A pythonic elegance: Unraveling iris classification through convolutional neural network
|
Full Text |
Pdf
|
|
Author |
Jessica S. Velasco, Clarabelle D. Bismonte, Krizea Mharie C. German, Alvin John A. Proceso, Rex Romero and Ryan C. Reyes
|
|
e-ISSN |
1819-6608 |
|
On Pages
|
1801-1806
|
|
Volume No. |
20
|
|
Issue No. |
20
|
|
Issue Date |
January 20, 2026
|
|
DOI |
https://doi.org/10.59018/1025203
|
|
Keywords |
iris flower, convolutional neural networks, python.
|
Abstract
Horticulturists and hobbyists appreciate the unique characteristics of irises, particularly their distinctive floral structure with three upward-pointing standards and three downward-drooping falls, contributing to their aesthetic appeal. The petal and sepal arrangement serves as a crucial identifier, forming the recognizable iris blossom. Despite a wide range of iris cultivars and hybrids, botanists and horticulturists find it hard to distinguish iris; they use specific traits like bloom size, petal features, and scent to distinguish closely related species. The image classification has undergone transformative advancements, with Convolutional Neural Networks (CNNs) playing a central role in enhancing accuracy and efficiency. This paper reviews the evolution of image classification methodologies, emphasizing the hierarchical feature extraction capabilities of CNNs. It also highlights the promising trajectory of image classification, fueled by innovations in regularization techniques, interpretability methods, and fine-tuning strategies using different models such as DenseNet, Xception, ResNet50, MobileNet, and InceptionV3. The DenseNet achieved an impressive accuracy of 92.40%, demonstrating its effectiveness in the given task. InceptionV3 followed closely with an accuracy of 89.96%and Xception has 83.54%, showcasing its robust performance. MobileNet outperformed the others, boasting an accuracy of 93.23%, suggesting its suitability for the specific application. However, ResNet50 displayed a significantly lower accuracy of 16.46%, indicating potential challenges or limitations for this model in the given context.
Back