Reliable of traditional cloth pattern Classification Using Convolutional Neural Network

Basically the traditional cloth pattern varieties increase every year so that they become more difficult to identify. Based on that fact, automatic traditional cloth patterns has become more important to help people recognize their patterns. This research can improve reliability in recognizing and understanding traditional cloth patterns originating from several regions in Indonesia by using a Deep Convolution neural network. In this study, several conditions were carried out, namely, the amount of training image data was more than the number of test images (3 conditions) and vice versa where the number of training images was much less but could recognize more test images (2 conditions). To support classification reliability, data processing is carried out starting from preprocessing with grayscale, resizing, noise reduction, and continued with image segmentation, and feature extraction and classification using the CNN Algorithm. After We do experiment with 42 classes of traditional cloth patterns from 22 provinces on collecting images that can be found in various scales and degrees than the average classification accuracy using Inception V3 3.0 is 83.9% with condition 1, but the highest is between 98.1 %–99.7%. Meanwhile, if using resnetV2 50 the average classification accuracy is between 78.3% for the highest between 93.7%–94.3%. This Research can be continued using antoher Deep CNN algorithm to get the optimal of classification accuracy.