Audio Keyframe Extraction of Belt Conveyor Inspection Robot Based on Frequency Band Entropy and Pixel Significance

As the main transportation equipment in coal mines, the safety and stability of belt conveyors are of great significance for the production efficiency and safety of coal mines. The traditional manual inspection method has problems such as low efficiency, poor real-time performance, and potential safety hazards. Fixed sensors have problems such as difficult deployment, difficult maintenance, and high cost. The emergence of belt conveyor inspection robots has provided a new mode for inspection methods. Unlike a fixed sensor with a single fixed mode of data, the data collected by the inspection robot during its movement is dynamic and time-varying. The traditional fixed sensor fault diagnosis methods often have poor results when directly applied to inspection robots. This article proposes an audio frame extraction technique based on frequency band entropy and pixel significance, based on the dynamic audio signals collected by a belt inspection robot and the comprehensive utilization of audio and image processing techniques. Frequency band entropy can extract faults or abnormal noise frames with concentrated audio energy, and image significance can extract faults or abnormal noise frames caused by slight continuous collisions. Through experimental verification, this method can achieve the extraction of abnormal audio frames dynamically collected by inspection robots, and can extract audio near faulty rollers such as breakage, wear, rotor blockage, and bearing damage. The extraction of audio keyframes can provide basic data for subsequent roller fault localization and recognition, and provide research ideas for roller fault diagnosis methods based on inspection robot audio.