欢迎访问作物学报，今天是
作物学报

作物学报 ›› 2025, Vol. 51 ›› Issue (7): 1861-1873.doi: 10.3724/SP.J.1006.2025.53008

• 耕作栽培·生理生化 • 上一篇    下一篇

基于改进SIFT算法的田间作物根系图像拼接方法研究

向梓薇(), 王韵博, 颜小飞()   

  1. 北京林业大学工学院, 北京 100083
  • 收稿日期:2025-02-17 接受日期:2025-04-25 出版日期:2025-07-12 网络出版日期:2025-05-13
  • 通讯作者: *颜小飞, E-mail: yanxf@bjfu.edu.cn
  • 作者简介:xiangziwei1113@163.com
  • 基金资助:
    国家自然科学基金项目(31971576)

Image stitching method for crop roots based on an improved SIFT algorithm

XIANG Zi-Wei(), WANG Yun-Bo, YAN Xiao-Fei()   

  1. School of Technology, Beijing Forestry University, Beijing 100083, China
  • Received:2025-02-17 Accepted:2025-04-25 Published:2025-07-12 Published online:2025-05-13
  • Contact: *E-mail: yanxf@bjfu.edu.cn
  • Supported by:
    National Natural Science Foundation of China(31971576)

RichHTML

3

PDF

23

摘要：

获取根系信息对研究作物养分吸收和水分利用效率具有重要意义。目前常用的微根管方法虽然能够获取根系图像, 但难以将分散拍摄的局部根系图像整合为连续的根系分布图, 限制了根系表型特征的连续提取和定量分析。因此, 本文基于课题组自主研制的根系图像自动监测管道机器人, 提出了一种高效、快速的根系图像拼接方法, 实现对根系全景图像的构建。首先, 利用机器人系统自动采集根系图像, 并采用Gamma校正和CLAHE算法增强图像的亮度和局部对比度; 然后基于改进SIFT算法设置重叠区域边界, 并利用自适应阈值筛选高响应特征点, 同时引入PCA降维方法降低计算复杂度; 最后, 使用多波段融合技术实现无缝拼接。试验选取3组玉米不同生长阶段的根系图像, 并将改进SIFT算法与传统特征提取算法(ORB、SURF、SIFT)进行对比。结果显示,预处理图像的平均对比度和平均信息熵分别提升19%和15%; 改进SIFT算法的正确匹配率较ORB、SURF、SIFT算法分别提升91.7%、35.9%和24.3%, 平均时间效率提升1.12倍、11.57倍和1.11倍。此外, 为验证本文所提方法的稳定性和鲁棒性, 设置了5组不同放缩比例的尺度变换试验。结果表明,改进SIFT算法在平均重叠面积和百分比2项指标上均达到最高值。综上, 该方法应用于根系图像自动监测管道机器人系统中, 可高效拼接根系图像, 为后续根系表型分析奠定基础。

关键词: 图像拼接, SIFT算法, 特征提取, 特征匹配, 管道机器人

Abstract:

Obtaining accurate information on crop root systems is essential for studying nutrient uptake and water use efficiency. While the widely used minirhizotron method can capture root images in situ, it remains challenging to integrate these localized images into a continuous map of root distribution, thereby limiting the ability to extract and analyze root phenotypic traits comprehensively. To address this limitation, we propose an efficient and rapid root image stitching method based on a root system image acquisition pipeline robot independently developed by our research group. The method consists of four main steps. First, the robotic system automatically captures root images, and image quality is enhanced using Gamma correction and the CLAHE algorithm to improve brightness and local contrast. Next, an improved SIFT algorithm is employed to define the overlapping boundaries, while adaptive thresholding is applied to filter high-response feature points. Simultaneously, PCA-based dimensionality reduction is introduced to lower computational complexity. Finally, multi-band fusion technology is used to achieve seamless image stitching. To evaluate performance, three sets of maize root images at different growth stages were tested, and the improved SIFT algorithm was compared with conventional feature extraction methods(ORB, SURF, and SIFT). Results showed that the average contrast and information entropy of the pre-processed images increased by 19% and 15%, respectively. The improved SIFT algorithm achieved correct matching rate improvements of 91.7%, 35.9%, and 24.3% over ORB, SURF, and SIFT, respectively, and enhanced time efficiency by factors of 1.12, 11.57, and 1.11. Additionally, to assess the robustness and stability of the proposed method, five groups of experiments involving different scaling ratios were conducted. The results demonstrated that the improved SIFT algorithm consistently achieved the highest average overlapping area and percentage. In conclusion, this method can be effectively integrated into automated root image monitoring systems, providing a reliable foundation for subsequent phenotypic analysis of crop root systems.

