堆石混凝土入仓堆石的分割识别与统计评价方法研究

Research on segmentation, identification, and statistical evaluation methods for rockfill placement in rock-filled concrete

摘要

【目的】堆石混凝土浇筑施工质量与仓内堆石体的粒径形状参数密切相关，但目前工程现场对于入仓堆石的评判和控制，强烈依赖工人的主观经验和判断，因而，有必要研究并提出快速有效的堆石混凝土入仓堆石分割识别与统计评价方法。【方法】基于YOLO算法框架提出了改进的堆石分割识别方法，给出了考虑图像边界堆石影响的粒径与形状参数的统计计算方法，结合DBSCAN算法提出了逊径料集中区域的判别方法；而后，结合规范要求和工程实际，提出了硬指标与软指标相结合的入仓堆石粒径形状参数的分区评价方法。【结果】结果显示：堆石、标杆与安全帽等分割识别模型的平均精度mAP50分别为92.3%、99.4%与97.5%,识别精度较高；给出的粒径形状参数估计与逊径料集中区域判别方法，经室内与现场算例验证，取得了较好的估计判别效果；提出的分区统计评价方法，结合典型工程进行了验证分析，评价结果与工程实际相符。【结论】提出的堆石分割识别与统计评价方法能够对现场仓内堆石质量进行有效分析与评价，可为堆石混凝土工程入仓堆石的分析评价提供必要技术手段。

Abstract

[Objective] The construction quality of rock-filled concrete placement is closely related to the particle size and shape parameters of the rockfill inside the placement areas. However, current on-site evaluation and control of rockfill placement heavily rely on workers' subjective experience and judgment. Therefore, it is necessary to develop a rapid and effective method for segmentation, identification, and statistical evaluation of the rockfill placement used in rock-filled concrete. [Methods] Based on this, an improved rockfill segmentation and identification method was proposed within the YOLO(You Only Look Once) algorithm framework. Furthermore, a statistical calculation method for particle size and shape parameters was presented that accounted for the influence of boundary rockfill in images. A discrimination method for identifying concentration zones of undersized rockfill materials was proposed using the DBSCAN(Density-Based Spatial Clustering of Applications with Noise) algorithm. Then, based on relevant standards and engineering practices, a zoning evaluation method combining hard and soft indicators for the particle size and shape parameters of rockfill placement was developed. [Results] The result showed that the mean average precision(mAP50) for the segmentation and identification models of rockfill, reference markers, and safety helmets were 92.3%, 99.4%, and 97.5%, respectively, demonstrating high recognition accuracy. The proposed method for particle size and shape parameter estimation and the identification of concentration zones of undersized rockfill materials were validated through laboratory and field case studies, demonstrating reliable estimation and discrimination performance. The zoning statistical evaluation method was verified in typical engineering applications, with result consistent with actual engineering conditions. [Conclusion] The segmentation, identification, and statistical evaluation method proposed in this study enables effective field analysis and evaluation of rockfill placement quality, providing an essential technical solution for evaluating rockfill placement in rock-filled concrete engineering.

关键词

堆石混凝土 / 堆石分割识别 / 改进YOLO算法 / 逊径料集中区域 / 统计评价 / 影响因素

Key words

rock-filled concrete / rockfill segmentation and identification / improved YOLO algorithm / concentration zones of undersized rockfill materials / statistical evaluation / influencing factors

基金资助

国家自然科学基金重点项目(52039005)