<![CDATA[PT XYZ requires a system to automatically validate items in storage with the system. Cardboard boxes containing items exhibit high visual similarity within a specific sub-category. Several studies have demonstrated the use of Convolutional Neural Network (CNN) as a method for image classification with a Fine Grained Image Classification (FGIC) approach for classifying data with high similarity, resulting in good accuracy, and this will be applied in this research. The ResNet architecture is used with and without ImageNet weight initialization, combined with the RMCSAM architecture, resulting in eight training configurations. Based on testing results using 172 images across 14 classes, the ResNet + RMCSAM configuration with ImageNet weight initialization and the 20 times augmentation dataset achieves the highest accuracy compared to other configurations, with an accuracy of 99.42% and a loss of 0.0004. This configuration is utilized for cardboard classification in the PT XYZ warehouse.]]>
Copyrights © 2024
