Belt conveyor inspection robot audio key frame extraction method based on frequency band entropy and pixel significance

The invention relates to a belt conveyor inspection robot audio key frame extraction method based on frequency band entropy and pixel significance, and belongs to the field of belt conveyor carrier roller fault detection. The method comprises the following steps: inputting audio signals collected by an inspection robot, and obtaining a time-frequency matrix of the input audio signals through short-time Fourier transform; calculating an entropy value of each row of the time-frequency matrix to obtain an information entropy of each frequency band, and obtaining a frequency band entropy; judging whether the frequency band entropy is lower than a lower threshold value or not, if yes, taking the audio frame as an abnormal key frame for extraction, otherwise, judging whether a condition that n frequency band entropies are lower than an upper threshold value within a period of time exists or not, and if yes, taking the audio frame as an abnormal key frame for extraction; and calculating a pixel saliency index based