An assistant for an incremental learning based image processing system

Supervised learning has been applied in image processing system for object recognition, inspection and measurement. However the teaching-learning mode of supervised learning is not practical in real application, because it is impossible to teach a system all possible samples in one time. Therefore, incremental learning is considered to be a promising solution which supports the iteration of teaching-learning in cycles.

An incremental learning based system can always be taught and can learn new samples of objects. However, from engineering perspective, incremental learning is not so practical for user in a teaching-learning cycle. For this reason, an assistant is proposed to support teaching-learning cycles. The assistant includes the following four functions: “result monitoring”, “auxiliary teaching”, “incremental learning” and “classifier evaluation”. With the help of an assistant, system user is able to control the whole teaching-learning cycle, and interact with the image processing system easily. The concept of an assistant is tested by experiments of classifying agricultural products. It is proved that the assistant is a practical manner in image processing system.