Foundation Model Enhanced Hierarchical Learning for White Blood Cell Classification

White blood cell (WBC) classification is essential for diagnosing hematological disorders, but automated methods face challenges from data heterogeneity and class imbalance. In this work, we propose a hierarchical deep learning framework for 13-class WBC classification. To address domain shift from various imaging conditions, we fine-tune foundation models and apply pseudo-labeling on unlabeled test samples. For severe class imbalance, we introduce a specialized pipeline combining binary voting and exemplar matching to improve rare subtype recognition. Our method achieves a macro F1 score of 0.77682 on the WBCBench 2026 dataset, demonstrating its effectiveness on heterogeneous data and clinically significant minority classes.