An End-to-End CNN-Based Framework for Butterfly Species Recognition and Visualization via Web Interface

Abstract

Butterflies are vital indicators of environmental 
health, and accurate species identification is essential 
for ecological research and conservation. This paper 
presents a deep learning-based butterfly classification 
system integrated with a web interface for real-time 
image-based prediction. The system utilizes a 
Convolutional Neural Network (CNN) to classify 75 
butterfly species using labeled image data. 
To improve generalization, the model employs data 
augmentation techniques such as rotation, zooming, 
and flipping. The CNN architecture consists of 
multiple convolutional and pooling layers followed 
by dense layers, enabling effective feature extraction 
and classification. The model is trained over 40 
epochs using the Adam optimizer and categorical 
cross-entropy loss, achieving high training and 
validation accuracy. 
A user-friendly Django web application is developed 
to allow users to register, log in, upload butterfly 
images, and receive real-time predictions. The system 
displays the predicted species along with the 
uploaded image, providing an accessible platform for 
students, researchers, and enthusiasts. 
By combining deep learning with a web-based 
interface, this work offers a practical and interactive 
solution 
for 
automated 
butterfly 
species 
identification, demonstrating the effectiveness of 
CNNs in ecological image classification tasks.