A real time intelligent detection and counting method of cells based on YOLOv5

Blood cell count plays an important role in the diagnosis and treatment of many diseases. This paper proposes and implements a blood cell detection and count method based on YOLOv5. On the basis of YOLOv5, 346 blood cell images screened by BCCD dataset were used for 300 epochs of training. The mAP of 46 samples in the verification set was as high as 0.96, and the actual counting accuracy of red blood cells, white blood cells and platelets was close to 100%. The detection accuracy of untrained new data is also close to 100
^%
, and the recall rate is slightly weaker than the verification set data. The inferencing time of single sample in batch detection is only 16 ms, which has the ability of real-time detection and counting. However, in the SSD blood cell detection and counting control model with MobileNet as the backbone, the verification set mAP is only 0.62, and the platelets can hardly be detected and counted. Experiments show that the blood cell detection and counting method based on YOLOv5 has the ability to replace and assist artificial cell counting in clinic.