An Intelligent Classification Model for Rubber Seed Clones Based on Shape Features through Imaging Techniques

This paper describes research work in developing an intelligent model for classifying selected rubber tree series clones based on shape features using image processing techniques. Sample of rubber tree seeds are captured using digital camera where the RGB color image are processed involving segmentation algorithm which includes thresholding and morphological technique. Shape features such as area, perimeter and radius are extracted from each image. Two models are being designed. Model 1 is represented by 38 input features while Model 2 is represented by a reduction of input size using Principle Component Analysis (PCA). The inputs for both models are then used to train a multi-layer perceptron Artificial Neural Network (ANN) using Levenberg-Marquardt algorithm. 160 samples are used as training set while another 100 samples are used for testing. The optimized ANN models are then evaluated and validated through analysis of performance indicators regularly applied in classification research work via pattern recognition. Findings in this work have shown that the optimized Model 2 has the best accuracy of 84% with more than 70% achievement for sensitivity and specificity.