High-Precision Image Segmentation and Feature Extraction Algorithm Design for Blood Cell Detection

Blood cell detection plays a key role in clinical medical diagnosis, while traditional manual microscopic observation methods have obvious deficiencies such as strong subjectivity and low efficiency. Addressing this problem, this research designs a high-precision image segmentation and feature extraction algorithm system by improving the U-Net network architecture through residual modules and dual attention mechanisms, optimizing the training process with multi-component loss functions, and constructing multi-scale feature fusion and attention-based feature selection methods. Experimental results show that the improved algorithm achieves 96.8% pixel-level accuracy and a Dice coefficient of 0.921 on the BCCD test set, while the feature extraction method reaches 97.2% accuracy in white blood cell five-classification tasks. The algorithm significantly improves the automation level and accuracy of blood cell detection, providing reliable technical support for clinical applications.