Identifikasi Tumbuhan Obat Menggunakan Fitur Citra Morfologi, Bentuk, dan Tekstur dengan Klasifikasi Probabilistic Neural Network

Abstract

Indonesia has no less than 2039 species of medicinal plants, but only 20-22% was cultivated, while around 78% obtained through direct collection (exploration) of forest. Generally, identification process of medicinal plants has been done manually by the herbarium taxonomist using guidebook of taxonomy/dendrology. It will be a difficult problem to identify directly it in the forest. Leave is usually used for identification because of its effectiveness and efficiency. This research proposed a new system to identify the medicinal plant leaves use Probabilistic Neural Network classification and to determine the character and specific classes of leaves regarding the best features. The method of identification is basic and derivative feature of leaf, Local Binary Patterns Variance, and Fourier Descriptors. After that, the features are classified by probabilistic neural network classifier which is combined Product Decision Rule (PDR). Then, the result will be analyzed based on overall effectiveness in class and image characteristics. The data of experiments consist of thirty species of flora at Biofarmaka Cikabayan and at Greenhouse Center of Ex-situ Conservation of Medicinal Indonesian Tropical Forest Plants, Faculty of Forestry, Bogor Agriculture University. The results without classifier showed that combination might only have a maximum until 63% accuracy, and if the three classifiers were combined with PDR, the accuracy will increase until 74.67%. Shape is feature with higher probability than morphology and texture feature.