The nondestructive detection of power cable body can be realized by X-ray digital imaging, but there is no depth processing and defect recognition method for X-ray digital images at present. The cable body and defects can not be detected and identified directly from the original digital image. Therefore, this paper studies the advanced processing method of power cable X-ray digital image and the intelligent identification technology of buffer layer defect, and puts forward the full convolution neural network (FCN) method. By using gray level processing technology, the original image gray scale range is compressed to the human eye recognizable range, then the defect identification is carried out, and then the image data is trained by the traditional convolution neural network (CNN) and the proposed method. The intelligent recognition of power cable buffer layer defects is realized. Compared with the traditional CNN, the proposed FCN has more clear and intuitive recognition effect.