Deep Learning-Based Transfer Learning for Classification of Cassava Disease

TL;DR

The paper addresses cassava leaf disease classification under dataset imbalance by comparing four pretrained CNN architectures (EfficientNet-B3, InceptionV3, ResNet50, VGG16) with transfer learning from ImageNet. Using a 21,367-image, five-class Ugandan cassava leaf dataset, the study applies stratified splits and data augmentation, training with up to 50 epochs and early stopping. EfficientNet-B3 emerges as the top performer with an F1-Score of 0.877 and high accuracy, while VGG16 underperforms due to excessive parameters and imbalance. The findings demonstrate the viability of efficient pretrained models for digital agriculture and suggest avenues for further improvement via hyperparameter tuning and dataset balancing to boost minority-class performance.

Abstract

This paper presents a performance comparison among four Convolutional Neural Network architectures (EfficientNet-B3, InceptionV3, ResNet50, and VGG16) for classifying cassava disease images. The images were sourced from an imbalanced dataset from a competition. Appropriate metrics were employed to address class imbalance. The results indicate that EfficientNet-B3 achieved on this task accuracy of 87.7%, precision of 87.8%, revocation of 87.8% and F1-Score of 87.7%. These findings suggest that EfficientNet-B3 could be a valuable tool to support Digital Agriculture.

Figures (3)

• Figure 1: Exemplos aleatórios selecionados para cada classe.
• Figure 2: Distribuição de amostras na base de dados
• Figure 3: Matriz de Confusão da EfficientNet-B3