1 College of Software Engineering, East China University o f Technology, Nanchang 330013, China
2 Department of Information and Engineering, East China University of Technology, Nanchang 330013, China
3 Jiangxi Engineering Laboratory on Radioactive Geoscience and Big Data Technology, East China University of Technology, Nanchang 330013, China
Show less
文章历史+
Received
Published
Revised
2025-04-02
2026-03-08
2025-12-09
Online Date
2026-05-28
PDF (22090K)
摘要
滑坡是一种严重威胁生态环境与人类安全的地质灾害,对其进行高效精准的识别对于灾害监测和预警具有重要意义。针对传统目标检测算法在滑坡识别中存在的精度不足、误检率较高等问题,提出了一种基于改进YOLOv7(you only look once version 7)的滑坡目标检测方法。具体方法为:①针对滑坡影像数据的特性,优化YOLOv7网络结构,以增强特征提取能力;②引入卷积注意力模块CBAM(convolutional block attention module)注意力机制,提高对滑坡区域的检测精度;③引入基于最大概率距离交并比的损失函数(maximum probability distance intersection over union,MPDIoU),使得模型在提升检测精度的同时有效降低误检率与漏检率。实验表明,与原始YOLOv7相比,改进后的算法在平均精度值(mean average precision,mAP)上提高了6.3%、精确率提升了5.8%,及参数量相比原模型减少了6.11 M。该方法能够高效识别基于高分辨率遥感影像的滑坡体,具有较强的实际应用价值,可为地质灾害监测和预警提供有效的技术支持。
Abstract
Landslides are geological disasters that severely threaten ecological environments and human safety. Efficient and accurate identification for landslides is crucial for disaster monitoring and early warning systems. To address the issues of low accuracy and high false detection rates in traditional object detection algorithms for landslide identification, a landslide detection method was proposed based on an improved YOLOv7 model. The specific approach includes fouows as these The YOLOv7 network was optimized to improve its feature extraction for landslide imagery; the convolutional block attention module (CBAM) was integrated, which improves the accuracy of detecting landslide regions; a new loss function, MPDIoU, was introduced. This function improves overall detection accuracy by reducing both false alarms and missed detections. Experimental results demonstrate that, compared with the original YOLOv7, the proposed method improves the mean average precision (mAP) by 6.3%, enhances precision by 5.8%, and reduces the model’s parameter size by 6.11 M. The method is capable of efficiently detecting landslide bodies in high-resolution remote sensing images and has strong practical application value. It can provide effective technical support for geological disaster monitoring and early warning systems.
滑坡直接威胁人类生命财产安全,引发次生灾害,破坏生态环境以及各种基础设施[1]。以往基于高分辨率遥感影像进行滑坡识别的目标检测方法,存在计算量大、速度慢,难以快速处理海量遥感数据的问题[2],现很难在精度与速度之间达到平衡[3],双阶段算法,精度较高,但速度较慢;一阶段算法如YOLO(you only look once )系列则在速度上具有优势,但对小目标识别的精度较低[4]。
针对现有边界框回归损失函数在目标尺度及形态适应性方面的不足,引入基于最大概率距离交并比(maximum probability distance IoU, MPDIoU)的改进方法。相较于传统IoU系列损失函数,MPDIoU通过概率空间建模增强了几何差异的敏感性,从而提升目标检测任务中的定位准确性和鲁棒性。MPDIoU的具体计算公式及损失函数分别为
从识别精度和处理速度两个方面,对耦合YOLO和Mask R-CNN算法的区域滑坡灾害识别模型进行评估。采用精确率P(precision)、召回率R(recall)、F1得分(F1-score)、平均精度(average precision, AP)以及平均精度均值(mean average precision, mAP)评价模型的识别精度。采用算法的参数量(params)、浮点运算次数(floating point operations, FLOPs)以及每秒传输帧数(frames per second, FPS)评估模型的参数规模、计算复杂度和识别速度。
ChengJ, DaiX, WangZ, et al. Landslide susceptibility assessment model construction usingtypical machine learning for the Three Gorges Reservoir Area in China[J]. Remote Sensing, 2022, 14(9): 2257.
