FMDet: Face Mask Detection Based on Improved Cascade rcnn

We propose the FMDet detection model to address the issues of low detection accuracy caused by occluded target features, poor detection performance of small targets, and small feature differences between correctly and incorrectly wearing masks in face mask detection. First, in order to ensure the accurate extraction of information, Switchable Atrous Convolution is used to optimize the backbone network, enhance the effective Receptive field, and improve the detection ability of occluded objects. Then use Recursive Feature Pyramids to process the multi-scale features extracted from the backbone network. Through recursive structure, ensure that the extracted features are more accurate and stable, and effectively detect small targets. Finally, Hybrid data augmentation is used to increase the number and diversity of training samples and enhance the robustness of the model. On the PWMFD dataset, the $\mathbf{A P}, \mathbf{A} \mathbf{P}_{75}, \mathbf{A} \mathbf{P}_{\text {without }}, \mathbf{A} \mathbf{P}_{\text {with }}$ and $\mathbf{A} \mathbf{P}_{\text {incorrect }}$ of $\mathbf{F M D e t}$ improved by $2.2,1.9,2.4,2.6$, and 1.5 compared to the baseline, respectively. We have also compared our method with several advanced object detection algorithms, and experimental results show that our method effectively improves the accuracy of mask detection and is superior to existing object detection algorithms.