<p>To address the problems of low detection efficiency and false detection of small target defects in the intelligent detection of X-ray image weld defects of industrial pressure pipelines, this paper proposed a more efficient and light EL-YOLOv8 weld defect detection algorithm. Firstly, data augmentation was performed to solve the problem of unclear defects and insufficient data sets. In order to improve the original YOLOv8 model, the FasterNetBlock was combined with the Efficient Multi-Scale Attention (EMA) module to design a lightweight multi-scale feature Faster-EMA module, which was fused with the CSPDarknet53 to Two-stage FPN (C2f) module in the backbone network. The C2f-Faster-EMA module is proposed to realize multi-scale feature object detection and enhance the feature extraction ability. The experimental results show that compared with five mainstream defect detection algorithms such as YOLOv8-Ghost, the proposed model achieves 91.5% mAP@0.5 defect accuracy on the self-developed X-ray welding defect image dataset, which is 2.8% higher than that of the baseline model. Compared with four mainstream lightweight models such as MobileNet V2, the parameter number of the proposed model is 2.5, the FPS reaches 205, and the processing speed is the fastest and the model is the lightest while maintaining a high accuracy, achieving lightweight. At the same time, on the public datasets COCO128 and ImageNet100, the AP@0.50:0.95 of our model is 2.5% higher than that of YOLOv8, which proves that this model also has good generalizability.</p>