earticle

영어

This study aimed to investigate the performance and factors affecting species classification of CNN architecture using whole-part and earlywood-part dataset of cross-sections in six Korean Quercus species. The accuracy of species classification for each condition using the datasets, data augmentation, and optimizers (SGD, Adam, and RMSProp) based on a CNN architecture with 3–4 convolutional layers was analyzed. The model trained with an augmented dataset yielded significantly superior results in classification accuracy compared to the model learned with a non-augmented dataset. The augmented dataset was the only factor affecting classification accuracy in the final five epochs, whereas four factors in the whole epochs, such as the Adam and SGD optimizer, and the earlywood-part and the whole-part dataset, affected species classification. The arrangement of earlywood vessels, broad ray, and axial parenchyma was identified as major influential factors for CNN species classification through Grad-CAM analysis. The augmented whole-part dataset with the Adam optimizer condition achieved the highest classification accuracy of 85.7% in the final five epochs of the test phase.