基于改进的PSPNet网络的毛白杨根系自动分割量化系统

doi:10.3969/j.issn.1004-1524.20230071

浙江农业学报 ›› 2024, Vol. 36 ›› Issue (2): 424-431.DOI: 10.3969/j.issn.1004-1524.20230071

基于改进的PSPNet网络的毛白杨根系自动分割量化系统

张鹏翀¹^,²^,³(), 韩巧玲¹^,²^,³^,⁴, 席本野⁵, 郑秋燕¹^,²^,³, 赵玥¹^,²^,³^,⁴^,^*()

1.北京林业大学工学院,北京 100083
2.城乡生态环境北京实验室,北京 100083
3.智慧林业研究中心,北京 100083
4.国家林业和草原局林业装备与自动化国家重点实验室,北京 100083
5.北京林业大学省部共建森林培育与保护教育部重点实验室,北京 100083

收稿日期:2023-01-16 出版日期:2024-02-25 发布日期:2024-03-05
作者简介:张鹏翀(1998—),男,山西忻州人,硕士,研究方向为图像处理。E-mail:zhangpengchong163@163.com
通讯作者: *赵玥,E-mail:zhaoyue0609@126.com
基金资助:
国家自然科学基金(32071838);国家自然科学基金青年科学基金(32101590);中国博士后科学基金(2022T150055);北京市共建项目

Automatic segmentation and quantification system for Populus tomentosa roots based on improved PSPNet

ZHANG Pengchong¹^,²^,³(), HAN Qiaoling¹^,²^,³^,⁴, XI Benye⁵, ZHENG Qiuyan¹^,²^,³, ZHAO Yue¹^,²^,³^,⁴^,^*()

1. School of Technology, Beijing Forestry University, Beijing 100083, China
2. Beijing Lab of Urban and Rural Ecological Environment, Beijing 100083, China
3. Research Center for Intelligent Forestry, Beijing 100083, China
4. Key Lab of National Forestry and Grassland Administration for Forestry Equipment and Automation, Beijing 100083, China
5. Key Laboratory for Silviculture and Conservation of Ministry of Education, Beijing Forestry University, Beijing 100083, China

Received:2023-01-16 Online:2024-02-25 Published:2024-03-05

摘要/Abstract

摘要：

针对毛白杨微根管图像中根系颜色不均匀、形态不一致、目标背景差异小的特点,以及现有根系图像处理软件无法批量分割的问题,提出一个基于深度学习网络的毛白杨根系自动分割量化系统。该系统包括根系图像自动分割和根系特征量化两个部分。首先,提出以EPSANet50为骨架的PSEPNet网络(pyramid scene efficient parsing network, PSEPNet)实现毛白杨根系图像的自动高精度分割;其次,使用骨架细化法提取根系中心像素轮廓;最后,运用数学统计方法提取根系数量、根系长度等多特征参数,实现对毛白杨根系特征的量化表达。结果表明:PSEPNet网络对毛白杨微根管根系图像具有良好的分割效果,其准确率为0.981 9,召回率为0.884 9,精确率为0.830 9,F1值为0.851 2,能够实现对根系数量、根系长度、根系投影面积等特征的量化,可为基于微根管技术对林木生长规律的研究提供技术支持和数据基础。

关键词: 微根管, 毛白杨根系, 语义分割, 特征量化

Abstract:

In the minirhizotron images of Populus tomentosa roots, the root color is uneven, the root morphology is inconsistent, and there is little difference between the target and the background. Moreover, the existing root image processing software can not segment roots in batches. To solve the above problems, an automatic segmentation and quantification system for Populus tomentosa roots is proposed in the present study. This system includes two parts, namely, automatic root image segmentation part and root feature quantification part. Firstly, the PSEPNet (pyramid scene efficient parsing network) based on EPSANet50 is designed to realize the automatic segmentation of Populus tomentosa roots. Secondly, the skeleton thinning method is adopted to extract the pixel contour of the root center. Finally, mathematical statistics are used to realize the quantitative description of root number, root length and other characteristics. The test results show that the segmentation method used by the system has the best segmentation effect for minirhizotron images of Populus tomentosa roots, as the accuracy rate is 0.981 9, the recall rate is 0.884 9, the accuracy is 0.830 9 and the F1 score is 0.851 2. This system can also realize the quantification of root number, root length, root projection area and other characteristics to provide technical support and theoretical basis for the study of tree growth laws based on minirhizotron technology.

Key words: minirhizotron, Populus tomentosa roots, semantic segmentation, feature quantification

中图分类号:

S718.42

张鹏翀, 韩巧玲, 席本野, 郑秋燕, 赵玥. 基于改进的PSPNet网络的毛白杨根系自动分割量化系统[J]. 浙江农业学报, 2024, 36(2): 424-431.

