An approach of classifying waste using transfer learning method

One of the most critical issues facing the world is waste management, regardless of whether the region is being established or becoming established. There is a waste partitioning process in waste management, and the main challenge is that the garbage space is flooded long before the beginning of the following cleaning process at clear spots. Only unskilled workers conduct waste separation, which is less accurate, time-consuming, and not utterly possible due to the enormous amount of waste. Using the Convolutional Neural Network, we propose an artificial waste classification problem to compile and organize a dataset into seven categories consisting of metal, plastic, glass, paper, cardboard, trash, and E-waste. We then distinguished between specific transfer learning algorithms for our project: Xception, DenseNet121, Resnet-50, MobilenetV2, and EffiecienNetB7. DenseNet121 achieved a high precision characterization of about 93.3% for our model, while Mobilenet also demonstrated an incredible conversion to different forms of waste of 93% and Resnet-50, Xception and EfffiecienNetB7 achieved 92%, 92.5%, and 87%, respectively. In the future, we would like to increase the accuracy by using some other hyperparameter tuning, and we would like to deploy the project on mobile devices. We will use dockers or Kubernetes to deploy and YOLO real-time object detection as a framework for the post