【摘要】 对石油管材落锤撕裂断口进行评定,目前采用的方法主要通过游标卡尺等测量工具进行测量和计算,存在对工作人员经验要求高、主观因素影响大、不规则形貌判别困难和效率低等缺点。针对以上问题提出了一种具有空洞卷积的编解码器模型的管材断口图像语义分割方法,首先对采集好的试样断口进行脆性区域的数据集标记,然后利用标记好的数据集对DeepLabV3+网络模型进行训练,该模型可以有效地分割试样断口中的脆性区域。最后对管材试样断口评定的计算方法进行了基于像素级别的改进,在对实验结果进行分析和对比后表明,所提出的方法具有更高稳定性、高准确率和良好分割效果。更多还原

【Abstract】 When evaluating the petroleum pipe steel drop weight tear fracture, the current method is mainly by measuring with tools such as vernier calipers and calculation. There are disadvantages such as high requirements for staff experience, large influence of subjective factors, difficulty in discriminating irregular morphology and low efficiency. Aiming at these problems, a steel fracture image semantic segmentation method with a codec model with atrous convolution is proposed. Firstly, the data set of the brittle region is marked on the fracture of the collected sample, then the DeepLabV3+ network model is trained by using the marked data set. The model can effectively segment the brittle region in the fracture of the sample. Finally, the calculation method of the fracture evaluation of steel samples was improved based on the pixel level. After analyzing and comparing the experimental results, the proposed method has higher stability, high accuracy and good segmentation effect.更多还原