为提高油气管道环焊缝缺陷的识别准确率和质量评价水平，梳理了管道环焊缝缺陷特征，从形状特征、图像质量和位置特征等方面构建了多维度图像特征体系，提高了数据集的利用效率，并从卷积层和损失函数两方面对模型进行了改进，形成iYOLOv8模型，用于实现缺陷的智能识别。结果显示，形状特征对分类准确率的贡献最大，其次为图像质量和位置特征；改进iYOLOv8模型的分类准确率、召回率和平均精度值分别为98.67%、75.49%和87.98%，与常规YOLOv8模型相比，分别提高了11.46%、21.57%和20.78%；可视化结果显示，改进iYOLOv8模型的识别精度均符合或高于规范要求，且优于其余对比模型的识别结果。

 To improve the identification accuracy and quality evaluation level of girth weld defects in oil and gas pipelines, the characteristics of girth weld defects were summarized. A multi-dimensional image feature system was constructed from the aspects of shape features, image quality, and position features to improve the utilization efficiency of data sets. The model was improved in terms of the convolution layer and loss function. An improved YOLOv8 model was then formed to realize intelligent defect identification. The results show that shape features contribute the most to classification accuracy, followed by image quality and position features. The classification accuracy, recall rate, and average accuracy of the improved iYOLOv8 model are 98.67%, 75.49% and 87.98%, respectively, 11.46%, 21.57% and 20.78% higher than those of the conventional YOLOv8 model. In the visualization results, the identification accuracy of the improved iYOLOv8 model meets or is higher than the specification requirements, and is superior to the identification results of other comparison models.