Comparison Between Faster R-CNN and CNN in Recognizing Weaving Patterns

Weaving is a particular type of cloth made specifically with distinctive motifs which is a traditional handicraft of Nusantara archipelago. It is necessary to have a motive recognition technology innovation that can identify weaving motifs in the Nusantara archipelago woven fabrics. Recognition of pattern in the woven fabric is very difficult because the patterns and types of Nusantara weaving are very diverse. This research will design the classification of woven fabric patterns using Faster R-CNN and compare it with The Convolutional Neural Network method. This research aims to compare the performance of Faster R-CNN and CNN in classifying weaving patterns by measuring the accuracy, precision, and recall levels of the recognition of Malay woven motifs with both Faster R-CNN and CNN. After collecting datasets, analyzing system requirements, designing preprocessing, and training processes for both CNN and Faster R-CNN, it is found that from the training data in the form of images of the Malay woven fabric and the gringsing woven cloth, it is found that through the K-Fold Cross Validation with a value of k = 5, the classification using Faster R-CNN obtained 82.14% accuracy, 91.38% precision and 91.36% recall. Meanwhile, the CNN method obtained 76% accuracy, 74.1% precision, and 72.3 recall. The faster R-CNN was superior in all parameter tests compared to CNN with a difference of 6.14% for accuracy, 17.28% more precision, and 19.06% for recall value. It found that the choice of Feature extractor architecture impact detection accuracy.