Spectrum sensing is one of the key tasks of cognitive radio to monitor the activity of the primary user. The sensing accuracy of the secondary user is dependent on the signal-to-noise ratio of the primary user signal. A novel Multi-head Attention-based spectrum sensing for Cognitive Radio is proposed through this work to increase the detection probability of the primary user at a low signal- to-noise ratio condition. A radio machine learning dataset with a variety of digital modulation schemes and varying signal-to-noise ratios served as a training source for the proposed model. Further, the performance metrics were evaluated to assess the performance of the proposed model. The experimental results indicate that the proposed model is optimized in terms of the amount of training time required which also has an increase of 27.6% in the probability of detection of the primary user under a low signal-to-noise ratio when compared to other related works that use deep learning.