无人机影像的松材线虫病半监督学习检测方法
Semi-Supervised Learning for Pine Wilt Disease Detection in UAV Images
-
摘要: 松材线虫病是对森林具有破坏性威胁的全球性疾病，每年在中国造成了极大的生态和经济损失，快速准确地监测和绘制松树的感染状况对于控制此类病虫害传播至关重要。为解决在少量有病害标签数据下用大量无标签数据提高模型精确率的问题，本文提出了一种基于深度学习的半监督目标检测方法，联合已有的标记数据和非标记数据训练YOLOv5目标检测模型，利用大范围无人机快拼图像快速自动地识别定位出感染松材线虫病的单株变色木。实验结果表明，半监督深度学习和无人机遥感结合能有效识别出疫木，精确率可达到85%以上，漏检率为9%，经过实际业务数据盲盒验证，算法指标满足大面积松材线虫病疫情快速动态监测要求。Abstract: Objectives: Pine Wilt Disease (PWD) is a global disease with devastating threat to forests, causing great ecological and economic losses in China every year. It is important to monitor and map the infection status of pine trees accurately for industry sectors to control the spread of such diseases timely. Traditional PWD detection methods such as manual inspection and visual interpretation of remote sensing images are inefficient. Recently, deep learning algorithms have applied in PWD detection task, bringing significant performance gains. However, the current methods usually adopt supervised learning strategy, which requires lots of professional labeling that is time-consuming and labor-intensive. To this end, we proposed a semi-supervised PWD detection method in a teacher-student mechanism. Methods: First, we collected 16310 images in Wudu River Region in Yichang City, Hubei Province, labeled 697 images for model training, and utilized unlabeled 15613 images for testing; then, the YOLOv5 model was adopted as a student model to learn features of input images, the learnt features were updated by exponential moving average algorithm to build a teacher model for generating pseudo-labels of unlabeled test images. The test images with pseudo-labels were mixed with train dataset and fed into the student model to optimize features and generated a stronger teacher model. With the corporation of student and teacher model, the network was more capable to learn robust features for detecting PWD in UAV images. Results: Extensive experiments on the dataset demonstrated that our semi-supervised method could effectively detect the diseased wood, achieving a promising performance with precision of 85.05% and recall of 91.36% on the test dataset, the ground truth of which was provided by third-party professionals. Conclusions: This paper proposed a semi-supervised object detection method using a small portion of labeled data to robustly learn features in large scale UAV images for high-quality PWD detection. The performance on test dataset showed that our method could be well applied to the application of rapid dynamic monitoring of PWD epidemics in forestry field, which could benefit for the sustainable development of forest ecosystem.
-
Keywords:
- Pine Wilt Disease /
- semi-supervised learning /
- deep learning /
- disease detection /
- UAV Images /
- GIS Application
-
-
[1] 汪晨,张辉辉,乐继旺,赵帅.基于深度学习和遥感影像的松材线虫病疫松树目标检测[J].南京师大学报(自然科学版),2021,44(03):84-89. Wang Chen,Zhang Huihui,Le Jiwang,et al. Object Detection to the Pine Trees Affected by Pine Wilt Disease in Remote Sensing Images Using Deep Learning[J].Journal of Nanjing Normal University(Natural Science Edition), 2021,44(03):84-89.
[2] Shin S C. Pine wilt disease in Korea[M]//Pine wilt disease. Springer, Tokyo, 2008:26-32.
[3] Mamiya Y. History of pine wilt disease in Japan[J]. Journal of nematology, 1988, 20(2):219.
[4] Zhao B G. Pine wilt disease in China[M]//Pine wilt disease. Springer, Tokyo, 2008:18-25.
[5] Han H, Chung Y J, Shin S C. First report of pine wilt disease on Pinus koraiensis in Korea[J]. Plant disease, 2008, 92(8):1251-1251.
[6] 黄华毅,马晓航,扈丽丽,黄咏槐,黄焕华.Fast R-CNN深度学习和无人机遥感相结合在松材线虫病监测中的初步应用研究[J].环境昆虫学报,2021,43(05):1295-1303. Huang Huayi,Ma Xiaohang,Hu Lili,et al.The preliminary application of the combination of Fast RCNN deep learning and UAV remote sensing in the monitoring of pine wilt disease[J]. Journal of Environmental Entomology, 2021,43(05):1295-1303.
[7] 黄麟,张晓丽,石韧.森林病虫害遥感监测技术研究的现状与问题[J].遥感信息,2006(02):71-75. Huang Lin,Zhang Xiaoli,Shi Ren.Current Status and Problems in Monitoring Forest Damage Caused by Diseases and Insects Based on Remote Sensing[J]. Remote Sensing Information, 2006(02):71-75.
