Abstract: In order to realize the objective and intelligent classification of the defects of welded joints of marine engineering steel parts, digital radiographic images were taken as the research object to study the defect identification based on improved Fuzzy C-means (FCM) clustering-algorithm as well as particle swarm optimization-support vector machine (PSO-SVM). Firstly, the interference noise in original images was removed based on contrast limited adaptive histogram equalization (CLAHE), and the pixel weighting coefficient ω was introduced to improve FCM clustering-algorithm for image segmentation. Then, the texture features of images were extracted based on gray level co-occurrence matrix, the dimensionality of feature data was reduced by principal component analysis, and the parameters were optimized by combining PSO and SVM classification. As a result, a continuous variable classification model between texture features and defect types was established. Finally, the effectiveness and accuracy of the defect identification model were verified by manual statistics of the completely correct evaluation results. The results show that the accuracy of the established recognition model is 96.1%, and after verification, the recognition accuracy of this model is 95.2%. Compared with the model without CLAHE enhancement and the back propagation (BP) neural network model, this model can well realize the identification of common defects with small error, which can be applied to the digital radiographic identification field of welding defects of marine steel.