A binary multi-band power distribution (BMBPD)-based low signal-to-noise ratio sound event detection method, which uses a combination of BMBPD and random forest (RF) to provide excellent performance. The BMBPD binarizes pixels the grayscale of which is less than a certain threshold in an MBPD image to 1, and the remainder to 0, which may highlight pixels related to a sound event in the MBPD image while suppressing the influence of noise, thereby reducing the influence of noise in a low signal-to-noise ratio environment on a sound event to be measured. By means of performing discrete cosine transform (DCT) on an BMBPD image block, a main part of Z encoding of a DCT coefficient is used as a feature of the sound event, i.e. BMBPD-DCTZ, and an RF classifier is used to train and detect the BMBPD-DCTZ. The method has a wide range of applications to sound levels and has strong robustness in severe non-stationary noise.
展开▼