A Machine Learning Based Smart Contact-less pH Sensing and Classification

Abstract

With the ever increasing world population, there is a critical need for healthy food resources. Fish are the most environmentally-friendly animal protein to produce, efficiently converting feed into meat while generating a fraction of the greenhouse gasses of livestock production. Therefore, fish farming is one of the most important fields for a sustainable future. Since there is no way for fishes in fish farming pools to migrate into healthier water, a key factor in this industry is to maintain the water quality in standard conditions. Out of different key measurements used to quantify water quality, pH is among the essentials. In this study a portable, cheap, non contact, reusable, and machine learning-based pH sensing system is introduced. This helps farmers to quantify the pH quality of their pools without spending significant amounts of money on measurement equipment. This work introduces a sensitive, non-invasive and reflection-based optical sensor along with an Autoencoder-ESN framework for pH sensing. Using the Autoencoder guarantees at least 5 percent better classification in comparison with simple Echo State Networks. Long lifetimes of the sensor along with high sensitivity of the machine learning algorithm makes this system valuable for local farmers.