Tropical Flower Image Classification Based on MobileNetV2 Using an Iterative Development Approach

Abstract

The diversity of tropical flower species and visual morphological similarities between classes pose significant challenges for manual identification, a process prone to inaccuracy. This study aims to implement a Convolutional Neural Network (CNN) algorithm based on the MobileNetV2 architecture to automate tropical flower image classification using an iterative modeling approach. The dataset comprises 7,500 digital images across 15 tropical flower classes. Pre-processing stages include image resizing, normalization, and data augmentation. Modeling was executed using the PyTorch framework, incorporating Early Stopping and a Learning Rate Scheduler. Quantitative evaluation on 1,125 test samples demonstrated that the MobileNetV2 model achieved an accuracy of 84.0%, precision of 84.0%, recall of 84.0%, and an F1-score of 83.0%. These results confirm the effectiveness and efficiency of MobileNetV2 in extracting specific visual features of tropical flora.