STEM-Based Analysis of Avocados (Persea americana Mill.) Leaf Classification using Deep Convolutional Neural Networks

Abstract

Avocado (Persea americana Mill.) is a high-value horticultural crop with great potential to support sustainable food security. However, its productivity is often limited by variety variability and the lack of efficient seed selection methods. This study proposes a machine learning-based framework for avocado variety classification through leaf morphology, including shape, size, texture, and colour. In this study, we integrated and evaluated Deep Convolutional Neural Networks technology to identify avocado types based on leaf image identification. In this study, we used a deep learning architecture and compared it with known approaches, EfficientNetV2, ConvNeXt, and Vision Transformer (ViT)-Hybrid CNN. A total of 1,400 image datasets were divided into training and testing data, containing 980 images and 420 images, respectively. Hyperparameters were considered, where the use of 100 epochs and a learning rate of 0.0001 provided the highest accuracy. The results show that the developed convolutional neural network (CNN) produced the highest accuracy of 97.83% on the EfficientNetV2 architecture, ConvNeXt produced an accuracy of 96.28% and Vision Transformer (ViT)-Hybrid CNN produced an accuracy of 95.14 %˙ The DCNN algorithm used produced the highest accuracy on the EfficientNetV2 architecture with an accuracy value of 97.83 % supported by stability with a FLOPS value of 1.34 GFLOPs.