A Deep Learning Approach to Potato Late Blight Detection: Developing and Evaluating a Lightweight Single-Stage Detection Model

Detecting and identifying crop diseases in their early stages remains a challenge in modern agricultural practices, affecting yield and food security worldwide. Potato late blight, caused by Phytophthora infestans is a serious threat to potato crops requiring fast and accurate detection methods to minimize losses. A new approach using deep learning was suggested in this study to spot this disease at an early stage. The study focused on developing and training a YOLOv8n model with a dataset of 14,150 images and annotations from Berkane, Morocco. This dataset was carefully chosen to cover a diverse range of symptoms and stages of potato blight under different environmental conditions to ensure the model’s reliability, robustness, and generalizability. The model is known for its suitable design for real-time decisions in agricultural settings due to its lightweight size. However, this experiment shows that YOLOv8n has a modest detection performance on complex and diverse datasets, making it an interesting opportunity for enhancing research towards early intervention against potato late blight using lightweight models - the potential of deep learning to improve farming methods.