The invention provides a thermodynamic diagram-based network structure for quickly counting stephanitis chinensis drakes. The network structure comprises a backbone network, a first feature fusion network and a second feature fusion network, wherein the backbone network comprises a convolution layer in a VGG16 network structure. Through the fusion of the backbone network and the first feature fusion network, shallow features are fused to deep features, and the spatial information and the detail information of the deep features are supplemented and enriched. In addition, through the fusion of the backbone network and the second feature fusion network, the deep features are reversely fused to the shallow features, so that the situation that the spatial information and the detail informationof the shallow features cannot be fully expressed after the deep feature information is enriched is avoided. Through the bidirectional feature fusion structure, the spatial information and the detailinformation of the shallow features and the deep features can be fully expressed, so that the stephanitis chinensis drakes can be identified and counted timely, quickly and accurately, and interference is avoided.