The invention discloses a precise facial paralysis degree evaluation method and device based on semantic segmentation. The method comprises the following steps: establishing a facial paralysis semantic segmentation model; acquiring to-be-detected data and processing the to-be-detected data; sequentially inputting the expression-free natural state static image, the sequence image I, the sequence image II, the sequence image III and the sequence image IV into the facial paralysis semantic segmentation model to output a plurality of corresponding groups of face shapes, and updating the pluralityof groups of face shapes; evaluating the facial paralysis degree of the user to be detected; calculating theta1, theta2, theta3, theta4, theta5, theta6, theta7, theta8, theta9, theta10, b1, b2, c1, c2, e1 and e2, and comparing the calculated values with the threshold values; and judging the facial paralysis degree of the to-be-detected user through the comparison result, and calculating a facial paralysis index. According to the invention, the detection model has high detection positioning precision, the precision and accuracy of comprehensive evaluation and detection of the facial paralysis degree of the to-be-detected user are greatly improved, and a powerful support is provided for prevention, discovery and treatment of facial paralysis patients.