The detection and segmentation of human blood cells can assist doctors to quickly make simple judgments on the current health of the human body, which is of great significance for disease diagnosis. In segmentation tasks of blood cells, the traditional image segmentation algorithm may wrongly segment the target and is unable to completely segment the target. To address these problems, this study proposes a blood cell segmentation algorithm XCA-Unet++ fusing Xception feature extraction and the coordinate attention mechanism. On the basis of the Unet++ network structure, the algorithm introduces the Xception feature extraction network in the encoder part to better extract low-level feature information. Moreover, a cell detection module based on the coordinate attention mechanism is designed to enhance the network’s feature extraction ability for blood cells with blurred edges and incomplete cells. DiceLoss is used as the loss function to optimize the imbalance of positive and negative samples in the dataset and speed up network convergence. The experimental comparison on the public blood cell dataset indicates that the XCA-Unet++ network achieves the results of 94.44%, 96.78%, and 97.12% for the evaluation indicators IoU, Acc, and F1, respectively, and the segmentation performance is better than that of other segmentation networks. Thus, it meets the high-precision requirements of blood cell segmentation tasks.