#TITLE_ALTERNATIVE#

• Main Author: HARLI WIDYA NARA UTAMA (NIM : 13306092), MUHAMAD
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/23166
• Institution: Institut Teknologi Bandung

Nowadays, identification of wood type in Indonesia is conducted by trained evaluators. Wood identification is based on general characteristics and anatomy features. General characteristics include colour, pattern, texture, fiber, odor, and rigidity of the wood. Given so many types of wood that are available in the market, evaluators need a lot of training and experience to identifiy the type of wood which may require a long time. This study develop a system to identify wood type based on pores and concentric curve which are wood anatomical features. Wood images that are used consist of 10 types of wood trading in Indonesia, which are : dahu, durian, jati, kamper, keruing, meranti putih, merbau, mersawa and ramin. The method for feature extraction using discrete wavelet transformation two <br /> <br /> <br /> dimension. In this method, initial image will be decomposed into four images for each level,which are approximation, detail horizontal, detail vertical and detail diagonal images. The result of feature extraction is energy value from each decomposition result. The energy value from feature extraction becomes the input <br /> <br /> <br /> of feature classification using artificial neural network with back propagation multilayer perceptron (MLP) as training method. Training and testing process in this system was done using 20 data from every type of woods for each training data and testing data. The results shows that wavelets have 92,9% in average for true identification. Basis wavelet db3 have the best result for identification wood type with 95% true identification. In the other hand, the worst wavelet to identificate is bior3.3 with 89% true identification.