Third-party face shooting recognition method based on global multilayer connection

The invention discloses a third-party face shooting recognition method based on global multilayer connection. According to the method, the shallow layer and the deep layer of the neural network are communicated in a global multi-layer connection mode, shallow layer information and deep layer information are fused, the shallow layer information comprises visual and simple information such as figure contours and angles, and the deep layer information comprises complex information such as figure expressions and actions; therefore, a better feature extraction effect can be obtained through global multi-layer connection; meanwhile, compared with the high-dimensional channel number of a traditional convolutional neural network feature map, the feature splicing module performs dimension reduction on the feature map through a pooling module and 1 * 1 convolution, the neural network parameter quantity and the reasoning calculation quantity are reduced, and faster and more efficient reasoning calculation can be realize