This study focuses on using the SegFormer segmentation model to accurately detect and segment microbial alterations in strawberry diseases, aiming to improve disease detection accuracy under natural conditions. The research evaluates three distinct Mix Transformer encoders—MiT-B0, MiT-B3, and MiT-B5—on a dataset of 2,450 raw images, expanded to 4,574 augmented images. The Segment Anything Model integrated into the Roboflow annotation tool facilitates efficient data annotation and preparation. The results show that MiT-B0 demonstrates balanced but slightly overfitting behavior, MiT-B3 adapts rapidly with consistent training and validation performance, and MiT-B5 offers efficient learning with occasional fluctuations. MiT-B3 and MiT-B5 consistently outperform MiT-B0 across various diseases, with MiT-B5 achieving the most precise segmentation in general. The study highlights the importance of selecting appropriate encoders based on disease characteristics and use cases. MiT-B3 is particularly noted for its rapid adaptation and stable performance, making it the best choice for real-world applications. The findings provide valuable insights for researchers and practitioners, guiding them in choosing the most suitable encoder for disease detection tasks in agriculture.This study focuses on using the SegFormer segmentation model to accurately detect and segment microbial alterations in strawberry diseases, aiming to improve disease detection accuracy under natural conditions. The research evaluates three distinct Mix Transformer encoders—MiT-B0, MiT-B3, and MiT-B5—on a dataset of 2,450 raw images, expanded to 4,574 augmented images. The Segment Anything Model integrated into the Roboflow annotation tool facilitates efficient data annotation and preparation. The results show that MiT-B0 demonstrates balanced but slightly overfitting behavior, MiT-B3 adapts rapidly with consistent training and validation performance, and MiT-B5 offers efficient learning with occasional fluctuations. MiT-B3 and MiT-B5 consistently outperform MiT-B0 across various diseases, with MiT-B5 achieving the most precise segmentation in general. The study highlights the importance of selecting appropriate encoders based on disease characteristics and use cases. MiT-B3 is particularly noted for its rapid adaptation and stable performance, making it the best choice for real-world applications. The findings provide valuable insights for researchers and practitioners, guiding them in choosing the most suitable encoder for disease detection tasks in agriculture.