摘要
为解决背景噪声干扰下,从微根管采集的原位根系图像中难以直接提取准确的表型参数问题,提出一种基于改进U-Net的微根管根系表型参数测量系统。在U-Net网络中引入优化后的空洞空间金字塔池化模块(Atrous Spatial Pyramid Pooling,ASPP)和高效通道注意力模块(Efficient Channel Attention,ECA),增大感受野,提升模型捕捉根系细节特征的能力,获取精确的根系分割图像。结果表明,改进的U-Net模型平均交并比和平均像素精度分别为87.07%和91.85%,相较原始U-Net分别提高了2.49%和2.3%。与WinRHIZO根系分析软件测量值相比,根长度和面积决定系数分别为0.951 8和0.984 9,Spearman相关系数分别为0.972 5和0.975 7,可以实现根系长度和面积的准确测量。
To address the challenge of accurately extracting phenotypic parameters from in situ root images collected from minirhizotrons amidst background noise interference,this paper proposes a minirhizotron root phenotypic parameter measurement system based on an improved U-Net model.In the U-Net network,optimized ASPP(Atrous Spatial Pyramid Pooling)and ECA(Efficient Channel Attention)modules are employed to increase the receptive field and enhance the ability to capture detailed features,thereby obtaining precise segmentation images.The experimental results show that the mean intersection over union and mean pixel accuracy of the improved U-Net model are 87.07%and 91.85%,which are 2.49%and 2.3%higher compared to the original U-Net,respectively.Comparing with measurements obtained using WinRHIZO root analysis software,the determination coefficients for the root length and area are 0.9518 and 0.9849.respectively.The Spearman correlation coefficients are 0.9725 for the root length and 0.9757 for root area.This indicates the system′s capability to accurately measure the root length and area.
作者
赵亚凤
刘晓璐
王冬冬
王孟雪
宋文华
胡峻峰
ZHAO Yafeng;LIU Xiaolu;WANG Dongdong;WANG Mengxue;SONG Wenhua;HU Junfeng(College of Computer and Control Engineering,Northeast Forestry University,Harbin 150040,China)
出处
《森林工程》
北大核心
2024年第4期127-136,共10页
Forest Engineering
基金
国家自然科学基金项目(32371864)。
关键词
根系表型
微根管
图像分割
参数测量
U-Net
Root phenotype
minirhizotron
image segmentation
parameter measurement
U-Net
