基于自旋式Unet++网络的小麦条锈病菌夏孢子自动检测方法  

Automatic detection of urediniospores method for wheat stripe rust pathogen based on spin Unet++network

作　　者：周晋兵 陈鹏 雷雨 黄林生 赵晋陵 梁栋 ZHOU Jinbing;CHEN Peng;LEI Yu;HUANG Linsheng;ZHAO Jinling;LIANG Dong(School of Internet/National Engineering Research Center for Agro-Ecological Big Data Analysis&Application,Anhui University,Hefei 230601,China)

机构地区：[1]安徽大学互联网学院/农业生态大数据分析与应用技术国家地方联合工程研究中心,安徽合肥230601

出　　处：《南京农业大学学报》2023年第5期1004-1010,共7页Journal of Nanjing Agricultural University

基　　金：国家自然科学基金项目(62273001,62072002);安徽省科技重大专项(202003a06020016);安徽省高等学校科学研究项目(2022AH050085);农业生态大数据分析与应用技术国家地方联合工程研究中心开放课题项目(AE2018009)。

摘　　要：[目的]条锈病菌夏孢子是引起小麦条锈病暴发的重要原因。针对夏孢子显微图像识别中存在的检测精度不高和分割精度较低的问题,提出了一种基于自旋式Unet++网络的小麦条锈病菌夏孢子自动检测算法。[方法]针对显微图像中夏孢子目标密集,易粘连等特点,通过改进Unet++特征提取网络,将原始网络的输出反向连接至输入端形成一种自旋结构,并增加超参数K控制自旋次数,以达到迭代训练的目的,优化Unet++网络的分割精度,提高检测的分割率;根据孢子的形态特征,提出一种加权映射矩形计算公式,计算其矩形热力图,以提高孢子的检测精确率。[结果]算法测试结果表明,改进算法的平均精确率达到99.03%,分割率为86.45%,较原始CenterNet模型分割率提高了10.35百分点,精确率提高0.46百分点,占用内存降低66.09%。[结论]该模型在保证较高精度检测夏孢子目标的同时,保证了较高的孢子分割率,为小麦条锈病早期预警提供了一种有效的方法。[Objectives]The urediniospores of wheat stripe rust pathogen is an important reason for the outbreak of wheat stripe rust.Aiming at the problems of low detection precision and low segmentation precision in the identification of microscopic image of urediniospores,an automatic detection method for urediniospores of wheat stripe rust pathogen based on spin Unet++network was proposed.[Methods]According to the characteristics of dense and conglutination of the urediniospores in the microscopic image,a spinning connection structure was proposed by improving the feature extraction network of Unet++,which optimized the segmentation precision of Unet++network and improved the segmentation rate of detection.According to the morphological characteristics of spores,a weighted mapping rectangle calculation formula was proposed to calculate its rectangular thermogram,so as to improve the precision of spore detection.[Results]The algorithm test results showed that the average precision of the improved algorithm reached 99.03%,and the segmentation rate was 86.45%,which met the requirements of accurate detection.Compared with the original CenterNet model,the overlap rate increased by 10.35 percentage points,the precision increased by 0.46 percentage points,and the memory usage reduced by 66.09%.[Conclusions]The model in this paper not only ensured a high precision of detection of urediniospores,but also ensured a high spore segmentation rate,which provided an effective method for the early warning of wheat stripe rust.

关 键 词：小麦条锈病 真菌夏孢子 改进Unet++ 图像分割 椭圆框 

分 类 号：S435.121.4[农业科学—农业昆虫与害虫防治]