College of Mechanical and Electrical Engineering,Northeast Forestry University,Harbin 150040,China
Show less
文章历史+
Received
Published
2025-03-05
2025-09-15
Issue Date
2025-10-30
PDF (8721K)
摘要
精准高效的森林树种识别是智慧林业实现的关键。传统的地面调查方法存在效率低、成本高等问题,而基于机器学习的树种识别方法通常依赖于大量的特征提取和先验知识。为解决这些问题,提出一种基于改进YOLOv10的无人机影像树种识别算法,通过轻量化网络设计和注意力机制增强,实现边缘设备的高效部署,为森林资源数字化管理提供技术支持。选取东北地区5种常见的树种落叶松(Larix gmelinii)、黄檗(Phellodendron amurense)、胡桃楸(Juglans mandshurica)、榆树(Ulmus pumila)和水曲柳(Fraxinus mandshurica)构建无人机影像数据集,通过采用轻量化卷积(Ghost)重构主干网络以降低计算复杂度。此外,在特征融合层引入卷积块注意力模块(Convolutional Block Attention Module,CBAM),通过通道和空间双维度的特征校准增强细粒度特征提取能力。通过双向跨尺度连接(BiFPN)优化多尺度特征融合,采用对称交并比损失函数(Symmetric Intersection over Union Loss,SIoU)改进边界框回归速度。最后于Jetson Nano嵌入式平台进行模型部署验证。改进后的YOLOv10模型在验证集上达到91.5%的查准率(Precision)和77.5%的mAP@0.5,分别较基线模型提升4.5%和3.8%。部署实测平均推理速度为43.5 FPS,较基线模型提升35.5%,mAP@0.5达75.7%。结果表明,YOLOv10改进算法通过轻量化架构和优化多尺度特征提取,在保持实时性的同时提升复杂林区场景的树种识别精度。该算法为无人机林业调查提供可嵌入式部署的解决方案,特别适用于林冠层重叠度高、光照条件多变的实际作业环境。
Abstract
Efficient and accurate tree species identification is critical for the realization of smart forestry. Traditional field survey methods are low efficiency and high cost, while machine learning-based tree species identification approaches often rely on extensive feature extraction and prior knowledge. To address these issues, a tree species identification algorithm based on improved YOLOv10 for UAV imagery is proposed in this paper. The improved architecture integrates lightweight network design and attention mechanisms to enable efficient edge device deployment, providing technical support for digital forest resource management. A UAV imagery dataset was developed for five common tree species (Larix gmelinii, Phellodendron amurense, Juglans mandshurica, Ulmus pumila, and Fraxinus mandshurica) in Northeast China. The backbone network was reconstructed using lightweight convolution (Ghost) for computational complexity reduction. The convolutional block attention module (CBAM) was introduced in the fusion layer to strengthen fine-grained feature extraction through channel and spatial dual dimensional feature calibration. Multi-scale feature fusion was optimized through bidirectional cross-scale connections (BiFPN), while bounding box regression efficiency was improved using a structured intersection over union (SIoU) loss function. Final deployment validation was conducted on the Jetson Nano embedded platform. The improved YOLOv10 model achieved 91.5% precision and 77.5% mAP@0.5 on the validation set, showing improvements of 4.5% and 3.8% compared to the baseline model, respectively. In practical deployment, the model achieved an inference speed of 43.5 FPS, 35.5% faster than the baseline model, with mAP@0.5 of 75.7%. Results showed that, the improved YOLOv10 algorithm successfully balances identification accuracy and real-time performance in complex forest environments through lightweight architecture and multi-scale feature optimization. The solution demonstrates particular effectiveness in scenarios with dense canopy overlap and variable illumination, offering an embeddable solution for UAV forestry surveys.
在遥感数据类型中,高光谱影像虽然可以提供更丰富的光谱信息,但需要进行数据降维和波段优选等预处理,过程较为繁琐;激光雷达虽然能提供精确的三维结构信息,但受限于高昂的设备和处理成本[4-5]。相反,无人机影像(red green blue,RGB)仅需搭载常规的数字相机即可获得,并可利用现有的技术手段进行快速处理,使其成为树种分类中较为便捷、高效的遥感数据类型。目前,树种识别大多数依赖于机器学习算法,周湘山等[6]运用支持向量机进行树种识别,该方法的整体精度为60%。王俊杰等[7]运用随机森林进行无人机多光谱树种识别,总体精度为70.83%。姚扬[8]运用随机森林进行多源遥感数据树种识别,精度达到了77.96%。虽然取得了较高的识别精度,但需要具备大量的先验知识。
WUX Q, QINX L, ZHOUR L,et al.Progress of study on forest cover change detection by using remote sensing technique[J].Forest and Grassland Resources Research,2010(4):82-87.
