To achieve refined management of public safety,this study proposes optimization and improvement strategies for three types of classic multi task cascaded convolutional neural networks,which have low accuracy in face detection under complex environ-mental conditions.The proposed improvement strategies include depthwise separable convolution,integrated image information convo-lution residual module,and data dimensionality reduction measures.Simulation experiments show that the proposed network,opti-mized network,and output network in the improved model can achieve face detection accuracy of 93.45％,95.77％,and 97.78％,respectively.They exhibit higher stability and convergence speed during the training process.Therefore,this model can provide tech-nical support for public safety.
关键词
公共安全/MTCNN/人脸检测/通道剪裁/权重量化
Key words
fine tuned governance of public safety/MTCNN/face detection/channel clipping/weight quantification
维普
万方数据