基于一种改进YOLOv7算法模型的滑坡识别研究——以四川省白格地区为例
|
刘浩然(1999-),男,硕士,主要从事深度学习、遥感滑坡检测研究。Email: 942435351@qq.com。 |
Copy editor: 李瑜
收稿日期: 2024-03-21
修回日期: 2024-06-25
网络出版日期: 2026-06-03
基金资助
浙江地质大数据应用中心有限公司项目“地质灾害遥感解译算法研发”(KZ230752)
河北省大学生创新创业训练计划项目“InSAR与深度学习技术相结合的白格地区滑坡形变监测与识别”(S202211775007)
Landslide identification based on an improved YOLOv7 model: A case study of the Baige area
Received date: 2024-03-21
Revised date: 2024-06-25
Online published: 2026-06-03
滑坡识别一直是地质灾害领域的一个研究热点,对于抢险应急指挥具有重要意义。针对滑坡检测的漏检误检以及识别精度不高的问题,该文通过对核心网络进行数据融合、增加卷积块注意力模块(convolutional block attention module,CBAM)、更改交并比(intersection over union,IoU)损失函数的优化等方法,提出一种改进的YOLOv7算法模型,实现了对滑坡同时进行目标检测与图像分割的功能,并以贵州省毕节市的滑坡数据集和四川省历史滑坡0.2 m高分辨率航空正射影像为例对模型的有效性进行了验证。研究结果表明,优化后的模型在山体滑坡的检测与分割任务中展现出较好的性能,相比于常规的YOLOv7模型以及Faster RCNN,Mask RCNN等主流模型方法,该模型对山体滑坡的识别更加高效、准确。以四川省白格地区为例,该模型在提高精度的同时可有效提高山体滑坡灾害信息获取的自动化程度。
刘浩然 , 严天笑 , 朱月琴 , 王艳萍 , 陈祖谊 , 杨昭颖 , 朱浩濛 . 基于一种改进YOLOv7算法模型的滑坡识别研究——以四川省白格地区为例[J]. 自然资源遥感, 2025 , 37(4) : 48 -57 . DOI: 10.6046/zrzyyg.2024110
Landslide identification has always been a research topic in the study of geological disasters, playing a significant role in emergency rescue and command. To address the limitations in landslide identification, such as missed/false detection, and low identification accuracy, this study proposed an improved YOLOv7 model that enables simultaneous object detection and image segmentation for landslides. The improved model optimized its core network by integrating data, adding the convolutional block attention module (CBAM), and changing the intersection over union (IoU) loss function. Its effectiveness was verified using the landslide dataset of Bijie City, Guizhou Province, and the 0.2 m high-resolution digital orthophoto map (DOM) of historical landslides in Sichuan Province. The results indicate that the improved model performed well in landslide detection and segmentation, achieving more efficient and accurate landslide identification compared to the conventional YOLOv7 model, and other prevailing models like Fast RCNN and Mask RCNN. Taking the Baige area in Sichuan Province as an example, this model can effectively enhance the automation level of landslide disaster information acquisition while improving accuracy.
表1 各项改进的消融实验Tab.1 Various improved ablation experiments |
| 模块 | 改进策略 | 精确率/% | 召回率/% | mAP@0.5/% | |||
|---|---|---|---|---|---|---|---|
| MPDIoU | CBAM | DEM | |||||
| YOLOv7-segment | 35.2 | 83.2 | 73.4 | 75.9 | |||
| YOLOv7-segment-A | √ | 36.3 | 84.7 | 74.5 | 78.2 | ||
| YOLOv7-segment-B | √ | 36.9 | 85.1 | 75.6 | 79.5 | ||
| YOLOv7-segment-C | √ | 35.2 | 86.3 | 76.1 | 78.2 | ||
| 本文方法 | √ | √ | √ | 37.9 | 89.4 | 79.8 | 83.5 |
表2 不同算法对比Tab.2 Comparison of different algorithms |
| 算法模型 | mAP@0.5/% | 参数量/MB | GFLOPs/GB |
|---|---|---|---|
| Mask R-CNN | 75.4 | 45.8 | 66.55 |
| YOLOv5L-seg | 75.9 | 46.56 | 109.60 |
| Faster-RCNN | 70.6 | 41.00 | 241.40 |
| YOLACT | 68.7 | 47.70 | 126.80 |
| YOLOv7-seg | 79.8 | 38.27 | 143.20 |
| 本文算法 | 83.5 | 37.89 | 141.50 |
| [1] |
谭永杰, 刘荣梅, 朱月琴, 等. 论地质大数据的特点与发展方向[J]. 时空信息学报, 2023, 30(3):313-320.
|
| [2] |
张潇远, 苏巧梅, 赵财胜, 等. 一种利用贝叶斯算法优化XGBoost的滑坡易发性评价方法[J]. 测绘科学, 2023, 48(6):140-150.
|
| [3] |
康孟羽, 朱月琴, 陈晨, 等. 基于多元非线性回归和BP神经网络的滑坡滑动距离预测模型研究[J]. 地质通报, 2022, 41(12):2281-2289.
|
| [4] |
洪韬, 朱月琴, 郭艳军, 等. 基于XR技术的交互式地质场景模型构建方法研究[J]. 地质通报, 2023, 42(s1):469-478.
|
| [5] |
|
| [6] |
|
| [7] |
|
| [8] |
|
| [9] |
|
| [10] |
|
| [11] |
|
| [12] |
杨昭颖, 韩灵怡, 郑向向, 等. 基于卷积神经网络的遥感影像及DEM滑坡识别——以黄土滑坡为例[J]. 自然资源遥感, 2022, 34(2):224-230.doi:10.6046/zrzyyg.2021204.
|
| [13] |
|
| [14] |
|
| [15] |
曾超, 曹振宇, 苏凤环, 等. 四川及周边滑坡泥石流灾害高精度航空影像及解译数据集(2008—2020年)[J]. 中国科学数据, 2022, 7(2):195-205.
|
| [16] |
黄震, 朱华波, 陶友瑞. 基于YOLOv5的打磨表面质量快速评价方法[J]. 智能计算机与应用, 2022, 12(11):247-252.
|
| [17] |
宋敏毓, 陈力荣, 梁建安, 等. 轻量化改进网络的实时光纤端面缺陷检测模型[J]. 激光与光电子学进展, 2022, 59(24):2415006.
|
| [18] |
|
| [19] |
|
| [20] |
|
| [21] |
|
/
| 〈 |
|
〉 |