The present invention provides an online detection method of cracks of an air channel insulation encapsulated layer of a dry-type hollow shunt reactor. The method mainly comprises the following stepsof: S01: reading endoscope detection image video of a detected dry-type hollow shunt reactor, and decomposing the endoscope detection image video into frames; S02: employing a local multi-scale dark channel endoscope image enhancement algorithm to perform enhancement and denoising operations of the images decomposed into the frames; S03: performing filtering processing of each image by employing wavelet multi scales taking the consideration of the dark channel priori criteria and same translucent color ratio assumption balance, and performing graying processing of each image; S04: employing alocal binary pattern operator according to local properties of endoscope image fracture distribution and gray contract to achieve dry-type hollow shunt reactor air channel insulation encapsulated layer crack edge detection and identification; and S05: determining crack positions of the dry-type hollow shunt reactor air channel insulation encapsulated layer. The method provided by the invention canaccurately achieve positioning of the cracks of the air channel insulation encapsulated layer.