Spectrum sensing is essential for cognitive radio, while wideband spectrum sensing (WSS) capabilities could enable secondary users to seize more transmission opportunities in frequency domain, where the signal location of primary users might be unknown in practice. To ensure the system stability, it is necessary for spectrum sensing under the condition of low signal-to-noise ratio (SNR). However, the detection performance of conventional energy detection (ED) methods is unsatisfactory when the SNR of the primary user signal is low. Motivated by the correlation attribute among frequency points of each signal in the wideband spectrum, this paper studies the theoretical relationship between SNR of the PU signal and frequency correlation parameters, i.e., the size of correlation window (CW) and the value of frequency correlation ratio. Based on this, an Adaptive Frequency Correlation-based (AFC) method is proposed. This method that does not require any prior knowledge specifically considers the detection of low SNR signals. These two parameters related to frequency correlation can be adaptively adjusted according to the SNR of each signal to be detected. Moreover, to ensure the detection performance of the proposed AFC method, we present a low-complexity inflection-point estimation algorithm which is capable of estimating the noise power, SNR and possible frequency-domain position of the signal quickly and accurately. Numerical results show that the proposed AFC method is suitable for WSS scenarios with unknown prior information. It is capable of detecting signals with bandwidths greater than their minimum detectable bandwidths which are determined by the size of CWs. Additionally, simulation results indicate that the AFC method outperforms the state-of-the-art ED methods, as it shows evidently better detection performance for low SNR signals with acceptable false alarm rate.