The invention provides a face turning detection method based on deep learning, and the method comprises the following steps: 1, training a deep learning target detection model, and detecting a bounding box; 2, creating a background image; 3, calculating and drawing a central point of the bounding box; 4, carrying out linear detection on the background image; 5, filtering and extracting straight lines, and judging the range that the straight lines cannot intersect and the slope; 6, calculating the center points of the remaining bounding boxes; step 7, fitting the center point into a straight line; 8, dividing the bounding box to a straight line; step 9, dividing the bounding box into rows, and dividing the straight lines into columns; step 10, acquiring the posture of the head of the student; 11, calculating an included angle between a yaw angle of the student head and a straight line corresponding to the bounding box, and judging whether the student turns the face or not; and 12, drawing the straight lines, the bounding boxes, the rows, the columns and the included angles on the image. According to the invention, row and column information of students and face turning can be automatically positioned, and classroom learning listening conditions can be digitally embodied.