基于卷积神经网络的扁桃体咽拭子采样机器人

针对医务人员咽拭子采样时有较高被传染风险，提出咽拭子采样机器人。采样需检测、识别和定位扁桃体的位置信息，由于扁桃体在原图中尺寸较小、特征提取困难，设计基于卷积神经网络的两步检测模型。第一步，使用口腔检测模块将口腔图像从自制的人体面部数据集中检测并切割出来；第二步，在口腔图像基础上，识别位扁桃体的位置信息。结合深度图像计算扁桃体在机器人世界坐标系的坐标，控制机械臂运动到指定位置，实现咽拭子采样。实验证明，系统使用基于卷积神经网络的扁桃体两步检测模型具有较高的准确性和检测效率，检测结果的AP50和检测平均时间均优于对比算法且能够准确地完成咽拭子采样。

Sampling Robot for Tonsil Swabs Based on Convolutional Neural Network

In view of the high risk of infection in throat swab sampling of medical staff, this paper proposes a throat swab sampling robot. Sampling needs to detect, identify and locate the position information of tonsils. Because the size of tonsils in the original image is small and feature extraction is difficult, a two-step detection model based on convolutional neural network is designed in this paper. The first step is to use the oral cavity detection module to centrally detect and cut out oral cavity images from self-made human face data. Step 2, based on the oral image, the position information of tonsils are identified. Then, combined with the depth image, the coordinates of tonsils in the robot world coordinate system are calculated, and the robot arm is controlled to move to the designated position, thus realizing throat swab sampling. Experiments show that the tonsil two-step detection model based on convolutional neural network has high accuracy and detection efficiency, and the AP50 of detection results and the average detection time are better than the contrast algorithm, and the throat swab sampling can be completed accurately.