The light guide plate is the main comp onent of the backlight module of the liquid crystal display, during the process of installation and cleaning,it is inevitable that some ligh t guide plates will be affected by sharp external forces and cause line scratch defects,which will cause decline of prod uct quality.In order to improve the detection accuracy of this type of defect,on the basis of analyzing the optical characteristics of light guide plate and its defects,this paper proposes a detection method for line scratch defects on the light guide plate based on self-adapting LoG filter and residual convolutional autoencoder.First,to make the difference between the defect, the light guide point and the background clearer,an adaptive LoG filter is desi gned to sharpen the edges of the light guide plate image.Second,design a residual convolutional autoencoder whi ch is better for repairing defects, the network structure of this autoencoder can make decoder identical share infor mation from encoder,input the sharpened image to trained residual convolutional autoencoder and obtain defect repaired light guide plate image. And then,subtract light guide plate image from repaired image and accurately ex tract line scratch defects by performing global threshold segmentation and region feature selection.Finally, a lot of experiments are carried out on the light guide plate images collected at the industrial site.The experiment al results show that this algorithm has high efficiency and accuracy,and the accuracy of line scratch defects detection can reach 98％.