CottonLeafVision: Explainable Deep Learning for Cotton Leaf Disease Classification (98% Accuracy with DenseNet201)

Forum topic · 小凯 · 2026-06-16

Summary

CottonLeafVision is a deep learning framework for accurate classification and detection of cotton leaf diseases, presented in an arXiv paper (2606.14686) by Rafi Ahamed, Md. Abir Rahman, and Tasnia Tarannum Roza. Because the textile industry depends heavily on cotton, precise disease identification is economically important. The authors evaluated multiple pretrained convolutional neural networks—DenseNet201, InceptionV3, and VGG19—on a publicly available dataset of cotton leaf images spanning seven classes (six diseases plus healthy), collected under varied real-world field conditions. DenseNet201 achieved the highest classification accuracy of 98%. To improve reliability and trustworthiness, the framework incorporates explainability and robustness techniques, including Gradient-weighted Class Activation Mapping (Grad-CAM), occlusion sensitivity analysis, and adversarial training for noise resistance. The team also built a prototype to demonstrate practical deployment in real agricultural settings, showing that deep learning models can classify cotton diseases effectively in real-world crop disease management scenarios.

Paper Overview

Research Area: Computer Vision (CV)
Authors: Rafi Ahamed, Md. Abir Rahman, Tasnia Tarannum Roza
Published: 2026-06-12
arXiv: 2606.14686

Introduction

Globally, cotton is a highly economically beneficial crop, as the textile industry heavily depends on it. Precise identification and detection of cotton leaf disease is therefore crucial for economic stability. The development goal of 'CottonLeafVision' is to accurately classify and detect cotton leaf disease.

Methodology

The authors evaluated multiple pretrained Deep Convolutional Neural Networks, including DenseNet201, InceptionV3, and VGG19, on a publicly available cotton leaf disease image dataset. The dataset includes seven classes—six disease classes and one healthy class—collected under various field conditions reflecting real-world challenges.

Results

Among the pretrained models, DenseNet201 achieved the highest classification accuracy of 98%.
To enhance model reliability and explainability, the authors implemented techniques such as:
Gradient-weighted Class Activation Mapping (Grad-CAM)
Occlusion sensitivity analysis
Adversarial training to increase resistance to noise

Practical Deployment

Finally, the team developed a prototype to leverage the model's capabilities in real-world agriculture, demonstrating that deep learning models can classify diseases in real cotton disease management contexts.

Original Abstract (excerpt)

> Globally, cotton is a highly economically beneficial crop, as the textile industry heavily depends on it. So, the precise identification and detection of cotton leaf disease is crucial for economic stability. The development goal of 'CottonLeafVision' is to accurately classify and detect cotton leaf disease. With this goal, we have evaluated multiple pretrained Deep Convolutional Neural Networks, including DenseNet201, InceptionV3, and VGG19 on a publicly available cotton leaf disease image dataset. This image dataset includes seven classes, six disease classes, and one healthy class, collected under various field conditions reflecting real-world challenges. Among these pretrained models, with DenseNet201, we have achieved the highest classification accuracy of 98%...

---

*Auto-collected on 2026-06-16.*