Leveraging Inception-V3 and Transfer Learning for Early Detection and Classification of Cotton Crop Diseases

Abstract

Background: Cotton is an important crop globally and early detection of plant diseases is crucial for maintaining yields. Traditional methods for disease detection are manual and inefficient, highlighting the need for advanced technology like AI to enhance productivity. Methods: The study utilized the Inception-v3 deep learning model along with techniques such as transfer learning and hyper-parameter tuning. These approaches helped design an efficient system to classify whether a cotton plant is healthy or diseased. Comparisons were made with other pre-trained models like VGG16, ResNet50 and ResNet152V2. Result: The Inception-v3 model showed exceptional performance: • Achieved 87.52% accuracy without tuning. • Achieved 98.85% accuracy after hyper-parameter tuning, marking an improvement of ~11%. This approach also demonstrated faster and more precise predictions for diseases like bacterial blight, army worms and aphids. It supports sustainable farming by reducing chemical usage while maintaining crop quality and yield.

Plant phenotyping relevance

綿花の健康・罹病状態を画像ベースの深層学習で分類する手法が研究の中心であり、植物病害状態のフェノタイピング手法に該当する。

Code and data availability

The paper describes an Inception-V3 transfer learning model for cotton disease classification using a Kaggle-sourced dataset, but provides no public code, model checkpoints, or dataset URL of its own. The Kaggle dataset is mentioned only by name without a link, and no availability statement for code or trained models (

No evidence-backed public reproduction asset is currently recorded.