Comparison of Convolutional Neural Network Models for Feasibility of Selling Orchids

Orchid flowers are one of the most popular ornamental plants, widely appreciated for their unique features and aesthetic appeal, making them highly potential for sales in the global market. While numerous studies have explored Orchid flower characteristics and disease detection, research on the classification of Orchid salability remains unexplored. This study addresses this gap by classifying Orchid flowers into three categories: saleable, potential saleable, and not saleable. Convolutional Neural Networks (CNN), known for their effectiveness in image-based classification, were employed in this study with performance enhancement through the application of transfer learning. Two prominent transfer learning architectures, VGG-16 and ResNet-50, were implemented and compared to evaluate their suitability for Orchid salability classification. The results demonstrated that the VGG-16 model significantly outperformed ResNet-50 in all evaluation metrics. The VGG-16 model achieved an accuracy of 98%, precision of 99%, recall of 97%, and an F1 score of 98%. In contrast, the ResNet-50 model yielded lower performance, with an accuracy of 69%, precision of 68%, recall of 56%, and an F1 score of 56%. The study also observed that increasing the training epochs from 25 to 50 had no significant impact on the performance of either model. This research highlights the superior performance of VGG-16 in Orchid salability classification and underscores the potential of transfer learning in advancing ornamental plant research.