[Paper Review] Cashew dataset generation using augmentation and RaLSGAN and a transfer learning based tinyML approach towards disease detection

Varsha Jayaprakash, K. R. Akilesh+3 more|arXiv (Cornell University)|Apr 18, 2023

Smart Agriculture and AIAgricultural and Biological Sciences3 citations

TL;DR

This paper proposes a novel, augmented cashew leaf and fruit dataset using photometric/geometric augmentation and RaLSGAN to address data scarcity in disease detection. It employs transfer learning with MobileNet and Inception, optimized via TensorFlow Lite for tinyML deployment on drones, achieving up to 99% accuracy with reduced memory and latency for real-time, resource-constrained applications.

ABSTRACT

Cashew is one of the most extensively consumed nuts in the world, and it is also known as a cash crop. A tree may generate a substantial yield in a few months and has a lifetime of around 70 to 80 years. Yet, in addition to the benefits, there are certain constraints to its cultivation. With the exception of parasites and algae, anthracnose is the most common disease affecting trees. When it comes to cashew, the dense structure of the tree makes it difficult to diagnose the disease with ease compared to short crops. Hence, we present a dataset that exclusively consists of healthy and diseased cashew leaves and fruits. The dataset is authenticated by adding RGB color transformation to highlight diseased regions, photometric and geometric augmentations, and RaLSGAN to enlarge the initial collection of images and boost performance in real-time situations when working with a constrained dataset. Further, transfer learning is used to test the classification efficiency of the dataset using algorithms such as MobileNet and Inception. TensorFlow lite is utilized to develop these algorithms for disease diagnosis utilizing drones in real-time. Several post-training optimization strategies are utilized, and their memory size is compared. They have proven their effectiveness by delivering high accuracy (up to 99%) and a decrease in memory and latency, making them ideal for use in applications with limited resources.

Motivation & Objective

• Address the challenge of limited, imbalanced datasets for cashew disease detection, particularly for dense-canopied trees.
• Develop a high-quality, authentic dataset of healthy and diseased cashew leaves and fruits using RGB color transformation to enhance diseased region visibility.
• Enhance dataset diversity and size using photometric and geometric augmentations combined with RaLSGAN to improve model generalization.
• Enable real-time, on-device disease detection in field conditions by deploying optimized deep learning models on resource-constrained platforms like drones.
• Optimize inference performance through post-training quantization and model compression techniques for minimal memory and latency.

Proposed method

• Apply RGB color transformation to highlight diseased regions in cashew leaf and fruit images for improved visual discrimination.
• Use photometric and geometric data augmentation to increase dataset diversity and robustness against variations in lighting and orientation.
• Generate synthetic images using RaLSGAN (Relativistic Average Least Squares Generative Adversarial Network) to expand the initial dataset and reduce data scarcity.
• Train and evaluate deep learning models using transfer learning with MobileNet and Inception architectures on the augmented dataset.
• Deploy trained models on edge devices using TensorFlow Lite, applying post-training quantization and optimization techniques to reduce model size and inference latency.
• Compare memory footprint and inference latency across multiple optimized model variants to identify the most efficient deployment configuration.

Experimental results

Research questions

• RQ1 Can data augmentation and GAN-based synthesis significantly improve the quality and diversity of a small, real-world cashew disease dataset?
• RQ2 To what extent does transfer learning with MobileNet and Inception improve classification accuracy on the augmented cashew dataset?
• RQ3 How effective are post-training quantization and optimization techniques in reducing model size and latency for edge deployment?
• RQ4 Can the optimized models achieve high accuracy while remaining suitable for real-time inference on low-power devices like drones?
• RQ5 What is the optimal balance between model accuracy, memory footprint, and inference speed for on-device cashew disease detection?

Key findings

• The proposed dataset augmentation and RaLSGAN-based data generation significantly improved model generalization and robustness on real-world cashew disease images.
• Transfer learning with MobileNet and Inception achieved classification accuracy of up to 99% on the augmented dataset, demonstrating high performance.
• Post-training quantization reduced model size and inference latency, making the models suitable for deployment on memory-constrained edge devices.
• The optimized models achieved high accuracy with minimal memory overhead, enabling real-time disease detection on drones.
• The combination of data augmentation, GAN-based synthesis, and model optimization resulted in a deployable, efficient, and accurate tinyML pipeline for agricultural monitoring.
• The final optimized models showed a substantial reduction in memory footprint and latency, confirming their viability for edge-based applications in precision agriculture.