SceBoost Learning Algorithm for Feature Selection

This paper proposes an improved boost learning algorithm, the SceBoost algorithm, and its application in developing fast and robust features for citrus canker detection by machine vision. The algorithm use symmetric cross entropy to eliminate redundancy among selected features using AdaBoost algorithm. Selected features are subjected to recognize citrus canker symptoms on given pictures of citrus foliage. Compared with related feature selection algorithm our method can get improvements in classification accuracy and significantly reduce computation time when reach the same requirements.