Feature-based transfer learning to train a novel cotton imaging system

In recent years, the transfer learning framework has gained increasing interest in the machine learning community. Fundamentally, this framework aims to train a new target system using existing data or knowledge from one or more previous source systems. By extending the theory of standard machine learning techniques, this framework allows us to solve many challenging problems directly and intuitively. This paper presents an application of this framework to train a novel target system whose goal is to measure a cotton fiber property named maturity using image analysis. In addition, this paper also presents a feature-based supervised domain adaptation approach named G2DA which performs mapping using the generalized (kernel) discriminant analysis. After domain adaptation is complete, model estimation is performed easily using traditional machine learning algorithms. Specifically, RANSAC-based regression is performed to learn a maturity function for the target system. This function is then used to estimate the maturity of any newly scanned fiber. Validation studies performed show good results for our overall approach.