AI-Powered Detection of Potato Leaf Diseases Using Deep Neural Networks

The potato (Solanum tuberosum), a crop essential to global food security, faces significant yield threats from diseases like Early and Late Blight. Traditional diagnosis via visual inspection is often inefficient and subjective, delaying necessary action. To overcome these limitations, this study evaluates six deep learning architectures for the automated identification of potato leaf diseases from images. A custom-built Convolutional Neural Network (CNN) is a benchmark against five prominent pre-trained models using transfer learning and subsequently investigates the impact of fine-tuning the top-performing architecture, EfficientNetB0. Experiments on the public PlantVillage dataset show that the fine-tuned EfficientNetB0 delivers a superior accuracy of 99.22%. A real-world test case provides strong support for this quantitative result, demonstrating the fine-tuned model’s ability to correct a misclassification made by its non-fine-tuned counterpart. The proposed work provides a clear benchmark for model selection and confirms that a strategic fine-tuning process is a critical step for achieving robust, state-of-the-art performance in agricultural computer vision and this research also contributes to SDG 9 by developing an innovative and robust deep learning framework to enhance the technological capabilities of the agricultural industry.