基于二维卷积神经网络的无线局域网室内定位系统

为改善现有无线局域网（Wireless Fidelity, WIFI）室内定位算法的精度与复杂度问题，提出了一种基于二维卷积神经网络（2D-Convolutional Neural Network，2D-CNN）的WIFI室内定位算法。该算法将在线阶段的复杂性转移到离线阶段，在线阶段中仅使用2D-CNN网络进行训练；在离线阶段中，采集定位区域各采集点可接收到的所有无线接入点（Access Point，AP）的接受信号强度（Received Signal Strength Indicator，RSSI）值，并根据其计算峰值，二者结合构成位置指纹图像。再使用滑动窗口进行数据集扩充，最后将其引入到2D-CNN网络模型中进行训练，建立定位模型并完成定位。实验结果表明，在当前室内环境中，该算法的平均定位精度达99.58%，证实了不同参数、优化算法及模型架构选择的正确性。

2D Convolutional Neural Network based WIFI indoor localization system

In order to improve the accuracy and complexity problems of existing WIFI indoor positioning algorithms, a WIFI indoor location algorithm based on 2D-CNN was proposed. The algorithm transfers the complexity of the online phase to the offline phase, in which only the 2D-CNN network is used for training; in the offline phase, the RSSI values of all wireless APs that can be received at each collection point in the localization area are collected and the peaks are calculated based on them, and the two are combined to form a location fingerprint image. The data set is then expanded using sliding windows, and finally introduced into the 2D-CNN network model for training to build the localization model and complete the localization. The experimental results show that the algorithm achieves an average localization accuracy of 99.58% in the current indoor environment, confirming the correctness of the different parameters, optimization algorithm and model architecture selection.