Abstract:
The identification of flowers is one of the most popular areas of research in the computer vision. This technology proves useful in many areas such as botany, agriculture, environmental science and also teaches in schools. This work is aimed at developing a robust flower recognition system using techniques of deep learning. The images were shot in natural conditions under different lighting, weather, and angles to introduce variability and robustness to the dataset. We have employed transfer learning for state-of-the-art deep learning models such as CNN, InceptionV3, and MobileNetV2. Among these, the best performance was given by MobileNetV2, which showed high accuracy with great computational efficiency. There, the accuracy is 98.33%, and all the models have been trained with great accuracy. Results from the study have showcased the performance of deep learning in the automation of recognizing flowers. The optimum model and FlowerNet collection serve to highlight opportunities not only for improving plant identification, biodiversity conservation, and ecological study but also to form a useful basis for upcoming developments in the field. These results really show how powerful deep learning can be for flower recognition automation. The best model proposed further, with the FlowerNet collection, provides a useful basis for further development in the area by underlining opportunities for improvement regarding plant identification, biodiversity conservation, and ecological study.