ZHANG Pengchong, HAN Qiaoling, XI Benye, ZHENG Qiuyan, ZHAO Yue. Automatic segmentation and quantification system for Populus tomentosa roots based on improved PSPNet[J]. Acta Agriculturae Zhejiangensis, 2024, 36(2): 424-431.

图/表 7

图1 采集到的图像原图(上)与裁剪后的图像(下)

Fig.1 Captured original image (up) and image after cropping (down)

图2 创建的PSEPNet的网络结构示意图

Fig.2 Structure diagram of the proposed PSEPNet

图3 毛白杨根系的分割结果示例一

Fig.3 Example 1 of segmentation results of Populus tomentosa roots

图4 毛白杨根系的分割结果示例二

Fig.4 Example 2 of segmentation results of Populus tomentosa roots

表1 不同方法分割效果的定量评估

Table 1 Quantitative evaluation of segmentation result of different methods

分割方法Segmentatin method	准确率Accuracy	召回率Recall	精确率Precision	F1值F1 value
PSPNet_ResNet50	0.971 5±0.033 4	0.840 2±0.111 8	0.731 3±0.138 5	0.771 5±0.110 7
PSPNet_MobileNetV2	0.979 7±0.025 4	0.855 1±0.094 8	0.818 4±0.124 4	0.829 0±0.098 1
PSEPNet	0.981 9±0.024 0	0.884 9±0.084 8	0.830 9±0.115 2	0.851 2±0.087 4

表2 毛白杨根系的特征参数

Table 2 Feature extraction of Populus tomentosa roots

特征参数 Feature parameter	参数值 Parameter value
根系数量Root number	3.66±2.37
根系长度Root length/mm	57.26±38.18
根系周长Root circumference/mm	130.5±79.29
根系平均直径Average diameter of root/mm	0.53±0.19
根系投影面积Projected area of root/mm²	27.81±17.47
根系表面积Surface area of root/mm²	87.36±54.87
根长密度Length density of root/(mm·mm^-3)	0.16±0.10
根表面积密度	0.08±0.05
Surface area density of root/(mm·mm^-3)

图5 毛白杨根系自动分割与量化系统的照片

Fig.5 Photos of the segmentation and quantification system for Populus tomentosa roots images

