This abstract describes the initial results of identifying foreign bodies in drinking water with the use of a machine learning algorithm in conjunction with portable microscopy. To do so, a magnified image is fed into an image analysis and machine learning program that makes use of a convolutional neural network (CNN). The algorithm will segment the image into parts and process them with multiple filters to determine qualitative information about the object, such as shape. During training, the CNN will adjust parameters (weights) to increase prediction accuracy. Initial results show the capability of identifying foreign bodies in distilled water, and further images are being prepared to test viability of identifying objects in tap water, and identifying hazardous materials, along with the quantity, density, and potential danger of the foreign body in accordance with government and private consensus on exposure to certain substances. If possible to replicate with minimal equipment and a developed machine learning algorithm, this method of foreign body identification could provide a method especially useful and simple for detecting potential contaminants and pollutants to homeowners, travelers abroad, and those in natural disasters or emergency survival situations.
