The invention discloses a human body key point detection method based on deep learning. The method comprises the steps of data acquisition, network construction, model training and evaluation, optimal model prediction and the like. According to the method, the ResNet50 network is improved, an expanded convolution residual network is provided, and a two-stage expanded convolution residual network is adopted to construct a human body key point detection network. During model training, feature extraction is performed on training data by the first-stage network, prediction is performed by using four channels, loss of all key points in a prediction result are calculated, and the loss is returned to adjust network parameters; the input feature map, the output feature map and the prediction result of the first-stage network are added by adopting an intermediate stage, and are transmitted to a second stage; and feature extraction is performed by the second-level network, prediction is performed on the finally obtained feature map after two-layer transposition, key point loss of a prediction result is calculated, the key point loss is sorted according to a descending order, and the first K * B losses are selected to return and adjust network parameters. An optimal training model is selected to predict human body key points of the to-be-detected image, the precision is high, and the practicability is good.