In this research, autonomous inspection of steel pipe weld lines for a single class of defects is done using a frequency analysis combined YOLOv5 (You Only Look Once) model. Since steel pipes are vastly implemented in high-risk applications, it is essential to accurately inspect them for defects. A new method is presented to enhance the training process of YOLOv5 using frequency analysis. Two YOLOv5 models are trained using frequencymodified and original images, and the results and training process of these two models are compared. In order to produce frequencymodified images, the first 50 frequencies are removed from the Fourier transform, resulting in a new image set. Results revealed that removing lower frequencies leads to more smooth behavior in indices of the YOLOv5 model during different epochs and reduces training time by 15%. I.