The invention discloses an object detection method and apparatus based on self-adaptive frequency domain noncoherent accumulation. The method comprises: first of all, carrying out FFT and operation on the input signals of a frame of an N point to obtain frequency domain data; then calculating an input signal-to-noise ratio SNR1 before accumulation of the frequency domain data; comparing the SNR1 with a first-stage threshold, determining whether to accumulate the frequency domain data, and obtaining output data; if the SNR1 is greater than the first-stage threshold, not performing the accumulation; if the SNR1 is smaller than the first-stage threshold, determining a second-stage threshold through the SNR1, performing the accumulation by use of an improved frequency domain noncoherent mode, and when an output signal-to-noise ratio reaches the second-stage threshold, stopping the accumulation; then performing autocorrelation operation on output data to obtain object data; and finally, performing peak detection on the object data, and taking a frequency point corresponding to the maximum spectrum value as an object frequency point. According to the invention, the noncoherent accumulation frequency is determined and adjusted through the signal-to-noise ratios before the accumulation and after the accumulation, the relaizaiton is simple, the disadvantage of fixed accumulation frequency by use of a conventional method can be overcome, the computational complexity is effectively reduced, and the real-time detection performance is improved.