[8] Dennison P E, Brunelle A R, Carter V A. Assessing canopy mortality during a mountain pine beetle outbreak using GeoEye-1 high spatial resolution satellite data[J]. Remote Sensing of Environment, 2010, 114(11):2431-2435.
[9] Johnson B A, Tateishi R, Hoan N T. A hybrid pansharpening approach and multiscale object-based image analysis for mapping diseased pine and oak trees[J]. International journal of remote sensing, 2013, 34(20):6969-6982.
[10] Park J, Sim W, Lee J. Detection of trees with pine wilt disease using object-based classification method[J]. Journal of forest and environmental science, 2016, 32(4):384-391.
[11] 徐伟萌,杨浩,李振洪,程金鹏,林哈特,杨贵军.利用无人机数码影像进行密植型果园单木分割[J/OL].武汉大学学报(信息科学版):1-11[2022-11-04]. Xu Weimeng,Yang Hao,Li Zhenhong,et al.Single Tree Segmentation in Close-Planting Orchard Using UAV Digital Image[J].Geomatics and Information Science of Wuhan University. 1-11[2022-11-04].
[12] Leckie D G, Jay C, Gougeon F A, et al. Detection and assessment of trees with Phellinus weirii (laminated root rot) using high resolution multi-spectral imagery[J]. International Journal of Remote Sensing, 2004, 25(4):793-818.
[13] Yu R, Luo Y, Zhou Q, et al. A machine learning algorithm to detect pine wilt disease using UAV-based hyperspectral imagery and LiDAR data at the tree level[J]. International Journal of Applied Earth Observation and Geoinformation, 2021, 101:102363.
[14] Hoshikawa T, Yamamoto K. Individual tree detection and classification for mapping pine wilt disease using multispectral and visible color imagery acquired from unmanned aerial vehicle[J]. Journal of The Remote Sensing Society of Japan, 2020, 40(1):13-19.
[15] Deng X, Tong Z, Lan Y, et al. Detection and location of dead trees with pine wilt disease based on deep learning and UAV remote sensing[J]. AgriEngineering, 2020, 2(2):294-307.
[16] Syifa M, Park S J, Lee C W. Detection of the pine wilt disease tree candidates for drone remote sensing using artificial intelligence techniques[J]. Engineering, 2020, 6(8):919-926.
[17] Zhang R, Xia L, Chen L, et al. Recognition of wilt wood caused by pine wilt nematode based on U-Net network and unmanned aerial vehicle images[J]. Trans. Chin. Soc. Agricult. Eng., 2020, 36(12):61-68.
[18] Huang R, Noble J A, Namburete A I L. Omnisupervised learning:scaling up to large unlabelled medical datasets[C]//International Conference on Medical Image Computing and Computer-Assisted Intervention. Springer, Cham, 2018:572-580.
[19] 周培诚,程塨,姚西文,韩军伟.高分辨率遥感影像解译中的机器学习范式[J].遥感学报,2021,25(01):182-197. Zhou Peicheng,Cheng Gong,Yao Xiwen et,al. Machine learning paradigms in high-resolution remote sensing image interpretation. National Remote Sensing Bulletin,25(1):182-197.
[20] 韩彦岭, 李鹏, 张云, 等. 主动学习与半监督技术相结合的海冰图像分类[J]. 遥感信息, 2019, 34(2):15-22. Han Yanling,Li Peng,Zhang Yun.Combining Activate Learning with Semi-supervised Learning for Sea Ice Image Classification[J].Remote Sensing Information,2019,34(2):15-22.
[21] 熊彪,江万寿,李乐林.基于高斯混合模型的遥感影像半监 督分 类[J]. 武汉 大学 学报(信息 科学 版),2011,36(01):108-112. Xiong Biao,JIANG Wanshou,Li Lelin.Gauss Mixture Model Based Semi Supervised Classification for Remote Sensing Image[J]. Geomatics and Information Science of Wuhan University, 2011,36(01):108-112.
[22] Ren S, He K, Girshick R, et al. Faster r-cnn:Towards real-time object detection with region proposal networks[J]. Advances in neural information processing systems, 2015, 28.
[23] Redmon J, Divvala S, Girshick R, et al. You only look once:Unified, real-time object detection[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2016:779- 788.
[24] Tarvainen A, Valpola H. Mean teachers are better role models:Weight-averaged consistency targets improve semi-supervised deep learning results[J]. Advances in neural information processing systems, 2017, 30.
[25] Kim M I, Lee W K, Kwon T H, et al. Early detecting damaged trees by pine wilt disease using DI (Detection Index) from portable near infrared camera[J]. Journal of Korean Society of Forest Science, 2011, 100(3):374-381.
计量
- 文章访问数: 558
- HTML全文浏览量: 62
- PDF下载量: 86
下载:
