<![CDATA[Automatic batik motif classification remains challenging due to the high visual similarity among many motifs, making manual differentiation difficult. This study proposes a weighted-sum multi-objective Genetic Algorithm (GA) to optimize a pre-trained ResNet50 model for batik motif classification. The novelty of this study lies in optimizing the top-layer architecture and transfer-learning hyperparameters of ResNet50 by jointly considering classification accuracy, trainable parameter count, and inference latency. Unlike many previous GA-based CNN studies that mainly focus on accuracy or construct architectures from scratch, this study employs GA as a resource-efficient optimizer for practical deployment. The research used a quantitative experimental design with a secondary dataset of 3,550 batik images from five motif classes, namely Kawung, Megamendung, Parang, Sidomukti, and Truntum. The optimization process searched for the best configuration of four hyperparameters, namely the number of neurons in the dense layer, dropout rate, learning rate, and fine-tuning depth, while Adam was used as a fixed optimizer throughout the experiments. The results show that the proposed model improved classification accuracy from 84.00% to 90.00%, reduced trainable parameters by 94.4% from 23.59 million to 1.31 million, and decreased inference time by 29.4% compared with the baseline ResNet50 model. These findings indicate that the proposed method can achieve a favorable balance between predictive performance and computational efficiency for cultural heritage recognition on resource-constrained devices.]]>
