Researchers have utilized deep learning techniques, particularly object detection technology, to address the labor-intensive task of detecting and counting various blood cells in microscopic smear images. In this study, we focused on detecting and counting the three main types of blood cell: platelets, red blood cells, and white blood cells. To achieve this, a You Only Look Once (YOLO)-based network was employed, using the YOLO algorithm with its CSPDarkNet53 backbone, to automatically detect and count these blood cells. We proposed a systematic approach using the uniform experimental design method to improve the accuracy of blood cell detection and counting by optimizing algorithm hyperparameters for the YOLO-based network. By applying this optimized algorithm hyperparameter combination (OAHC) to the YOLO-based network, we enhanced the detection and counting of blood cells in the blood cell counting and detection (BCCD) dataset, resulting in the YOLO-based–OAHC model. The experimental results showed that the YOLO-based–OAHC model achieved a higher detection accuracy than the Resnet50–single shot detector model, particularly on the test set of blood cell images. The YOLO-based–OAHC model demonstrated superior accuracy in detecting and counting blood cells, particularly white blood cells and platelets, in images from the BCCD dataset.