Pembuatan Perangkat Lunak untuk Menentukan Panjang Gelombang Berdasarkan Spektrum Cahaya Tampak dengan Metode Jaringan Syaraf Tiruan Menggunakan Matlab 7.0

Abstract

Neural network program for determined wavelength in visible spectrum has been made using back propagation method with levenberg Marquardt algorithm. The programming language which is used is MATLAB 7.0. This neural network has a single hidden layer. Before used, program must have been training. Training and testing data are value of primary color RGB with its wavelength. Neural network training process is done by introducing input and target pattern into network. Required epoch along training process is determined by the value of training parameters (number of hidden layer’s neuron, error, mu, mu_dec and, mu_inc). Training process was separated to be feed forward and back propagation phase, all values of weights and biases had been fixed during training process until error or maximum epoch value were met. Neural network program has been successful to identify connection pattern from value of primary color RGB and value of wavelength. This neural network program will produce effective training process if used 30 neuron in hidden layer’s with additional parameter (mu=10 -5 , mu_dec=0.1, and mu_inc=10). Accuracy neural network to identify wavelength’s value for each input colors is determined by the value of error. The level of acuracy in error 10-8 is 100% for data after training (data1) and 97.11% for data never been training (data2).