Longan's leaf analysis for chemical-substance usage

• Main Authors: Chouvatut V., Putanamatada K., Jindaluang W.
• Format: Conference or Workshop Item
• Published: 2015
• Online Access: http://www.scopus.com/inward/record.url?partnerID=HzOxMe3b&scp=84925868215&origin=inward; http://cmuir.cmu.ac.th/handle/6653943832/38915
• Institution: Chiang Mai University

© 2015 IEEE. Flower cluster in longan production can be separated into two types, one with young leaf underneath and the other without. Each type of cluster requires different treatments of chemical fertilizer. The method to differentiate the type of longan flower clusters is challenging, as the color tone of a young leaf and fully mature one are extremely similar. This paper presents image processing techniques to distinguish the types of longan flower cluster, immature or mature cluster. Furthermore, if the immature flower cluster has been detected, the leaves beneath the cluster must then be classified whether most of them are young or fully-grown leaves. Since the colors of longan's leaves are very similar and they are all green, the appropriate color space must be considered. Even in the cases with or without the fully grown flower cluster, the color space must also be carefully selected due to its very close color of the young leaves. After conversion into some appropriate color space for each process of classification, changes in environmental brightness or illumination is another issue which must be carefully concerned. Finally, the correctly retrieved information then helps to determine the required chemical substance. Several image enhancement methods adaptively chosen according to the requirements for each classification process are thus applied to adjust the tone contrast in order to create high contrast tone of longan leaf needed to analyze the type of longan flower cluster. In this paper, we applied easy and well-known algorithms which are appropriate for our purposes in each processing step while still provide perfect classification results. In our experiment, we use 150 images to separate the two types of flower cluster, which yields an accuracy of 100%.