DETERMINATION OF ICUMSA ON GRANULATED SUGAR BASED ON IMAGE PROCESSING AND ARTIFICIAL NEURAL NETWORK

Alfiah Rizky Diana P, Prof. (Emr.) Adhi Susanto, M. Sc. Ph.D.

2014 | Tesis | S2 Teknik Elektro

Klasifikasi dan identifikasi gula kristal di Indonesia sebelumnya dilakukan tanpa standardisasi. Pada produksi gula kristal, terdapat beberapa proses yang masing-masing menghasilkan jenis gula yang berbeda. Proses-proses tersebut memerlukan pengawasan dengan standar tertentu. Standardisasi didesain untuk mengikuti ICUMSA, standar berdasarkan analisis kimia. Sistem didesain untuk mengidentifikasikan nilai ICUMSA pada gula kristal dari citra. Sistem didesain sebagai jaringan syaraf tiruan multi-level perceptron. Jaringan syaraf tiruan didesain dengan satu lapisan tersembunyi dengan 5 neuron dan dilatih menggunakan algoritma Levenberg-Marquardt untuk mengikuti nilai ICUMSA sampel. Fitur warna dan tekstur sebanyak 20 diekstrak dari 180 citra gula kristal untuk masukan jaringan syaraf tiruan. Momen warna, fitur Haralick dan alih ragam gelombang singkat digunakan sebagai fitur. Sistem dapat mengidentifikasi nilai ICUMSA dari 6 sampel gula kristal dengan galat sebesar 1.32%. Setelah menggunakan algoritma reduksi fitur berbasis korelasi (CFS), fitur direduksi menjadi 6 buah. Sistem juga dites untuk gula yang ICUMSA-nya belum diketahui. Kata kunci: ICUMSA, citra, jaringan syaraf tiruan, gula kristal, ekstraksi fitur

Classification and identification of granulated sugar in Indonesia are previously done with no quantitative standardization. In the production of granulated sugar, several stages and condition produce different kinds of sugar, resulting in the need of supervision to obtain a standard. Standardization is designed to follow ICUMSA, a standard based on chemical process. System was designed to identify ICUMSA value of granulated sugar from its image. System was designed as Multi-Level Perceptron Artificial Neural Network with one hidden layer comprised of 5 neurons. Levenberg-Marquardt algorithm was used for the Artificial Neural Network with output trained to follow ICUMSA. Total of 20 colour and textural features were extracted from 180 images of granulated sugar for Artificial Neural Network inputs. Colour moments, Haralick features, and symlet wavelet transform were used as features. Systems successfully identified ICUMSA and classify the 6 samples of granulated sugar with 1.32% of error. After using Correlation-Based Feature Selection (CFS), the amount of features was reduced to 6. System was also tested to identify ICUMSA values of granulated sugar with unknown ICUMSA value. Keywords -- ICUMSA, image, artificial neural network, granulated sugar, feature extraction

Kata Kunci : ICUMSA, citra, jaringan syaraf tiruan, gula kristal, ekstraksi fitur