Developing a Deep Learning-Based Mobile Application to Detect Solid Wastes

The main purpose of this study is to detect solid waste through a mobile application using a deep learning-based model. However, some features must be compromised to use a model developed using deep learning on mobile devices. Multi-layered and high-dimensional deep learning models cannot perform well on devices with low hardware capacity such as mobile or embedded devices. The MobilnetV2-SSD model proposed in this study performs better in embedded systems and mobile applications due to its light structure. The deep learning model trained with the TACO dataset was able to detect and classify solid wastes according to classes such as metal, plastic, and cardboard. The developed deep learning model detected solid wastes with an accuracy value of 44%. In future studies, increasing the sample size in the training dataset and reducing the number of classes will contribute to increasing the accuracy value.