Traditional imaging technology requires that there must be no obstacles between the detector and object. Single-pixel imaging technology does not have this requirement. Due to the various advantages of single-pixel imaging, such as easy construction of the physical system, low cost, and wide spectrum imaging, it has quickly gained significant attention and development in the imaging field. However, single-pixel imaging uses a single-point detector, which requires measurements equal to the pixel number of the target object to obtain a relatively perfect imaging quality. As a result, how to balance the imaging quality and the imaging speed has become not only a technical challenge but also a meaningful task. An adaptive sampling method based on Fourier transform single-pixel imaging is proposed to solve this problem. This method takes advantage of the characteristics of Fourier transform spectrum energy concentration to reasonably plan the sampling path in the frequency domain. Then it measures the spectral coefficient along the planned path, calculates the variance of the spectral coefficient of each sampling path, and performs curve fitting online. When the slope of the curve reaches a range small enough (or close to 0), it will automatically stop sampling and perform an inverse Fourier transform to realize image reconstruction. Thus, the proposed method greatly improves imaging efficiency.