摘要
白喉乌头(Aconitum leucostomum)是伊犁的主要毒害草之一。快速精准识别白喉乌头可为监测其分布和扩散提供必要的数据和技术支持。本研究利用无人机影像获取白喉乌头影像并构建数据集,分别采用深度残差网络ResNet50和ResNet101提取特征,应用深度学习目标检测算法Mask-RCNN对白喉乌头自动识别。结果表明,通过对比测试集的检测精度,ResNet50的平均精确度(mAP)最高,达到66.0%,ResNet101的mAP最低,为65.3%,ResNet50网络在检测白喉乌头的性能表现优异。Mask-RCNN应用于无人机影像识别白喉乌头,实现白喉乌头的自动检测切实可行。
Aconitum leucostomum was one of the main poisonous grasses in Yili. Rapid and accurate identification of Aconitum leucostomum could provide necessary data and technical support for monitoring its distribution and diffusion. In this study, the images of Aconitum leucostomum were obtained by UAV images and the data set was constructed. The deep residual networks ResNet50 and ResNet101 were used to extract the features, and the deep learning object detection algorithm Mask-RCNN was used to automatically recognize Aconitum leucostomum. The results showed that by comparing the detection accuracy of the test set, the mean average precision(mAP) of ResNet50 was the highest(66.0%), and that of ResNet101 was the lowest(65.3%). ResNet50 network had excellent performance in detecting Aconitum leucostomum. Mask-RCNN was applied to recognize Aconitum leucostomum in UAV image, and it is feasible to realize the automatic detection of Aconitum leucostomum.
作者
梁俊欢
董峦
孙宗玖
马海燕
艾尼玩·艾买尔
阿仁
魏鹏
田聪
阿斯娅·曼力克
郑逢令
LIANG Jun-huan;DONG Luan;SUN Zong-jiu;MA Hai-yan;AINIWAN·Aimaier;Aren;WEI Peng;TIAN Cong;ASIYA·Manlike;ZHENG Feng-ling(College of Grassland Science,Xinjiang Agricultural University,Urumqi 830052,China;Institute of Grassland Research,Xinjiang Academy of Animal Science,Urumqi830057,China;Field Orientation Observation and Research Station of Grassland Ecological Environment on the Northern Slope of Tianshan Mountains,Xinjiang Academy of Animal Science,Urumqi830057,China;College of Computer and Information Engineering,Xinjiang Agricultural University,Urumqi 830052,China)
出处
《草食家畜》
2023年第1期44-51,共8页
Grass-Feeding Livestock
基金
国家自然科学基金项目(31860679)。
