Comparison of Crop Image Classification Model Performance According to Image Augmentation Techniques

Currently, the agricultural population is rapidly decreasing in Korea, and the aging population is getting worse. Therefore, the development of smart farm-related technologies is more necessary, and image processing technologies using big data and AI are very important. In this study, after collecting crop images directly from farms, three datasets were constructed according to the preprocessing method, and images between crops were classified using a CNN-based image classification model. As a result of the analysis, it was confirmed that the method using dataset B and NASNetLarge with five data augmentation was the best with an average classification accuracy of $\mathbf{9 6 . 7 1 \%}$.