Multispectral Blood Cell Image Analysis via Deep Learning With YOLOv5

Blood cell counting is vital for medical diagnosis, and image recognition offers an automated approach. While most studies rely on microscopic images, these provide limited information. In contrast, multispectral imaging captures additional optical characteristics, improving the delineation of cellular boundaries and structures. This paper presents a blood cell recognition method based on multispectral imaging and YOLOv5. Blood cell images at five wavelengths were fused for multispectral information. The standard and modified YOLOv5 models were trained and tested on single-wavelength and multispectral images. Experimental results demonstrate that, compared with single-wavelength imaging, multispectral imaging markedly enhances the recognition performance of blood cells, yielding identification precision values of 99.9% for red blood cells and 96.1% for platelets. For white blood cells, which are relatively scarce, the recognition precision reached 98.9%, representing a 12.26% improvement over the best-performing single-wavelength model. Multispectral imaging shows significant potential for high-precision detection of rare cells.