<![CDATA[The more the development of the carving motifs, the more varied the shapes and variations. It complicates the determination of a carving with Jepara motif. In this paper, the transfer learning method with developed FC was used to identify Jepara's distinctive motifs in a carving. The dataset was divided into three color spaces, i.e., LUV, RGB, and YcrCb. Besides, sliding windows, non-max suppression, and heat maps were utilized for the process of tracing the area of the engraved object and identifying Jepara motifs. The test results of all weights showed that the Xception on the Jepara motif classification had the highest accuracy values, namely 0.95, 0.95, and 0.94 for each LUV, RGB, and YCrCb color space dataset. However, when all the model weights were applied to the Jepara motif identification system, ResNet50 was able to outperform all networks with motif identification percentage values of 84%, 79%, and 80%, for the LUV, RGB, and YCrCb color spaces, respectively. These results prove that the system is able to assist in the process of determining whether a carving is included in the Jepara carving or not, by identifying the typical Jepara motifs contained in the carving.]]>
Copyrights © 2020
