The invention discloses a mobile phone curved glass typical defect online detection device and method. The online detection device comprises a work-piece transmission device, wherein the work-piece transmission device comprises a transmission guide rail, a plurality of transparent objective tables, a motion control card and a motor, and a rotating table; a plane part detection device, a curve edgedetection device and a R angle part detection device for forming three stations are orderly arranged along a transmission direction of the work-piece transmission device; the online detection methodperforms previous training learning on the background and scratch, burning, edge breakage, pit and like typical defect sample images through a convolutional neural network, and original images collected at three stations are divided into multiple sub-images consistent with the size of the training sample, the segmented sub-images are quickly identified, thereby judging whether the original imageshave defects or judging the defect variety. The scratch, burning, edge breakage, pit and like typical defect sample images can be online detected in real time, and the detection device has the features of being fast, efficient, precise and stable.