摘要
本文以林业松材线虫病识别为例,对某地开展基于无人机和人工智能技术的松材线虫病异常林木试验和研究。通过采集精度达到0.1 m的可见光航拍数据,自行开发人工智能的语义分割和小目标识别框架,并基于该框架,设计和开发出适用于异常林木的深度学习的连接层和池化层。实验证明,识别准确率可以达到90%以上,速度达到人工的70倍。整体方案技术稳定可行,能够根本性解决异常林木快速识别问题。
In this paper, a case experimental study was made on technologies for identification of pinewood nematodiasis in abnormal forest trees at a sample plot. By collecting data from visible aerial photograph with 0.1-m resolution, a semantic segmentation and smallobject identification framework of artificial intelligence was independently developed. Based on the framework, the in-depth-learning connecting layer and pooling layer suitable for identifying abnormal forest trees were designed and developed. The experiment proved that the identification accuracy could reach above 90% and the identification speed could be 70 times faster than that of the artificial identification. The overall plan was technologically stable and workable and could be a fundamental solution to rapid identification of abnormal forest trees.
引文
[1]郝燕湘.松材线虫病工程治理及对策[J].中国林业,2000(5):14.
[2]李祥康,黄焕华,黄咏槐,等.松材线虫病发病过程特征变化的研究[J].广东林业科技,2010(1):92-96.
[3]马菁.基于遥感技术的松材线虫病早期预警[D].北京:北京林业大学,2012.
[4]申卫星.泰山松材线虫病、美国白蛾风险分析评估与预警技术研究[D].山东:山东农业大学,2012.
[5]石雷.基于3S技术的云南省松材线虫病风险评估研究[D].中国林业科学研究院,2008.
[6]Kaiming He,Xiangyu Zhang,Shaoqing Ren et al.Deep Residual Learning for Image Recognition.Microsoft Research.
[7]Jonathan Huang,Vivek Rathod,Chen Sun et al.Speed/accuracy trade-offs for modern convolutional object detectors.Google Research.
[8]Tsung-Yi Lin,Piotr Dollar,Ross Girshick et al.Feature Pyramid Networks for Object Detection.Facebook AI Research(FAIR),Cornell University and Cornell Tech.
[9]Shaoqing Ren,Kaiming He,Ross Girshick et al.Faster R-CNN:Towards Real-Time Object Detection with Region Proposal Networks.Microsoft Research.
[10]Kaiming He,Georgia Gkioxari,Piotr Dollar et al.Mask R-CNN.Facebook AI Research(FAIR).
[11]Matthew D.Zeiler,Dilip Krishnan,Graham W.Taylor and Rob Fergus Dept.of Computer Science.Deconvolutional Networks.Courant Institute,New York University.
[12]Ross Girshick,Jeff Donahue,Trevor Darrell,Jitendra Malik Rich feature hierarchies for accurate object detection and semantic segmentation.
[13]Wei Liu,Dragomir Anguelov,Dumitru Erhan et al.SSD:Single Shot MultiBox Detector.
[14]Joseph Redmon,Santosh Divvala,Ross Girshick,Ali Farhadi.You Only Look Once:Unified,Real-Time Object Detection.
[15]Jifeng Dai,Yi Li,Kaiming He,Jian Sun R-FCN:Object Detection via Region-based Fully Convolutional Networks.