Research And Implementation Of Citrus Canker Detection System

Citrus is the most important fruit of world, but its production is under the threaten of citrus canker. While biologic detection technologies are expensive, restricted and hard to satisfy the demands in China. In this paper we propose a method for fast detection of canker areas on citrus leaf images by using computer. The method is a combination of digital image processing, statistical learning and pattern recognition. We developed citrus canker auto detection system guided by this theory. The paper has studied on citrus canker detection technique, and the primary tasks of the paper are as follows:Discussed the feature extraction problem for citrus canker. We tried different color feature and texture feature and using them to recognize the citrus canker. We choose some effective features, which could be useful for other disease detection system.Proposed new cascade AdaBoost architecture to detect citrus canker. We introduce assistant discriminate function into cascade AdaBoost. After one sample is rejected by certain classifier at a layer, assistant discriminate function is used for further classification. The assistant discriminate function combines all the former classification results of this sample and gets a general classification result. The new cascade AdaBoost framework is tested in citrus canker detection and experiments show that it produces better results than other cascade AdaBoost framework.Proposed a windows combination hierarchy strategy for object detection. The strategy aims at the object which has distinct local characteristic such as citrus canker area. It first searches small area, and than combine the hit area to larger area for further searching. It runs much quickly because of the dramatically decrease of search space. Compared with popular moving window recognition and pyramid detection strategies, the proposed strategy is faster and gets better performance in citrus canker detection.Design the archetypal citrus canker detection system and realize it in order to implements the theory proposed in this paper. Detection system has two parts, one is for local detection, and the other is for remote detection. Remote detection subsystem can be used by experts, which can detect the citrus canker through the subsystem.In the very end, based on a summary of the research results, several questions for further research and exploration are proposed.