Advancing Explainability in Multi-Label Classification for Tomato Disease Detection Using Machine Learning Interpretability Techniques

<p>Plant diseases pose a significant threat to global food security, affecting crop yield, quality, and overall agricultural productivity. Traditionally, diagnosing plant diseases has relied on timeconsuming visual inspections by experts, which can often lead to errors. With the rapid growth of technology, machine learning (ML) and artificial intelligence (AI) have opened new possibilities for automating this process. One of the most promising technologies for plant disease diagnosis is Convolutional Neural Networks (CNNs), which have proven effective in image classification tasks. Plant leaves, often exhibiting symptoms such as discoloration and irregular textures, serve as key indicators for disease detection. By processing large datasets of leaf images, CNNs can automate disease diagnosis without the need for manual inspection, offering faster and potentially more accurate results. Despite their strong performance, CNNs are often criticized as "black box" models due to the lack of transparency in their decision-making processes. In high-stakes fields like agriculture, where farmers and agronomists must rely on the predictions of these models, this opacity can hinder trust in their outputs. To address this issue, machine learning interpretability techniques, known as Explainable AI (XAI) has emerged as a solution, providing insights into how these models make predictions and which features of the input data most influence their decisions. This research aims to combine CNNs with XAI techniques to enhance transparency in plant disease diagnosis. Specifically, we focus on tomato leaf diseases, where the plant may suffer from multiple diseases simultaneously. Unlike traditional multiclass classification, which labels diseases as individual categories, multilabel classification allows for the identification of iv multiple diseases in a single leaf. This multilabel approach, coupled with XAI, allows agricultural specialists to better understand the model's decision-making process and increases trust in automated disease detection systems. Our findings show that incorporating XAI techniques, such as Grad-CAM, intergraded gradients, and LIME, significantly improves model interpretability, making it easier for practitioners to identify the underlying symptoms of plant diseases. This study not only contributes to the field of plant disease detection but also offers a novel perspective on improving AI transparency in real-world agricultural applications. With a training accuracy of 96.88% for EfficientNetB7, 93.75% for EfficientNetB0, and 87.50% for ResNet50, our proposed models outperform earlier research on the same dataset, demonstrating a notable increase in accuracy over previous models. Additionally, we implemented three explainable AI techniques to enhance the transparency and trustworthiness of our models, ensuring that their decisions are interpretable and reliable.</p>

Name thesis:degree_name

Computer Science - Applied Computing Track

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

TSYS School of Computer Science

Year dc:date.available

2024

Author dc:creator

• Nurullah, Md.

Contributors dc:contributor

• Dr. Rania Hodhod
• Dr. Hyrum Carroll
• Dr. Yi Zhou

• Tomato Disease Detection
• Machine Learning Interpretability Techniques
• Computer Sciences
• Food Science
• Life Sciences

Language dc:language

English

Repository record dc:identifier

https://csuepress.columbusstate.edu/theses_dissertations/544

OAI identifier oai:identifier

oai:csuepress.columbusstate.edu:theses_dissertations-1546