基于改进YOLO Nano的嵌入式手背静脉检测

Detection of Dorsal Hand Vein Based on Improved YOLO Nano and Embedded System

静脉穿刺作为最基本的医疗手段,仍然是医疗工作者面临的一大挑战。针对近红外图像提出一种静脉检测和定位方法。首先,设计基于近红外的可穿刺静脉定位嵌入式系统,采集43名实验对象左右手手背静脉图像,经预处理后建立了含325张手背静脉图像的数据库;其次,改进YOLO Nano算法,通过裁剪网络结构减小模型大小,并缩减输出尺度适应检测目标的尺寸,同时引入空间金字塔结构以提升模型的特征表达能力和计算能力,最终达到更高检测准确率。按7∶3的比例将数据库划分为训练集和测试集并进行标注和数据扩充,并在嵌入式设备上进行了测试。结果表明,改进的YOLO Nano模型参数量减小了15%,平均精确度由91.68%提升至93.23%,检测时间缩短为529 ms,较YOLO Nano减少了22%,在检测速度和准确率上均优于原版YOLO Nano,实现了穿刺静脉的快速准确检测与定位。

As the most fundamental medical means,venipuncture remains challenging for medical workers.This paper proposed a vein detection and location method for near infrared image.Firstly,an embedded system based on the near-infrared was designed,by which the vein images of both left and right dorsal hand from 43 subjects were captured to finally build a database composed of 325 dorsal hand vein images after preprocessing.Secondly,YOLO Nano algorithm was improved by trimming the network structure to reduce the model size and the output scale to adapt to the size of the detection target.The spatial pyramid pooling structure was introduced to improve the detection accuracy for its strong detail feature description and efficient feature computation.The database was divided into training set and test set in a proportion of 7∶3 and labeled and expanded.After tested on our embedded system,the results showed that the size of the improved YOLO Nano was reduced by 15%,while the average precision(AP)was increased from 91.68%to 93.23%and the detection time reached 529 ms,reduced by 22%compared to YOLO Nano.The improved YOLO Nano outperformed the original YOLO Nano in terms of both detection speed and accuracy,which realized the detection of puncturable veins.