Abstract: The research was carried out for real-time radiographic inspection system in this paper.Combining with the characteristics of real-time radiography,the automatic extraction and recognition techniques for weld defects were studied.The GFO algorithm was employed to extract weld defects,which was proved to be very effective.Meanwhile,a set of parameters were presented for the description of defects and the detailed calculation methods for these parameters were presented.Moreover,a method based on backpropagation neural network was proposed for the recognition of weld defects and gave the corresponding results.It was indicated through experiments that this method was more reliable and adaptable than other traditional methods.In addition,the image preprocessing methods and the modification method for defects extraction results were introduced.The methods presented in this paper had been applied in practical inspection of weld defects by which very good results were obtained.