The invention discloses a sugarcane seed screening system based on deep learning, which comprises a pretreatment conveying device, a qualified sugarcane seed conveying device, an image recognition device, a screening device and an unqualified sugarcane seed conveying device, the pretreatment conveying device arranges piled sugarcane seeds into a single conveying state to enter the qualified sugarcane seed conveying device, so that the image recognition device recognizes the sugarcane seeds conveniently, and the screening device removes inferior sugarcane seeds conveniently; the invention also discloses a method for screening sugarcane seeds by using the system, and the inferior seed feature recognition system is constructed based on a deep learning convolutional neural network, so that surface features which are not beneficial to sugarcane growth, such as too close incision to sugarcane buds, incision damage, too long sugarcane buds, wormholes, cracks and the like, can be recognized and detected; an effective mode is provided for screening sugarcane seeds in a sugarcane segment form generated by a pre-cutting type sugarcane cutting machine, a segment cutting type sugarcane harvesting machine and the like, the sugarcane inferior seed screening rate can be effectively increased, and the sugarcane planting cost and the manual labor intensity are reduced.