摘要
卷积神经网络是一种特殊的人工神经网络,通过卷积核遍历图获取更多的目标特征信息。近年来,遥感技术发展迅速并被广泛应用于土地利用分析、作物分类识别、作物生长监测和病虫害检测,卷积神经网络为提取农业遥感图像的有效信息提供了新方法。卷积神经语义分割网络可根据语义信息对遥感图像像素点进行标注分割,在计算机计算能力逐步提高的前提下,分割网络结构优化,深度加深,分割准确率提高,性能提升。针对实际需要网络侧重于模块化设计改进,在土地利用分析和农作物分类识别应用中,改进的网络分割边缘细化、清晰,且像素准确率较高。在作物生长监测和病虫害识别方面,模块化改进使网络可高效实现分割任务、满足实际需要,卷积神经网络对农业遥感图像信息的语义分割为农业现代化和精细化管理提供了信息支撑。
Convolution neural network is a special kind of artificial neural network,through convolution kernels traversal graph for characteristic information.In recent years,remote sensing technology has developed rapidly and been widely applied in the analysis of land use,crop classification recognition,crop growth monitoring and insect pests detection,convolution neural network provides a new method of extracting effective information from agricultural remote sensing images.The convolutional neural semantic segmentation network can annotate and segment remote sensing image pixels according to semantic information.Under the premise of gradual improvement of computer computing ability,the structure of the segmentation network is optimized,the depth is deepened,the segmentation accuracy is improved,and the performance is improved.In accordance with the need of actual network focus on the improvement of modular design,the analysis of land use and crop classification recognition applications,the improved network segmentation edge is thinning and clear,and pixel accuracy is higher.In terms of crop growth monitoring and plant diseases and insect pests identification,improvement allows modular network can efficiently implement segmentation task and meet the actual need,and the semantic segmentation of agricultural remote sensing image information by convolutional neural network provides information support for agricultural modernization and fine management.
作者
徐乐园
毛克彪
郭中华
葛非凡
赵瑞
Xu Leyuan;Mao Kebiao;Guo Zhonghua;Ge Feifan;Zhao Rui(School of Physics and Electronic-Engineering,Ningxia University,Yinchuan 750021,Ningxia;Institute of Agricultural Resources and Regional Planning,Chinese Academy of Agricultural Sciences,Beijing 100081;Jiaxing Meteorological Bureau of Zhejiang Province,Jiaxing 314001,Zhejiang)
出处
《农业展望》
2024年第2期70-75,共6页
Agricultural Outlook
基金
2021年宁夏自治区科技创新团队柔性引进人才项目“‘北斗+’土壤水分和植被含水量监测仪器设备研发及应用创新团队”(2021RXTDLX14)
风云卫星应用先行计划资助(FY-APP-2022.0205)
中央级公益性科研院所基本科研业务费专项(1610132020014)。
关键词
卷积神经网络
语义分割
遥感
像素准确率
convolutional neural network
semantic segmentation
remote sensing
pixel accuracy
