New Calibration Method for Implementing Machine Learning in Low-Cost Sensor Applications

Abstract

Hardware cost and manufacturing process variation constrain the use of machine learning in low-cost sensor applications. This letter describes a novel method to manage some of those critical limitations. Reference calibration mapping is a method that creates a reference space from a single sensor and then transforms the output from the remaining sensor population into that reference space. The method results in the ability to utilize low-cost hardware and reduce the required training set. This letter applies the method to a media sensing system in a laser printer application. The resulting system lowered component and development costs while meeting stringent manufacturing and performance requirements