CHENL W, ZHOUX C, LIC X,et al.Classification of tree species based on UNet-ResNet14 semi-supervised learning using UAV images[J].Transactions of the Chinese Society of Agricultural Engineering,2024,40(1):217-226.
[7]
吴晓明.改进的混合空洞卷积神经网络高光谱影像树种分类算法[D].廊坊:北华航天工业学院,2023.
[8]
WUX M.Improved hybrid hollow convolutional neural network for tree species classification in hyperspectral images[D].Langfang:North China Institute of Aerospace Engineering,2023.
ZHOUX S, PIAOH Y, ZHOUJ,et al.Research on classification method of drone hyperspectral tree species based on machine learning algorithms[J].Energy and Environment,2023(4):134-137.
WANGJ J, ZHANGS Y, TENGP C.Forest tree species identification based on UAV multispectral and HMLS[J].Journal of Heilongjiang University of Science & Technology,2023,33(5):774-778.
[15]
姚扬.基于多源遥感数据的城市植被优势树种分类[D].昆明:云南大学,2022.
[16]
YAOY.Classification of dominant tree species of urban vegetation based on multi-source remote sensing data[D].Kunming:Yunnan University,2022.
[17]
NEZAMIS, KHORAMSHAHIE, NEVALAINENO,et al.Tree species classification of drone hyperspectral and RGB imagery with deep learning convolutional neural networks[J].Remote Sensing,2020,12(7):1070.
[18]
KRIZHEVSKYA, SUTSKEVERI, HINTONG.ImageNet classification with deep convolutional neural networks[C]//Advances in Neural Information Processing Systems 25 (NIPS 2012.Curran Associates Inc.,2012.
[19]
HEK, ZHANGX, RENS,et al.Deep residual learning for image recognition[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition(CVPR).Las Vegas,NV,USA,June 27-30,2016.IEEE,2016:770-778.
[20]
HINTONG E, SALAKHUTDINOVR R.Reducing the dimensionality of data with neural networks[J].Science,2006,313(5786):504-507.
[21]
KNAUERU, VON REKOWSKIC S, STECKLINAM,et al.Tree species classification based on hybrid ensembles of a convolutional neural network (CNN) and random forest classifiers[J].Remote Sensing,2019,11(23):2788.
ZHOUZ X.Identification of urban tree species in UAV RGB images based on FC-DenseNet[D].Changsha:Central South University of Forestry and Technology,2023.
[24]
CHENC, JINGL, LIH,et al.Individual tree species identification based on a combination of deep learning and traditional features[J].Remote Sensing,2023,15(9):2301.
[25]
REDMONJ, DIVVALAS, GIRSHICKR,et al.You only look once:Unified,real-time object detection[C]//2016 IEEE Comference on Computer Vision and Pattern Recognition (CVPR).Las Vegas,NV,USA,June 27-30,2016.IEEE,2016:779-788.
[26]
GIRSHICKR.Fast R-CNN[C]//2015 IEEE International Conference on Computer Vision (ICCV),Santiago,Chile,December 07-13,2015.IEEE,2015:1440-1448.
RENS, HEK, GIRSHICKR,et al.Faster R-CNN:Towards real-time object detection with region proposal networks[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2017,39(6):1137-1149.
[29]
XUS C, WANGR R, SHIW,et al.Classification of tree species in transmission line corridors based on YOLO v7[J].Forests,2023,15(1):61.
HUANGL M, WANGY X, XUQ,et al.Recognition of abnormally discolored trees caused by pine wilt disease using YOLO algorithm and UAV images[J].Transactions of the Chinese Society of Agricultural Engineering,2021,37(14):197-203.
[32]
HANY H, DUANB C, GUANR X,et al.LUFFD-YOLO:A lightweight model for UAV remote sensing forest fire detection based on attention mechanism and multi-level feature fusion[J].Remote Sensing,2024,16(12):2177.
WANGY Y, MENGL, HAND H,et al.Species diversity in urban forest-A case study of City Forestry Demonstration Base in Harbin[J].Journal of Northeast Forestry University,2017,45(3):34-38.
[36]
HANK, WANGY, TIANQ,et al.GhostNet:More features from cheap operations[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).Seattle,WA,USA,June 13-19,2020.IEEE,2020:1577-1586.
YANGK, FANX J, BOW H,et al.Plant disease and pest detection based on visual attention enhancement[J].Journal of Nanjing Forestry University (Natural Sciences Edition),2023,47(3):11-18.
SUNF G, WANGY L, LANP,et al.Identification of apple fruit diseases using improved YOLOv5s and transfer learning[J].Transactions of the Chinese Society of Agricultural Engineering,2022,38(11):171-179.