YOLO-MaskRCNN：一种改进的血细胞分类统计模型

YOLO-MaskRCNN: An Improved Statistical Model for Blood Cell Classification

摘要: 血液中血细胞的数量对疾病诊断具有重要意义。尽管全自动血细胞计数仪已普及，但仍需人工涂片和显微镜形态学检查，工作量较大。为此本文提出一种改进的血细胞识别和计数模型：在早期YOLO模型中引入颜色注意力机制，利用染色后不同血细胞在颜色上的差异，通过小型全卷积网络对特征图进行加权，使模型更加关注颜色特征，从而提升血细胞的检测与分类效果，并保持较低的计算开销。另外，针对血液涂片中粘连和重叠红细胞的计数难题，采用Mask R-CNN算法实现实例分割，相较于传统的分水岭和轮廓检测方法，能更准确地区分粘连红细胞。实验证明改进后的模型训练成本更低且识别准确度更高，最后，开发了一个基于Python的开源桌面程序，实现血细胞识别、计数与分割的一体化，降低了工作者的负担。

Abstract: The quantity of blood cells plays a crucial role in disease diagnosis. Although automated hematology analyzers have become widely used, manual smear preparation and microscopic morphological examination are still required, resulting in heavy workloads. To address this issue, this paper proposes an improved blood cell detection and counting model. A color attention mechanism is introduced into an early YOLO model to exploit the color differences among various blood cells after staining. A lightweight fully convolutional network is employed to weight the feature maps, enabling the model to focus more on color features and thereby enhance detection and classification performance while maintaining low computational cost. In addition, to tackle the challenge of counting adherent and overlapping red blood cells in blood smear images, the Mask R-CNN algorithm is adopted for instance segmentation. Compared with traditional watershed and contour-based methods, this approach can more accurately distinguish adherent red blood cells. Experimental results demonstrate that the improved model achieves lower training cost and higher recognition accuracy. Finally, a Python-based desktop application has been developed to integrate blood cell detection, counting, and segmentation, effectively reducing the workload of laboratory personnel.