Color Based Feature Extraction and Backpropagation Neural Network in Tamarind Turmeric Herb Recognition

Abstract

The aim of this paper is finding the optimum image pattern of the tamarind turmeric herb. So far, in the process of producing tamarind turmeric herb, it is not constant in terms of taste and color, which is influenced by maturity and the amount of turmeric. Image pattern recognition will use Backpropagation algorithm applied to typical Content-based image retrieval systems. The main purpose is to apprehend various parts of tamarind turmeric herb in the retrieving processing. The camera is applied to classify the tamarind turmeric herb product, process into 5x5 pixels, and take an average of the RGB value so the stable RGB values will be obtained in each category and used as input for Backpropagation algorithm. The most suitable and the fastest process from the Backpropagation algorithm will be searched and applied in a real-time machine. In this paper will be using two methods, first, train the algorithm using ten data by change neuron, layer, momentum, and learning rate, and the last is testing with ten data. The results obtained from the training and testing algorithm that the two hidden layers can recognize 100% inputs, with three input layers used for R, G, and B value, ten neurons in the first hidden layers and the second hidden layers, one output layer with a parameter used is Learning rate 0.5 and Momentum 0.6. The best image pattern standard for tamarind turmeric herb is dark yellow with RGB values of 255, 102, 32 up to 255, 128, 48.