Aiming at 304 austenitic stainless steel paramagnetic material,a multi-class classification support vector machine LIBSVM algorithm was carried out for the magnetic anomaly inversion of weak magnetic signals. Orthogonal test design was used to make 4 testing samples with an artificial defect. The characteristics of artificial defect such as width,amplitude,and area were extracted for the collected magnetic anomalies of the defect signals. The LIBSVM algorithm was used to establish the relationship between the magnetic anomalies of the defect signals and the artificial defects. A total of 16 groups testing results indicate that 10 of the 16 defects inversion are correct in length and width,and 8 defects inversion are correct in depth. The inversion accuracy of defect length,width and depth is high,which provides a reference basis for the study of magnetic anomaly inversion for weak field signals,related experimental methods and equipment development.