Predicting and recognizing plant diseases at an early stage is a critical requirement for expanding agriculture, which contributes significantly to the economy and food security of our country. Early detection can help to save crops and prevent further damage. Deep learning methods are commonly used for image-based disease classification and prediction. This paper studies sunflower diseases such as Alternaria leaf spot and Verticillium wilt, and presents a deep learning segmentation model to classify them. The study enhances the YOLOv5 architecture, the scales of the fusion layer, and the multiscale detection layer to make the first-stage prediction more effective at detecting small, subtle defects with high similarity on the leaf surface. In addition, a modified version of the Improved Dwarf Mongoose Optimization Algorithm (IDMO) is used for hyperparameter tuning. This optimization method makes three minor but significant changes to the original algorithm (DMO). First, IDMO's alpha selection is different from DMO's, where calculating the probability value of each fitness member is an unnecessary computational burden that adds nothing to the alpha's or the group's overall quality. The Kaggle dataset images are pre-processed to improve classification accuracy. The experimental results validate that the proposed model outperforms state-of-the-art deep learning approaches by a margin of approximately 95%. This research has the potential to revolutionize the way plant diseases are detected and classified. By using deep learning, farmers can quickly and accurately identify diseases, which will help them to save crops and prevent further damage. This could lead to increased crop yields and improved food security.