This paper proposes a Tri-Training framework based on the Weighted Fuzzy Rough Sets and integrates the Granular-Ball method to enhance semi-supervised learning through multi-view granulation. Traditional Tri-Training methods, while effective in leveraging unlabeled data, often suffer from label propagation inaccuracies. Moreover, despite their strengths in handling high-dimensional and uncertain data, Weighted Fuzzy Rough Sets face challenges in robustness and feature selection. To address these limitations, we introduce Granular-Ball computing to generate multi-level granular representations, promoting classifier diversity and reducing redundancy. Additionally, we employ the Weighted Fuzzy Lower Approximation classifier to improve robustness in uncertain environments. Experimental results on multiple benchmark datasets demonstrate that our approach consistently outperforms traditional semi-supervised learning methods, particularly in handling noisy data.