Clustering Optimization using Particle Swarm Optimization for Medicinal Plant Identification System based on Digital Image

Abstract

Indonesia is a country with a high level of biodiversity and there are 22,500 species of medicinal plants. The percentage of medicinal plant used by Indonesian people is only 4.4% or as much as 1,000 species. The people’s lack of knowledge becomes one of the cause of this low use of medicinal plant in Indonesia. People’s insight about medicinal plant can be increased by developing an identification system of medicinal plant. In this research, we used fuzzy local binary pattern (FLBP) to extract the feature. The number of digital image that used in this research are 1,440 leaf image. Unsupervised learning technique known as clustering is proposed to compare the results of the identification of medicinal plants with classification techniques. Clustering methods used is the k-means clustering and fuzzy c-means clustering. The disadvantages of both method is that the results are sensitive to the selection of initial cluster centers and the calculation of local solutions to achieve optimal conditions. This research applied a particle swarm optimization (PSO) method to overcome the weaknesses of k-means clustering and fuzzy c-means clustering. The application of PSO method in the identification system as a purpose to improve the identification of medicinal plants. This research builds four models of clustering is the k-means clustering, fuzzy c-means clustering, PSO based k-means clustering and PSO-based fuzzy c-means clustering. The development of clustering model using C++ programming language. Evaluation model of clustering using an accuracy value indicates that the PSO method can improve the identification of the k-means clustering and fuzzy c-means clustering. Accuracy of k-means clustering resulted 41.67% and PSO-based k-means clustering is 48%. Accuracy of fuzzy c-means clustering resulted in 50% and PSO based fuzzy c-means clustering is 52.33%. Evaluation of clustering models using computational time indicates the addition of time on the PSO method. The addition of algorithms led to growing number of data processing. Computation time on k-means clustering 291.3 seconds and PSO based k-means clustering 403.3 seconds. Computation time on fuzzy c-means clustering 59.65 seconds and PSO based fuzzy c-means clustering 60.1 seconds. PSO based fuzzy c-means clustering is a better method for generating high accuracy and faster computing time.