Oral mucosal cell sampling is primarily utilized to gather samples of sloughed oral epithelial cells or cells in the throat for influenza DNA testing. In the robot sampling process, the image segmentation of the oral and pharyngeal swab sampling area plays a crucial role. However, owing to the complex nature of the oral sampling area, the accuracy of image segmentation can be slightly affected. A pharyngeal swab sampling image segmentation method based on an enhanced Deeplabv3+ model in the field of machine learning is proposed in this paper. The method applied hollow convolution to capture comprehensive information of each convolution output. We utilized deep learning of machine learning methods to enhance and refine the segmentation accuracy of the sampling area and gathered 1774 oral images from 81 volunteers, including children, youth, and middle-aged and elderly individuals, for training, validation, and testing. By comparing the experimental results of U-Net, Mobilenetv2, and Xception, it has been proven that the improved Xception network model has good segmentation performance, with accuracy, recall, and precision of 92.12, 92.86, and 97.69%, respectively. The experimental results indicate that this method accurately and efficiently segments the M-region of the pharyngeal swab sampling area, overcomes boundary discontinuity or ambiguity issues common in other segmentation methods, and possesses a high segmentation accuracy.