DS-ECA-CNN：一种新型轻量化CNN的WIFI指纹室内定位模型

DS-EC-CNN: A Novel Lightweight CNN Model for Indoor Localization of WIFI Fingerprints

摘要: 针对在大规模室内环境下多建筑、多楼层定位场景中定位精度不高、模型参数量大的问题，本文设计了一种WIFI指纹室内定位(DS-ECA-CNN)模型，该模型基于卷积神经网络(CNN)进行改进，包括特征提取模块和分类模块，其中特征提取模块由基于深度可分离卷积(DS)模块与注意力模块(ECA)融合的模块(DS-ECA)组成。DS-ECA模块在降低模型参数量的同时，能有效地增强了模型的整体性能表现。在UJIIndoorLoc数据集、Tampere 数据集这两个公共数据集上对模型性能进行了评估，实验结果显示，UJIIndoorLoc数据集上的建筑定位准确率为100%，楼层定位准确率为99.2%；Tampere 数据集上的建筑定位准确率为100%，楼层定位准确率为99.7%。提出的模型和与其他室内定位模型相比，定位精度更高；模型参数量少；存储空间更小。

Abstract: Aiming at the problems of low positioning accuracy and large number of model parameters in multi-building and multi-floor localization scenarios in large-scale indoor environments, this paper designs a WIFI fingerprint indoor localization (DS-ECA-CNN) model, which is improved based on the Convolutional Neural Network (CNN) and includes a feature extraction module and a classification module, in which the feature extraction module consists of a module based on the fusion of the Depth Separable Convolutional (DS) module fused with an attention module (ECA) (DS-ECA). The DS-ECA module can effectively enhance the overall performance of the model while reducing the number of model parameters. The model performance is evaluated on two public datasets, UJIIndoorLoc dataset and Tampere dataset, and the experimental results show that the accuracy of building localization on UJIIndoorLoc dataset is 100% and the accuracy of floor localization on Tampere dataset is 100% and the accuracy of floor localization on UJIIndoorLoc dataset is 99.7%. 99.7% on the Tampere dataset. Compared with other indoor localization models, the proposed model has higher localization accuracy, fewer model parameters, and smaller storage space.