[2]
BoussaadL, BoucettaA. Deep-learning based descriptors in application to aging problem in face recognition[J]. Journal of King Saud University-Computer and Information Sciences, 2022, 34(6): 2975-2981.
[3]
ZhaoY, ZhangX, FengW, et al. Deep learning classification by ResNet-18 based on the real spectral dataset from multispectral remote sensing images[J]. Remote Sensing, 2022, 14(19): 4883.
[4]
PrakashN, ManconiA, LoewS. Mapping landslides on EO data: performance of deep learning models vs. traditional machine learning models[J]. Remote Sensing, 2020, 12(3): 346-368.
[5]
WangC Y, BochkovskiyA, LiaoH. YOLOv7: trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[J]. arXiv Preprint, arXiv: 2207. 02696, 2022.
[6]
XiaJ, LiuH, ZhuL. Landslide hazard identification based on deep learning and sentinel-2 remote sensing imagery[J]. Journal of Physics: Conference Series, 2022, 2258(1): 012031.
[7]
XieW, NieW, SaffariP, et al. Landslide hazard assessment based on Bayesian optimization-support vector machine in Nanping City, China[J]. Natural Hazards, 2021, 109(1): 931-948.
CaoJiale, LiYali, SunHanqing, et al. A review of visual object detection technology based on deep learning[J]. Journal of Image and Graphics, 2022, 27(6): 1697-1722.
ChenPeng, YangTao. Research on the application of YOLOv7 algorithm in landslide disaster remote sensing images[J]. Disaster Science, 2024, 41(1): 34-42.
ChenYu, LiYong, JiangShuai, et al. Research on real-time detection and early warning methods for landslide disasters[J]. Journal of Geological Hazards and Environment Protection, 2023, 34(4): 112-120.
DuanYuying, TangJun, LiuYuangang, et al. Spatial sensitivity evaluation of loess landslides in Liulin County, Shanxi Province based on random forest[J]. Geography Science, 2022, 42(2): 343-351.
GaoMengjie, ChenFang, WangLei, et al. Landslide extraction research based on Markov random field for PLANET high-resolution satellite images[J]. Remote Sensing Technology and Application, 2023, 38(5): 1180-1191.
GaoZhijie, LiChunxiao, WangGuowei, et al. Research on landslide monitoring based on convolutional neural networks[J]. Computer Science and Exploration, 2022, 15(11): 2125-2133.
HanJing, YangShuai, MaYudong. Identification of geological disaster hazards in the Shangluo Area based on InSAR and optical remote sensing technologies[J]. Western Resources, 2023 (4): 144-147.
JiangWandong, XiJiangbo, LiZhenhong, et al. Mask R-CNN landslide segmentation and recognition based on difficult sample simulation[J]. Journal of Wuhan University (Information Science Edition), 2023, 48(12): 1931-1942.
JuYuanzhen, XuQiang, JinShichao, et al. Automatic identification of loess landslides using deep learning methods[J]. Journal of Wuhan University (Information Science Edition), 2020, 45(11): 1747-1755.
LiHua, WangPeng, ZhangMing, et al. Research on landslide disaster detection methods based on deep learning[J]. Journal of Geological Hazards and Environmental Protection, 2023, 34(2): 45-56.
LiJie, ZhangLei. Research progress on landslide monitoring technology based on deep learning[J]. Journal of Geological Hazards and Prevention, 2023, 42(2): 76-85.
ZhangZhijuan, ShenZheping, XueQitao. Improved lightweight neural network for photovoltaic maximum power point tracking based on knowledge distillation[J]. Science Technology and Engineering, 2025, 25(16): 6781-6788.
JiaXianguang, LiuHuan, FengChaoqin, et al. Prediction of shared bicycle inflow and outflow based on Conv3D-GRU neural network[J]. Science Technology and Engineering, 2025, 25(5): 2127-2134.
YangQingsong, HaoRujiang, FanYafei, et al. A gearbox fault diagnosis method based on improved lightweight neural network[J]. Science Technology and Engineering, 2024, 24(7): 2699-2705.