| Popis: |
For the complex problems of invoice occlusion, invoice deformation, dark environment, excessive noise and so on in invoice detection, this paper proposes an improved YOLOv5s invoice detection and classification method. In order to improve the generalization ability of the model, the attention mechanism is introduced to improve the feature extraction ability of the network. By adding cavity convolution to the YOLOv5S backbone network and the neck network, and adding context transformation network to the backbone network, the robustness of the model is improved. For model output, flexible non-maximum suppression is used to replace non-maximum suppression to improve the detection effect. Comparative experiments show that the accuracy, recall and average accuracy of the proposed method are greatly improved. |
