Face Recognition Based on The Improved MobileNet
You Zhou, Yiyue Liu, Guijin Han, Fu Yiping · 2019
Within the average pooling area of the lightweight MobileNet network, The problem that the global feature performance is insufficient due to the difference of contribution of each element, A weighted pooling method based on sensitive location is proposed. Different adaptive contribution values are assigned to the elements in the pooling area, which makes the spatial information more fully expressed and reduces the information loss. At the same time, the improved MobileNet model is applied to face recognition. The experimental training set is derived from the CASIA-WebFace data set, and the improved MobileNet algorithm is validated by LFW data set. Through comparing with the original model, it is found that the improved MobileNet algorithm makes the model more effective and obtains a better recognition accuracy, which can be increased by 1% at most.