摘要
为有效解决玉米雄穗无人机识别过程中因雄穗尺寸过小造成的漏检、识别速度慢、模型体积大等问题,通过添加注意力机制、融入轻量化模块和优化损失函数,建立了一种改进YOLOv8n卷积神经网络的玉米雄穗检测模型YOLOv8n-maize。结果表明:改进后的模型在测试集上的平均精度均值(Mean average precision,m AP)达97.8%,比原模型提高了2.6%;模型计算量(Floating point operations,FLOPs)减少了15.8%,参数量(Parameters,Params)体积缩小了17.6%。这种高精度、小体积模型能够满足玉米雄穗快速识别的需求,可为无人机机载平台的部署提供关键技术支持。
To effectively address the issues of missed detection,slow recognition speed,and large model size in the unmanned aerial vehicle(UAV) identification of maize tassels due to their small size,an improved maize tassel detection model,YOLOv8n convolutional neural network,was developed by incorporating an attention mechanism,integrating lightweight modules,and optimizing the loss function.The results showed that the average precision(m AP) of the improved model on the test set was 97.8%,which was 2.6% higher than that of the original model,the amount of model computation(Floating point operations,FLOPs) was reduced by 15.8%,and the volume of parameters(Params) was reduced by 17.6%.This high-precision,compact model is capable of meeting the requirements for rapid maize tassel identification and provides crucial technical support for the deployment of UAV-borne platforms.
作者
胡冬
班松涛
马超
田明璐
袁涛
李琳一
庄洁
HU Dong;BAN Songtao;MA Chao;TIAN Minglu;YUAN Tao;LI Linyi;ZHUANG Jie(Institute of Agricultural Science and Technology Information,Shanghai Academy of Agricultural Sciences,Shanghai 201403,China;Key Laboratory of Intelligent Agricultural technology(Yangtze River Delta),Ministry of Agriculture and Rural Affairs,Shanghai 201403,China;Rural Development Promotion Center of Jiading District Shanghai,Shanghai 201800,China)
出处
《上海农业学报》
2025年第1期102-107,共6页
Acta Agriculturae Shanghai
基金
上海市科技兴农技术创新项目[沪农科创字(2022)第4-1号]
上海市农业科学院卓越团队建设项目[沪农科卓(2022)015]。
