| X-ray defect detection is an important technology in industrial non-destructive testing.In most actual production process,the detection of X-ray film is manually determined by the inspector.The quality of manual detection is not only strict with the detection environment but also depends on the experience and working status of the quality inspectors.It is difficult to ensure the detection efficiency and accuracy.Therefore,efficient and reliable computer-aided diagnosis algorithms are critical to the X-ray detection.In this paper,we first propose a multi-scale multi-feature fusion damage location method based on classifier and sliding window strategy.This method first enhances the features by the G-CLAHE method proposed in this paper.The method add the experience of manually detecting damage in the weld to the algorithm by extracting weld location information by clustering.Multi-feature fusion data was obtained by combining the first two special feature with the original data.On the other hand,the method uses a multi-scale detection model to locate the damage on original image,so that the model obtains more spatial information and uses a smaller bonding box to locate damage.Finally,the method introduces the active learning strategy,which alleviates the problem of high cost of labeling in the field of industrial non-destructive testing using deep learning algorithms for computer-aided diagnosis algorithm.Experimental results confirm the effectiveness of each module of the method proposed in this paper.Finally,an improved damage location model based on target detection method is proposed.In order to improve the visualization effect of the model’s overall damage location results,this paper uses the non-maximum value suppression algorithm to remove the non-essential low-confidence bonding box and uses the region merging algorithm to connect multiple bonding box.The visual effect of the damage location that the model presents to the user is ultimately improved. |
PDF Full Text Request