基于改进YOLOv5s6的母胎外周血细胞检测

Maternal Fetal Peripheral Blood Cell Detection Based on Improved YOLOv5s6

摘要: 母胎外周血细胞的检测和计数为医务人员的产前诊断提供了高效、可靠的技术支持。本文提出了一种改进YOLOv5s6的目标检测模型REPYOLO-TS2，用于母胎外周血细胞的检测。首先，我们通过对HSV颜色空间中的色调(H)，饱和度(S)，亮度(V)三个通道添加扰动来丰富数据集的环境背景，以实现图像增强并通过马赛克数据增强方法以提高模型的泛化能力。其次，我们在主干网的尾部用C3TR模块代替原模型的C3模块并将主干网的前两个C3模块替换成Repvggblock模块，然后在模型的颈部添加了一种高效的空间转移注意力机制(S2Attention)以提高模型对不同环境背景细胞的检测精度。最后，我们将α-CIoU替换原模型的坐标损失函数CIoU，使检测器能够更快地学习高IoU目标。经过参数调试和实验验证，我们最终确定α值为0.15。最终的实验数据显示，REPYOLO-TS2模型对母体和胎儿外周血细胞的mAP值@0.5为94.2%。

Abstract: The detection and counting of maternal and fetal peripheral blood cells provide efficient and reliable technical support for prenatal diagnosis of medical personnel. In this paper, an improved YOLOv5s6 target detection model, REPYOLO-TS2, was proposed for the detection of maternal fetal peripheral blood cells. First, we added disturbance to hue (H), saturation (S) and brightness (V) in the HSV color space to enrich the environmental background of the dataset, so as to achieve image enhancement and improve the generalization ability of the model through Mosaic data enhancement method. Secondly, we replaced the C3 module of the original model with C3TR module at the tail of the backbone network and replaced the first two C3 modules of the backbone network with Repvggblock module. Then, an efficient S2Attention mechanism was added to the neck of the model to improve the detection accuracy of cells with different environmental backgrounds. Finally, we replace the coordinate loss function CIoU with alpha-CIoU, so that the detector can learn the high IoU target faster. After parameter debugging and experimental verification, the α value is finally determined to be 0.15. Final experimental data showed that the mAP@0.5 value of maternal and fetal peripheral blood cells induced by REPYOLO-TS2 model was 94.2%.