Automated Flower Species Detection and Classification Using Image Preprocessing and Deep Learning

Accurate identification of flower species plays a vital role in the fields of agriculture, botany, biodiversity monitoring, supporting smart agriculture, plant breeding, and environmental studies. The use of deep learning and machine learning techniques has greatly improved this task, allowing for highly accurate automated flower detection, particularly in variable, real-world settings. For this study, we collected and improved a data set of 7,864 flower images, including Rosa hybrida, Tagetes erecta, Rosa floribunda, Tagetes patula, Zinnia angustifolia and Zinnia elegans, from local nurseries and gardens. To enhance the performance and to generalize the model, we made use of data augmentation techniques and tested several models like MobileNetV2, DenseNet201, VGG19, ResNet50, Xception and a CNN model. From these models, MobileNetV2 provided the best accuracy of 99.81 % while the customized CNN provided the accuracy of $\mathbf{9 8. 3 5 \%}$. The results of the experiments carried out on MobileNetV2 and CNN models clearly suggest deployment in real-time IoT-based flower recognition systems in the areas of agriculture and the environment.