The invention discloses a bobbin yarn appearance defect classification method based on a deep convolutional neural network. The bobbin yarn appearance defect classification method comprises the following steps of S1, carrying out image cutting; S2, carrying out manual sorting; S3, cleaning the data; S4, reorganizing the data; S5, carrying out data augmentation; S6, carrying out a neural network structure; S7, carrying out performing model training; S8, adjusting parameters; S9, carrying out model testing; and S10, after obtaining the optimal data, starting from the step S3 until the loss valueof the verification set approaches 0 and cannot be reduced, and finally obtaining an optimal model. Compared with a traditional machine vision image processing method, the glass fiber bobbin yarn appearance detection method based on the convolutional neural network has the advantages that mathematical fitting is carried out on surface layer characteristic data extracted from a picture; and finally, a satisfactory result can be achieved by using a simple classifier, so that the reliability and the detection speed are improved, and the labor cost is reduced.