| Autor: |
Zhang, Rui, Li, Ji, Fu, Liuhu, Pan, Lihu, Ren, Wenyu, Jin, Mengyan, Song, Jinlong |
| Zdroj: |
Multimedia Tools & Applications; Jan2024, Vol. 83 Issue 3, p8241-8265, 25p |
| Abstrakt: |
The intelligent identification of the shape, category, and location of X-ray stainless-steel weld defects is one of the important links to ensure the safety and reliability of welding structures. At present, detection and research on X-ray stainless-steel weld defects has mainly focused on classification and recognition of weld defects in static images using deep learning methods, whereas relevant research on semantic segmentation is rarely seen. To achieve semantic segmentation of stainless-steel weld defects, this paper proposes an Salient Region-Guided Error Correction Network, whose Salient Region Enhancement module performs salient object detection on the input image, and which is used to extract significant clues of X-ray weld sample defects by reducing interference by redundant information. In addition, the new Error Correction Attention module, by controlling semantic information flow, corrects the insufficient extraction of effective feature information caused by invalid, redundant, or wrong judgments and the weighting of potential operations in convolution operations. To fully and strongly verify the effectiveness of the network, the joint mechanism proposed in this paper was combined with various state-of-the-art methods on the X-ray stainless-steel weld data set, the VOC2012 data set and a publicly available medical image dataset, and a large number of experiments were conducted. Experimental results show that the network proposed in this paper is helpful in improving the accuracy of semantic segmentation. The model offers both high category recognition accuracy rate and good generalization ability. [ABSTRACT FROM AUTHOR] |
| Databáze: |
Complementary Index |
| Externí odkaz: |
|
Full text from SpringerLink