One-vs-All Methodology Based Cassava Leaf Disease Detection

Introduction

Cassava is a vital crop for food security in many developing countries. However, it is susceptible to several diseases that can significantly reduce yields. This project proposes a one-vs-all methodology to classify the four most prevalent types of cassava leaf diseases and healthy leaves.

Materials and Methods

• Dataset: The dataset consists of images of cassava leaves, annotated with labels for four disease types and healthy leaves.

• Preprocessing: Image preprocessing techniques, including resizing, normalization, and augmentation (rotation, flips, brightness variation), were applied to enhance model performance.

• Model Training: Five binary classifiers were trained using the EfficientNet B4 model as the base. Each classifier was responsible for distinguishing one class from the rest.

• Deployment: The final model was deployed on Android using Android Studio, Java, and XML, making it accessible to farmers with low-cost mobile devices.

Results and Discussion

The one-vs-all methodology achieved an accuracy of 85.64% on highly skewed test data. This approach allowed each classifier to specialize in detecting a specific disease, improving overall classification performance.

Technologies Used

• Deep Learning Frameworks**: TensorFlow, Keras
• Model: EfficientNet B4
• Mobile Development: Android Studio, Java, XML

Conclusion

This project demonstrates the effectiveness of deep learning in agricultural disease detection. By deploying the model on mobile devices, we provide a valuable tool for farmers to diagnose diseases early and take appropriate action.

Publication

This work was published in the 2021 12th International Conference on Computing Communication and Networking Technologies (ICCCNT).