Computer vision based medical waste separator

Abstract

In this study we have aimed to build an automatic computer vision based medical waste separator that detects the presence of medical waste and categorizes them into one of the four categories namely gloves, mask, syringe and cotton. This embedded system uses transfer learning on the AlexNet deep learning network to train a model which classifies medical waste. The hardware set up is designed to detect the movement of the lid of the input bin and then capture an image of the waste object dropped into the bin. This image is then fed to our trained model which classifies the object with a 86.17\% validation accuracy. Once the model classifies it, the waste object is dropped into the correct bin with the help of servo motors. This embedded system has been tested with different types of gloves, mask, syringe and cotton samples and presents a convenient way to segregate medical waste successfully. Thus this system aims to eliminate the need for manual labour to segregate hazardous medical waste and avoids harmful exposure to contaminated waste.

Keywords: Computer vision \and medical waste, waste separation,convolutional neural network, transfer learning

JEL Classification: i19