IMPLEMENTASI JARINGAN SYARAF TIRUAN (JST) DAN PENGOLAHAN CITRA UNTUK KlASIFIKASI KEMATANGAN TBS KELAPA SAWIT

Abstract

The clasification of ripeness stages of oil palm fresh fruit bunches (FFBs) can be done using color parameters. These parameters are often evaluated by human vision, whose degree of accuracy is subjective which can cause doubt in judgement. Automatic clasifications offreshfruit bunches (FFBs) based on color parameters can be done using computer vision. This method is known as a nondestructive, fast and cost effective method. In this research, a MATLAB computer program has been developed which consists of RGB and HSV GUI which is used to record, display, and process FFB image data. The backpropagation artificial neural network (ANN) program is also developed which is used to classify the oil palm fruit fresh bunches (FFBs). Samples are fresh fruit bunches (FFB) of oil palm varieties of Tenera which comprise of Topaz, Marihat, and Lonsum clones. Each clone composed of three levels of ripeness represented by five fractions. The measurements were started by capturing images of oil palm, extracting RGB and HSV values, calculating weight values from the image database to make anANN program, preparing grid programs for oil palm FFBs, and comparing grading levels of oil palm FFBs using program and by harvester. This program successfully classified oil palm (FFBs) into three categories of ripeness which are unripe (F0 and F1), ripe (F1 and F1) and over ripe (F4 and F5). The RGB and HSV programs successfully classified 79 out of 216 FFBs or 36.57% and 106 out of 216 TBS or 49.07%. Respectively the HSV program is better than RGB program because the representation of HSV color space are more understood by human perception hence can be used in calibration and color comparison.

Comparison of Feature Histograms and Co-occurrence Matrix on Analysis of the Light Spectrum Effect for Identification of Surface Roughness with Speckle Images

“…The mean value of each feature in the coexistence matrix is the mean value of 60 coarse sandpaper speckle images and 60 fine sandpaper speckle images obtained from 405 nm diode laser illumination for each Histogram Feature. Each feature of the Co-occurrence Matrix is obtained from the calculation results using Equation ( 6) to Equation (11). Data for the testing process consists of 12 images of coarse sandpaper speckles and 12 images of fine sandpaper speckles resulting from red (wavelength 650 nm), green (wavelength 550 nm), and blue (wavelength 405 nm) laser diodes.…”

“…Using a laser diode with a 3-5 times longer wavelength will affect the contrast of the speckle image [8]. Minaruni's work shows that the average intensity of a pixel is affected by the wavelength of the light used [11].…”

Peramalan Jumlah Permintaan Produksi Menggunakan Jaringan Saraf Tiruan Algoritma Backpropagation

“…Selain itu juga penelitian tentang prediksi dan klasifikasi buku menggunakan metode backpropagation [12]. Penelitian lainnya yang melibatkan Algoritma Backpropagation adalah dalam menentukan klasifikasi 5 varietas padi melalui citra scan laser tiga dimensi (3D) [13], penentuan tingkat kematangan sawit pada 116 citra tandan buah segar [14]. Metode ini juga berhasil mengenali aksara Lontara Bugis pada foto aksara yang diambil pada manuskrip kuno Bugis [15].…”

Akurasi dalam Mengidentifikasi Citra Anggrek Menggunakan Backpropagation Artificial Neural Network

“…Penelitian lainnya yang melibatkan BPNN adalah dalam menentukan klasifikasi 5 varietas padi melalui citra scan laser 3D [12], penentuan tingkat kematangan sawit pada 116 citra tandan buah segar [6]. Metode ini juga berhasil mengenali aksara Lontara Bugis pada foto aksara yang diambil pada manuskrip kuno Bugis [13].…”