The invention relates to a gymnasium body-building movement identification method and device based on depth learning, takes real-time videos as media and employs a method based on depth learning to carry out movement normalization identification. The method comprises steps that (1),
data acquisition, normative movements are recorded as normative movement images; (2),
data annotation, normative movement classification of the normative movement images is carried out; (3), data training, an
object detection identification framework based on the
convolutional neural network of the depth learning method Caffe is employed to acquire a normative movement identification model; (4), movement identification, user shooting is carried out, user movements in line with normative movement classificationof the normative movement identification model are identified; and (5), movement scoring, normative similarity scores and correction plans are outputted. The method is advantaged in that a user has noneed to wear or replace expensive equipment, cost is controllable, videos which are popular in intelligent terminals at present and are convenient to use are taken as media, precise movement data isgenerated for the user, false movements are corrected, and the better body-building effect is realized.