Classification of Indoor Building Materials for Future Wireless Networks: A Machine Learning Technique

Abstract

The ever increasing indoor wireless traffic has necessitated an accurate classification indoor building materials to improve the performance of wireless networks in enclosed environments. This paper proposed a convolutional neural network technique to accurately classify building materials and structures on indoor environments. The model is trained, tested and evaluated using the received signal strength (RSS) signatures from indoor corridors made up of bricks, concrete and glass. The RSS data was collected using a custom-made channel sounder with a measurement environment of 10 m long. The data was captured at five different points in all the three scenarios. The findings from CNN model outperformed the traditional classifiers which have applied RSS for classification of indoor building materials. The model achieved a performance accuracy of 90%. The model also showed little impact of distance from the transmitter of its ability to accurately classify the building materials. The model’s accurate of classification of indoor building materials will improve the efficiency and adaptability of indoor wireless networks.