Automated Flower Classification Using ResNet50: A Deep Learning Approach

This research revolves around the application of pretrained ResNet50 for the classification of flower species into five main categories i.e., Rose, Tulip, Sunflower, Daisy, and Dandelion. The study's results in ResNet50 have been achieved to be of great accuracy and robust performance on all classes. With its training and validation over 10 epochs with a batch size of 62, it produced notable results such that the precision, recall, and F1-scores of most categories are greater than 0.87. The confusion matrix indicates good classification accuracy with slight misclassifications between closely related flower types. The best classified classes include Dandelion and Sunflower, with F1-scores of 0.93, while the class that faced the worst case is Rose, which had presented an F1-score of 0.88. Data augmentation and regularization techniques have also been applied to avoid overfitting, because the validation metrics are proof that can be given in this regard. In general, the experiment showed that ResNet50 was efficient in the classification of flowers and may be further improved for precision by limiting more misclassifications between similar species. Consequently, the notion that ResNet50 can be a great helping tool in the automation of flower classification finds its way into botanical research and autonomous agricultural systems.