Spectrum Sensing Based on Spectrogram and SE-Resnet

With the increasing prominence of the contradiction between limited wireless spectrum resources and the growing demand, spectrum sensing, which enables secondary users to detect target frequency bands for spectrum reuse, is crucial. The previous single - node spectrum sensing methods, including traditional methods and deep - learning - based methods, have certain drawbacks. For example, traditional methods such as energy detection face problems in threshold determination. Some deep - learning - based methods perform poorly under low signal - to - noise ratios and are prone to overfitting. This paper presents a spectrum sensing algorithm based on spectrogram and SE - Resnet. The algorithm converts time - series signals into spectrograms and inputs them into the SE - Resnet network for feature extraction and determination of channel occupancy. Experimental results demonstrate that this algorithm outperforms CNN and SSCL methods in terms of overall performance and accuracy. Moreover, spectrogram processing can prevent overfitting and enhance the generalization ability of the model.