Abstract: The girth weld defect of pipelines is an important factor affecting the safety of long-distance oil and gas pipelines in service.However, the MFL inline inspection signals at pipeline girth welds are relatively complex, and thus it is hard to classify the defects with the traditional manual method. Hence, an intelligent classification method was proposed based on the Deep Convolutional Neural Network(DCNN) for the girth weld defects in MFL inline inspection. Specifically, the MFL inline inspection signal images of pipeline girth welds were used as the samples, and a database was established with the defect types found through radiographic inspection after excavation at the girth welds. On this basis, the data set, after being extended and enhanced with the Deep Convolution Generative Adversarial Network(DCGAN), was used for the improvement and iterative training of the residual network. Then, the MFL inline inspection signal images of the girth welds were classified with the trained residual network. The results of practical application show that this method can be used to identify and classify the common strip and circular defects at girth welds, the accuracy of the classification test is 83%-88%, and the recall rate is over 97% for circular defects. Generally, this method breaks through the limitation of manual analysis on MFL inline inspection signal of girth welds, and is capable of providing a reference for the intelligent classification of girth weld defects.