Abstract: Aiming at the problems of low efficiency, high missed detection rate and others in manual inspection of belt conveyors, an integrated intelligent inspection scheme of "track-type inspection platform-multi-source perception-edge decision-making" is proposed. The system adopts a state machine/event driven architecture to integrate infrared thermal imaging, gas/temperature and humidity, visible light, and acoustic channels; Complete time sequence alignment and denoising at the feature layer, and introduce weighted voting and conflict resolution at the decision-making layer. The industrial testing shows that the average response delay of edge decision-making is≤200 ms (emergency stop channel ≤50 ms), and multi-source fusion can comprehensively identify three types of events: "deviation, idler roller and motor overheating, and smoke" with F₁≥0.90; With a single inspection of 29 minutes, single charging of 5.5 h, and a missed detection rate of <2.17%, which can replace 1-2 inspection positions; The direct economic benefit in 6 months is 496 500 yuan, and the personnel exposure time is reduced by about 175 people per hour. The scheme has generalability and practical engineering value in long-distance belt conveyor scenes.