Solid Waste Image Classification Using Deep Convolutional Neural Network

Separating household waste into categories such as organic and recyclable is a critical part of management systems to make sure that valuable materials are recycled utilised. This beneficial human health the environment because less risky treatments used at landfill and/or incineration, ultimately leading improved circular economy. Conventional separation relies heavily on manual objects by humans, which inefficient, expensive, time consuming, prone subjective errors caused limited knowledge classification. However, advances in artificial intelligence research has led adoption machine learning algorithms improve accuracy classification from images. In this paper, we dataset evaluate performance bespoke five-layer convolutional neural network when trained with two different image resolutions. The publicly available contains 25,077 images categorised 13,966 11,111 waste. Many researchers have same their proposed methods varying results. these results not directly comparable our approach due fundamental issues observed method validation approach, including lack transparency experimental setup, makes it impossible replicate Another common issue associated high computational cost often development prediction model size. Therefore, lightweight level methodology particular importance domain. To investigate issue, resolution sizes (i.e., 225×264 80×45) explore terms time, size, predictive accuracy, cross-entropy loss. Our intuition smaller will lead relatively than higher resolution. absence reliable baseline studies compare loss, random guess classifier show small leads lighter training produced (80.88%) better 76.19% yielded larger model. Both large models performed 50.05% accuracy. encourage reproducibility results, all artifacts preprocessed source code experiments made public repository.