Comparison of MobilenetV2 and NASNetMobile for Lavender Flower Analysis using Convolutional Neural Network

Abstrak

The identification of lavender flower varieties is a critical challenge in botany and agriculture, primarily due to the high morphological similarity among different varieties and the influence of environmental conditions on their appearance. Traditional methods of identifying lavender varieties, which often rely on manual observation, face significant limitations. These methods are time-consuming, prone to subjective error, and may not account for subtle environmental variations that affect flower morphology. The specific goal of this research is to develop an automated classification model using Deep Learning techniques, specifically Convolutional Neural Networks (CNN), to improve the accuracy and efficiency of lavender variety identification. The study leverages a dataset from Kaggle, which contains images of three lavender varieties—Lavandula angustifolia, Lavandula viridis, and Lavandula multifida. By applying data augmentation techniques to address dataset variability, the research compares two advanced CNN architectures, MobileNetV2 and NASNetMobile, for their classification performance. The key contribution of this work is demonstrating that NASNetMobile achieves superior performance, with 91.87% accuracy and a lower loss value, compared to MobileNetV2, which reaches 81.67% accuracy. This study highlights the novelty of using CNN models for lavender classification, offering a significant advancement over traditional methods by enhancing the identification process's accuracy and reducing reliance on manual and inefficient approaches. The findings have broad implications for botanical research, agricultural practices, and plant conservation efforts, showing that CNNs can significantly improve the efficiency of plant species identification.