Predicting Response of Printed Potentiometric Nitrate Sensors Using Image Based Machine Learning

Abstract

Solid-contact nitrate sensors have wide applications in agriculture. In manufacturing, fabrication is an essential step and strongly affects the sensor performance. We focus on controlling the fabrication process to develop an economical thin-film nitrate sensor with an ion-selective membrane (ISM). However, direct long-time measurement of sensor performance for monitoring fabrication is expensive and costs human labor. Thus, in this work, we propose an automatic system to predict the temporal potentiometric response based on non-contact images acquired in real time. Our prediction systems are generated by exploiting image-processing techniques and machine learning approaches. To improve the prediction accuracy, we also fuse manufacturing factors to the image inputs. The comparison of prediction performance with different inputs also helps us to understand their effects on the fabrication process.