摘要
草原植物花朵计数可以帮助我们了解草原植物的生长状况、繁殖能力、群落结构等信息,通过人工计数方法获取草原上不同物种花朵丰度是一个费时费力的过程。本研究基于深度学习目标检测方法,在鄂尔多斯荒漠草原上10个样地和50个样方上开展检测模型训练、评估和应用。从YOLOv7的3个模型整体表现来看,YOLOv7-E6E的F1-sorce和mAP@0.5均可达到0.7以上,具有较高的识别精度。从YOLOv7的3个模型在5种花朵检测的表现来看,YOLOv7-X、YOLOv7-E6E模型在北芸香、蒙古韭、细叶韭的检测上mAP@0.5高于0.8,而3个模型中仅有YOLOv7-E6E在蒺藜、兔唇花的mAP@0.5超过0.6。从模型在50个样方的花朵计数应用来看,YOLOv7-E6E模型花朵计数的总体正确率为0.91,能满足这5种草原开花植物检测和计数的需要。综上所述,通过深度学习花朵快速计数可以提高样方尺度花期植物调查效率,但为满足大规模物种调查和计数的任务需求,仍需扩大样本量和不断改进模型结构,以提高模型植物花朵检测的整体性能。
Desert steppe plant flower counting can help us understand the growth status,reproductive capacity,community structure and other information of plants.It is a time-consuming and labor-intensive process to obtain the flower abundance of different species in grassland by manual counting.Based on the deep learning target detection method,this study trained detection model,and then evaluated and applicated on 10 plots with 50 quadrats in the desert steppe.From the overall performance of the three models of YOLOv7,the results of YOLOv7-E6E model were very good for some types of flowers,with F1-sorce and mAP@0.5 higher than 0.7.The performance of the 3 models of YOLOv7 in the detection of 5 kinds of flowers,the mAP@0.5 of the YOLOv7-X and YOLOv7-E6E models is higher than 0.8 in the detection of Haplophyllum dauricum,Allium mongolicum,and Allium tenuissimum,while only the mAP@0.5 of YOLOv7-E6E in Tribulus terrestris and Lagochilus ilicifolius exceeds 0.6.Application of the model in flower counting of 50 quadrats,the overall accuracy rate of flower counting of the YOLOv7-E6E model is 0.91,which can meet the needs of detecting and counting these 5 grassland flowering plants.To sum up,the rapid counting of flowers by deep learning can improve the survey efficiency of flowering plants at the quadrat scale.However,in order to meet the task requirements of large-scale species survey and counting,it is still necessary to expand the sample size and continuously improve the model structure to improve the overall performance of flower detection in model plants.
作者
王永财
万华伟
高吉喜
胡卓玮
WANG Yong-cai;WAN Hua-wei;GAO Ji-xi;HU Zhuo-wei(College of Resource Environment and Tourism,Capital Normal University,Beijing 100048,China;Ministry of Ecology and Environment Center for Satellite Application on Ecology and Environment,Beijing 100094,China)
出处
《环境生态学》
2024年第2期1-8,共8页
Environmental Ecology
基金
国家重点研发计划项目(2021YFB3901102)资助。
关键词
深度学习
目标检测
花朵计数
草地植物
Deep learning
object detection
flower counting
grassland plant
