Spectrum sensing stands as a fundamental technology within Cognitive Radio, pivotal for efficient spectrum allocation and utilization enhancement. To expedite and refine spectrum occupancy determination, we introduce a semi-supervised signal detection algorithm leveraging spectral morphology. Our approach entails initial analysis of spectrally stable signals’ homogeneity, facilitating the extraction of morphological features from the spectral envelope for steady-state signal detection. Simultaneously, in addressing random signals present within spectral gaps, we propose an adaptive threshold line algorithm for their detection. Central to our method is the implementation of a sliding window mechanism, where preceding windows furnish feature and threshold line insights for subsequent ones, thereby ensuring real-time detection effectiveness. Notably, we incorporate prior information, such as known signal bandwidth and peak values within the monitored band, typically obtained through manual labeling. Through empirical evaluations across broadcast, aviation communication, wireless microphone, and mobile communication frequency bands, our experiments demonstrate superior signal detection performance compared to existing schemes.