首页|期刊导航|北京测绘|优化Mask R-CNN的测量标志检测与智能巡检

优化Mask R-CNN的测量标志检测与智能巡检OA

Survey mark detection and intelligent inspection based on an optimized Mask R-CNN

中文摘要英文摘要

针对测量标志巡检中人工巡查效率低、复杂场景下自动识别易受光照变化、遮挡干扰和背景复杂度影响等问题,本文提出一种优化掩膜区域卷积神经网络(Mask R-CNN)的测量标志检测与智能巡检方法.该方法以Mask R-CNN为基础,通过超参数优化策略,对模型训练过程中的关键超参数进行自动搜索与优化,提高模型在复杂场景下的特征提取与目标检测能力,并将模型部署于省级自然资源巡查管理平台开展应用验证.结果表明,该方法精确率达到94.30%,F1分数达到91.13%,在复杂环境下具有较好的识别效果与稳定性,平台应用后巡检效率明显提升,漏检风险有效降低.该方法能够提升测量标志智能识别与巡检管理水平,为测量标志保护与巡检管理的信息化、智能化提供技术参考.

To address the problems of low manual inspection efficiency for survey marks and the susceptibility of automatic tar-get recognition in complex field scenarios to interference from illumination variations,occlusions,and high background com-plexity,this paper proposed a survey mark detection and intelligent inspection method based on an optimized mask region-based convolutional neural network(Mask R-CNN).Built on the baseline Mask R-CNN framework,the proposed method adopted a hyperparameter optimization strategy to automatically search and optimize key hyperparameters in the training process,so as to improve the model's feature extraction and object detection capabilities under complex scenes.The optimized model was subse-quently deployed on a provincial-level natural resources inspection and management platform for practical application valida-tion.Experimental results show that the proposed method achieves a precision of 94.30%and an F1-score of 91.13%,exhibit-ing robust recognition performance and stability in complex environments.Following platform deployment,inspection effi-ciency has been significantly improved,and the risk of missed detections has been effectively reduced.This method can improve the level of intelligent survey mark recognition and inspection management,providing technical reference for the infor-matization and intelligentization of survey mark protection and inspection management workflows.

潘诗辰;徐攻博;王紫琦;严灏;陈铭鸿

浙江启真信息科技有限公司,浙江 杭州 310012浙江启真信息科技有限公司,浙江 杭州 310012浙江省测绘科学技术研究院,浙江 杭州 311122杭州师范大学 信息科学与技术学院,浙江 杭州 311121杭州师范大学 信息科学与技术学院,浙江 杭州 311121

天文与地球科学

测量标志目标检测掩膜区域卷积神经网络(Mask R-CNN)超参数优化智能巡检

survey markobject detectionmask region-based convolutional neural network(Mask R-CNN)hyperparam-eter optimizationintelligent inspection

《北京测绘》 2026 (8)

1152-1158,7

浙江省自然资源厅科技项目(2024ZJHY001)

10.19580/j.cnki.1007-3000.2026040129

评论