LEAFLET: A Web-Based Leaf Classification System Using Convolutional Neural Networks

Abstract

The Convolutional Neural Network (CNN), a crucial component of deep learning neural networks, is a network model frequently used in the classification of images, target recognition, and other applications. Utilizing this network model, this study aims to find the best CNN model compatible for the web-based application that can identify endemic trees found locally at FAITH College’s Serenity and Labyrinth Garden with 90% accuracy rate. Similar to this study, Salem et al. (2019) tested different classifiers with both Flavia and their own dataset and found that K-Nearest Neighbors (KNN) displayed the highest precision among all. The study focuses on a web application utilizing the five pre-trained models namely VGG-16, ResNet50, Inception V3, Xception and Sequential, and it was found that the best performing among those is the VGG-16 which was then used for leaf classification, achieving a remarkable 94% accuracy, surpassing the 90% objective. The application allows users to scan leaves of specific trees in a designated area, with accuracy not guaranteed beyond this zone. The application's performance is influenced by the user's device, internet connection, and image quality.