Abstract:
In horticulture, botanical study, and landscaping, accurate rose identification is
critical. The ability to tell the difference between red, yellow, and white rose varieties
benefits in the creation of visually pleasing gardens, contributes to botanical
databases, and aids biodiversity conservation efforts. In the floral sector, precise
classification is also useful in agricultural techniques, ensuring optimal production,
disease management, and informed decision-making. The emphasis on deep learning
models in this study emphasizes the essential role that technology plays in improving
the efficiency and reliability of rose identification operations. This study explores the
effectiveness of four deep learning models—EfficientNet, ResNet50, MobileNetV2,
and FNet—in classifying roses by color (red, yellow, and white). Advanced computer
vision algorithms help with rose identification, which is important for agriculture and
botanical research. Each model is rigorously trained and evaluated using a varied
collection of high-resolution rose photos. For identifying between rose colors,
performance parameters such as accuracy, precision, recall, and F1 score are
examined. The most promising model is FNet, which achieves an astonishing 98.17%
accuracy, demonstrating the efficacy of transformer-based topologies in rose color
recognition. The study emphasizes the importance of selecting proper deep learning
models and positions FNet as a reliable option for accurate rose color detection. These
findings add important insights to computer vision in botanical research, aiding
academics and practitioners in the selection of best models for rose identification
based on color features. The success of FNet motivates further investigation of
transformer architectures, which could lead to advances in plant species recognition
via deep learning approaches.
