Improvement of Face Detection Algorithm Based on Lightweight Convolutional Neural Network

In recent years, the application of visual processing algorithm based on deep network on mobile terminal has become a popular research problem. In this paper, a new face loss function is designed to solve the problems of limited computing power and storage resource of mobile devices when face detection is applied to mobile terminal. The face detection algorithm based on SSH (single stage headless face detector) is improved in two aspects: firstly, the feature of sample data is extracted by using the lightweight convolutional neural network based on MobileNet, which can effectively reduce the amount of parameters and calculation of the model; Secondly, the deformable convolution layer is introduced into the detection module of SSH network to improve the modeling ability of face deformation. At the same time, the method of knowledge distillation is introduced. The high-precision large network model is used as the teacher network, and the small network model of the improved SSH algorithm is used as the student network to further improve performance of the detection model. Experimental results on the wider face dataset show that the proposed method can significantly reduce the complexity of the model, effectively reduce the size of the face detection model and improve the speed of model detection, while ensuring the detection accuracy. Therefore, the proposed method is expected to be better applied to face detection equipment in mobile terminals.