一种轻量化CNN的WIFI指纹室内定位模型

A Lightweight CNN-Based WiFi Fingerprint Indoor Positioning Model

摘要: 为了提高室内WiFi指纹定位技术的定位精度，减少模型参数量。本文提出一种基于卷积神经网络(Convolutional Neural Network, CNN)的轻量化室内定位模型。首先将接收信号的强度指示 (RSSI) 处理为二维灰度图，然后使用深度可分离卷积进行特征提取，将提取后的特征通过自适应池化层(Adaptive Avg Pool 2d)固定输出大小，减少全连接层参数量；最后输入到全连接层进行分类。在UJIIndoorLoc数据集和Tampere数据集上的实验结果表明：模型的楼层定位分别达到了99%和99.7%的准确度，坐标定位的平均误差为6.51 m，训练参数为48685个。与现有的先进室内定位模型相比，定位精度更高、模型参数更少。

Abstract: To improve the positioning accuracy of indoor WiFi fingerprinting technology and reduce the number of model parameters, this paper proposes a lightweight indoor positioning model based on Convolutional Neural Network (CNN). Firstly, the received signal strength indication (RSSI) values are processed into a two-dimensional grayscale image. Then, deep separable convolutions are used for feature extraction, and the extracted features are passed through adaptive pooling layers to maintain a fixed output size and reduce the number of parameters in fully connected layers. Finally, the features are inputted into the fully connected layer for classification. Experimental results on the UJIIndoorLoc dataset and Tampere dataset show that the model achieves accuracies of 99% and 99.7% respectively for floor-level positioning, with an average coordinate localization error of 6.51 m and 48,685 training parameters. Compared to existing advanced indoor positioning models, this model achieves higher positioning accuracy with fewer model parameters.