bottles and Paper Using Convolutional Neural Network and Transfer Learning

Most object classification models use steady images in a fully controlled environment as a dataset to get the highest possible accuracy. Existing classification systems aligned with waste materials also have generalized outputs ie. biodegradable or nonbiodegradable. This study aims to integrate images dataset of handheld PET bottles and paper wastes on deep learning methods CNN, YOLOv5, and ResNet50. Mentioned transfer learning techniques are evaluated and compared to conclude which of them produce the highest classification accuracy. The tests conducted resulted in a similar result of 100<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup> classification accuracy for all 3 models in consideration of the 2,400 image datasets equally divided between the two proponent materials, wherein 1,980 are designated as training data, while 400 are testing data.