The work aims to achieve the goal of fully extracting paper defect features, improving detection accuracy and reducing detection rate of small targets in paper defect detection. The detection algorithm was improved based on Faster R-CNN. The main improvements were as follows:the backbone feature extraction network VGG16 of the original model was replaced by the deep residual network ResNet-50 to retain more feature information of paper defect and enhance the feature network's ability to extract paper defects. The dual attention mechanism CBAM of space and channel was added to the algorithm to improve the accuracy of paper defect detection. ROI-Pooling was replaced with ROI-Align to enhance the generalization ability of network. The experimental results indicated that the average accuracy of the improved algorithm reached 98%, which was 9% higher than that of the original algorithm. The improved algorithm can fully extract the feature information, effectively improve the detection accuracy of paper defect, improve the detection rate of small target paper defect, and reduce the error and miss detection rate.