An Agro Automation System for Endurable Sunflower Disease Recognition: A CNN-Based Transfer Learning Approach

The Helianthus annuus, commonly known as the sunflower, holds significant ornamental and economic value due to its vibrant appearance and high-quality oil-yielding seeds. Every part of the plant offers utility—leaves serving as fodder, flowers contribute to natural dye production, and the plant enhances aesthetic appeal in various events. Despite its value, sunflower productivity is severely affected by plant diseases, largely due to the lack of timely detection and inadequate knowledge among farmers. This study proposes an automated system for early sunflower disease recognition using transfer learning techniques. A dataset comprising 467 real-life images collected from sunflower fields was categorized into four classes: three disease types—Downy Mildew, Leaf Scars, Gray Mold, and healthy leaves. The dataset quality and variability has enhanced using data augmentation and preprocessing techniques. Deep learning models—DenseNet201, Xception, and ResNet50—were trained and evaluated using six performance metrics. Among these, the Xception model achieved the highest accuracy of 97.33 %, demonstrating its superior performance and potential for practical application in precision agriculture for sunflower disease management.