Complete Blood Cell Detection Based on Improved YOLOV8

Whole blood cell (CBC) detection and recognition plays an important role in general medical examination. However, manually locating and identifying different types of blood cells is time-consuming and laborious, and the accuracy of classification depends on the ability and experience of the operator. In this paper, we propose an architecture based on depth neural network, which can accurately detect and identify blood cells in blood smear images. We propose an improved algorithm based on YOLOV8. In order to solve the problem that it is difficult to identify correctly due to the overlap of different cell types, we use ODConv instead of ordinary convolution. ODConv uses parallel strategy in any convolution and internal to learn the attention value of convolution kernel from four dimensions, so as to obtain full-dimensional convolution kernel attention value, which is helpful to distinguish different types of blood cells. In addition, we introduced BiFPN to learn more detailed features. Our method obtains the mAP50 of WBC 0.972 and WBC 0.893 on the BCCD dataset respectively, which is better than the existing methods.