Spoofing attacks are generally difficult to detect and can have potentially harmful consequences on computer networks and applications. Wireless IoT networks, in the context of smart buildings or smart factories, are particularly vulnerable to these attacks. In this paper, we present a new physical device fingerprinting approach aiming at detecting spoofing attacks in wireless IoT environments. The proposed approach is based on the analysis of some properties of the physical signals emitted by connected devices, using their Power Spectral Density (PSD) to extract a frequency profile of their communications. This approach does not require any expensive equipment, is easy to deploy, and is resilient to non predictable phenomena in transmissions. The detection of spoofing attacks consists in comparing the fingerprint of a transmitting device with previously stored fingerprints of legitimate devices, by measuring the similarity of the corresponding PSDs and applying a community detection algorithm. The efficiency of this approach has been successfully tested using various experimental setups with connected devices supporting different wireless protocols (BLE, Zigbee). We also discuss the practical applicability of our approach, e.g. in an industrial environment by analysing its scalability and proposing solutions to tune and optimize its deployment at a large scale.