参考文献 20

[1]	刘悦. 间伐对水曲柳和落叶松混交林细根形态、生物量和根长密度的影响[D]. 哈尔滨: 东北林业大学, 2021.
	LIU Y. Effects of thinning on the morphology, biomass and length density of fine roots in mixed forest of Fraxinus mandshurica and Larix olgensis[D]. Harbin: Northeast Forestry University, 2021. (in Chinese with English abstract)
[2]	PENG Y F, GUO D L, YANG Y H. Global patterns of root dynamics under nitrogen enrichment[J]. Global Ecology and Biogeography, 2017, 26(1): 102-114.
[3]	王延平, 许坛, 朱婉芮, 等. 杨树人工林细根数量和形态特征的季节动态及代际差异[J]. 应用生态学报, 2016, 27(2): 395-402.
	WANG Y P, XU T, ZHU W R, et al. Seasonal dynamics of quantitative and morphological traits of poplar fine roots and their differences between successive rotation plantations[J]. Chinese Journal of Applied Ecology, 2016, 27(2): 395-402. (in Chinese with English abstract)
[4]	姚继周. 水杉人工林细根生产和周转及对氮沉降的响应[D]. 南京: 南京林业大学, 2016.
	YAO J Z. Fine root production and turnover and the response to nitrogen deposition in Metasequoia glyptostroboides plantation[D]. Nanjing: Nanjing Forestry University, 2016. (in Chinese with English abstract)
[5]	DORNBUSH M E, ISENHART T M, RAICH J W. Quantifying fine-root decomposition: an alternative to buried litterbags[J]. Ecology, 2002, 83(11): 2985-2990.
[6]	TAYLOR H M, HUCK M G, KLEPPER B, et al. Measurement of soil-grown roots in a Rhizotron1[J]. Agronomy Journal, 1970, 62(6): 807-809.
[7]	史建伟, 于水强, 于立忠, 等. 微根管在细根研究中的应用[J]. 应用生态学报, 2006, 17(4): 4715-4719.
	SHI J W, YU S Q, YU L Z, et al. Application of minirhizotron in fine root studies[J]. Chinese Journal of Applied Ecology, 2006, 17(4): 4715-4719. (in Chinese with English abstract)
[8]	李燕丽, 王昌昆, 卢碧林, 等. 基于微根管技术的盐胁迫下小麦根系生长原位监测方法[J]. 土壤学报, 2021, 58(3): 599-609.
	LI Y L, WANG C K, LU B L, et al. In-situ monitoring method of wheat root growth under salt stress using minirhizotron technique[J]. Acta Pedologica Sinica, 2021, 58(3): 599-609. (in Chinese with English abstract)
[9]	王传宇, 郭新宇, 温维亮, 等. 基于玉米根系图像的表型指标获取方法[J]. 农业工程学报, 2021, 37(8): 169-176.
	WANG C Y, GUO X Y, WEN W L, et al. Phenotyping index acquisition method based on maize root images[J]. Transactions of the Chinese Society of Agricultural Engineering, 2021, 37(8): 169-176. (in Chinese with English abstract)
[10]	席本野. 杨树根系形态、分布、动态特征及其吸水特性[J]. 北京林业大学学报, 2019, 41(12): 37-49.
	XI B Y. Morphology, distribution, dynamic characteristics of poplar roots and its water uptake habits[J]. Journal of Beijing Forestry University, 2019, 41(12): 37-49. (in Chinese with English abstract)
[11]	刘凯, 李文彬, 赵玥, 等. 基于微根管图像的根系形态特征快速提取技术[J]. 中国水土保持科学(中英文), 2021, 19(4): 129-136.
	LIU K, LI W B, ZHAO Y, et al. Rapid extraction technology of the root morphological characteristics via minirhizotron image[J]. Science of Soil and Water Conservation, 2021, 19(4): 129-136. (in Chinese with English abstract)
[12]	ZHAO H S, SHI J P, QI X J, et al. Pyramid scene parsing network[C]// 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). July 21-26, 2017, Honolulu, HI, USA. IEEE, 2017: 6230-6239.
[13]	HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]// 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). June 27-30, 2016, Las Vegas, NV, USA. IEEE, 2016: 770-778.
[14]	SANDLER M, HOWARD A, ZHU M L, et al. MobileNetV2: inverted residuals and linear bottlenecks[C]// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. June 18-23, 2018, Salt Lake City, UT, USA. IEEE, 2018: 4510-4520.
[15]	石锐, 朱鑫, 刘辉, 等. 微根窗图像中根系的自动分割方法[J]. 计算机工程与应用, 2013, 49(23): 181-185.
	SHI R, ZHU X, LIU H, et al. Automatic segmentation method in minirhizotron images[J]. Computer Engineering and Applications, 2013, 49(23): 181-185. (in Chinese with English abstract)
[16]	郭海宇. 植物根系图像的特征分析方法研究与实现[D]. 成都: 电子科技大学, 2013.
	GUO H Y. Research and implementation of characteristic analysis method using plant root images[D]. Chengdu: University of Electronic Science and Technology of China, 2013. (in Chinese with English abstract)
[17]	BURTON A J, PREGITZER K S, HENDRICK R L. Relationships between fine root dynamics and nitrogen availability in Michigan northern hardwood forests[J]. Oecologia, 2000, 125(3): 389-399.
[18]	陈倩秋. 蓄水坑灌不同坑深及灌水对苹果树根系生长的影响研究[D]. 太原: 太原理工大学, 2018.
	CHEN Q Q. Study on effect of pit depths and irrigation under water storage pit irrigation on the root growth of apple trees[D]. Taiyuan: Taiyuan University of Technology, 2018. (in Chinese with English abstract)
[19]	KINGMA D P, BA J. Adam: a method for stochastic optimization[EB/OL]. (2014-12-22) [2023-01-16]. https://arxiv.org/abs/1412.6980.
[20]	韩巧玲, 柏浩, 赵玥, 等. 采用染色示踪技术的土壤优先流自动分割与量化系统[J]. 农业工程学报, 2021, 37(6): 127-134.
	HAN Q L, BAI H, ZHAO Y, et al. Automatic segmentation and quantitative analysis of soil preferential flow using dye tracer technology[J]. Transactions of the Chinese Society of Agricultural Engineering, 2021, 37(6): 127-134. (in Chinese with English abstract)

基于改进的PSPNet网络的毛白杨根系自动分割量化系统

Automatic segmentation and quantification system for Populus tomentosa roots based on improved PSPNet

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 7

参考文献 20

相关文章 2

编辑推荐

Metrics

本文评价

[1]	王英允, 龙燕, 杨智优, 黄铝文. 基于改进U-Net网络的苹果叶部病害语义分割方法[J]. 浙江农业学报, 2023, 35(11): 2731-2741.
[2]	朱世松, 马婉丽, 赵理山, 郑艳梅, 郑先波, 芦碧波. 基于改进的LinkNet的苹果叶片图像分割算法[J]. 浙江农业学报, 2023, 35(1): 202-214.