关键词: 室内定位, RSSI, 胶囊网络, 卷积神经网络, 图像分类

Abstract: Aiming at the problem of low positioning accuracy of Wi-Fi indoor positioning method based on location fingerprint, this paper proposes a Wi-Fi indoor positioning algorithm model that integrates convolutional neural network(CNN)and capsule network(CapsNet) which is recorded as CNN-CapsNet. Firstly, the collected RSSI time series information is used to generate the location fingerprint image dataset. Then the CNN primary feature extractor composed of convolution layer and pooling layer is used to complete the conversion from the positioning image to the primary feature map. Finally, the primary feature map is input into CapsNet to obtain the final classification result. The experimental results show that the accuracy of this model is as high as 99.99% and the loss function is as low as 0.009 91 under different vector dimensions and iteration times, which is better than other traditional positioning methods.

Key words: indoor positioning, received signal strength indicator, capsule network, convolutional neural network, image classification