Treffer: 基于CT-ST半监督模型的城市地下管道缺陷语义 分割研究.

Defect segmentation of urban underground pipes using machine vision technology is an industrialized intelligent development trend. Since conventional supervised methods require a large number of annotations for defect segmentation task, an improved CT-ST semi-supervised semantic segmentation model based on the ST semi-supervised model is proposed, which is firstly applied to the field of defect segmentation of urban underground pipelines. The model is based on the self-training method of semi-supervised semantic segmentation domain, combined with the idea of Co-teaching algorithm, distinguishes different quality pseudo-labels, and utilizes a one-time pseudo-label screening strategy instead of the traditional set-threshold iterative method, to reduce the impact of erroneous feature training due to low-quality labels; for the problems of complex background of underground pipelines, multiple defect categories, multiple scales, and multiple noises, we introduce a NAM attention mechanism into each residual block, to give each important defects a more accurate and more accurate labeling. NAM attention mechanism is introduced in each residual block to increase the weight of each important feature and weaken the proportion of unimportant features. The experiments verify the effectiveness of CT-ST semi-supervised segmentation model, and the mIoU is improved on different proportions of labeled sample sets, in which the mIoU of 1/2 proportion of labeled dataset is 67.36%, which is increased by 2.33% compared with the original model. Compared with many mainstream pseudo-labeling and consistency regularization methods, CT-ST has better performance in terms of accuracy. [ABSTRACT FROM AUTHOR]

利用机器视觉技术进行城市地下管道缺陷分割是工业化智能化发展趋势. 由于常规的监督方法进行缺陷 分割任务时需要大量的标注, 提出了一种基于 ST 半监督模型改进的 CT-ST 半监督语义分割模型, 并首次应用到城 市地下管道缺陷分割领域. 该模型基于半监督语义分割领域自训练方法, 结合 Co-teaching 算法思想, 区分不同质量 伪标签, 利用一次伪标签筛选策略代替传统设置阈值迭代方法, 降低因低质量标签带来的错误特征训练影响; 针对 地下管道背景复杂、缺陷类别多、多尺度、多噪声等问题, 在每个残差块引入 NAM 注意力机制, 给每个重要特征增加 权重, 弱化不重要特征的占比. 实验验证了 CT-ST 半监督分割模型的有效性, 在不同比例有标签样本集上 mIoU 均有 提升, 其中 1/2 比例有标签数据集 mIoU 为 67.36%, 对比原模型增加了 2.33% .与多种主流的伪标签、一致性正则化方 法相对比, 所提出的模型在精度上均有较好的表现. [ABSTRACT FROM AUTHOR]

Copyright of Journal of South-Central Minzu University (Natural Science Edition) is the property of Journal of South-Central Minzu University (Natural Science Edition) Editorial Office and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)