Key words: image stitching, SIFT algorithm, feature extraction, feature matching, pipeline robot

图1

管道机器人系统实物图1 : 集成驱动模块。2: 图像采集模块。3: 电源模块。4: 数据基站。"

图2

根系图像采集试验示意图"

图3

作物根系图像拼接方法流程图"

图4

预处理结果 a: 原始图像。b: Gamma处理结果。c: CLAHE处理结果。"

表1

预处理效果比较"

试验组
Experimental group		 对比度增强指数
Contrast improvement index		 信息熵
Entropy
原始图像
Original image		 预处理图像
Pre-processed image		 原始图像
Original image		 预处理图像
Pre-processed image
明显主根
Apparent primary root		 29.19
 33.57
 6.43
 7.05
细根较密
Density fine root		 25.30
 27.00
 6.13
 6.77
细根较少
Few fine root		 17.10
 24.73
 5.15
 6.53

图5

自适应阈值参数α对特征匹配性能的影响 a: 试验组别1。b: 试验组别2。c: 试验组别3。d: 试验组别4。"

图6

自适应阈值参数α的特征匹配效果对比 a: 自适应阈值α=0.50。b: 自适应阈值α=0.75。c: 自适应阈值α=1.00。"

图7

待匹配的根系图像 a: 参考图像。b: 待匹配图像。"

图8

4种算法之间的匹配结果比较 a: ORB算法。b: SURF算法。c: SIFT算法。d: 改进SIFT算法。"

图9

4种算法的时间消耗量的比较 a: 特征重叠较多。b: 特征重叠较少。c: 根系特征丰富。d: 根系特征匮乏。"

图10

4种算法的正确匹配率比较 a: 特征重叠较多。b: 特征重叠较少。c: 根系特征丰富。d: 根系特征匮乏。"

图11

4种特征提取算法的拼接结果比较 不同算法处理结果的不一致原因: 相邻原始图像之间的匹配点对不足, 难以得到一幅连续、完整的拼接长图。a: 原始图像。b: ORB算法。c: SURF算法。d: SIFT算法。e: 改进SIFT算法。"

表2

第1组试验中重叠面积与重叠百分比的比较"

图像尺度
Image scale		 重叠面积
Overlapping area (×104 pixel)		 重叠百分比
Overlapping percentage (%)
ORB SURF SIFT Improved-SIFT ORB SURF SIFT Improved-SIFT
0.50 1.88 1.88 1.88 3.04 32.85 32.85 32.85 53.15
0.75 7.85 6.93 6.90 6.96 61.07 53.92 53.49 54.14
1.00 12.35 12.13 12.43 12.43 53.92 52.96 54.29 54.30
1.25 17.79 18.60 17.90 19.04 49.65 51.93 49.98 53.15
1.50 24.14 27.52 24.29 28.20 46.87 53.43 47.14 54.74

表3

第2组试验中重叠面积与重叠百分比的比较"

图像尺度
Image scale		 重叠面积
Overlapping area (×104 pixel)		 重叠百分比
Overlapping percentage (%)
ORB SURF SIFT Improved-SIFT ORB SURF SIFT Improved-SIFT
0.50 1.88 2.30 1.88 3.16 32.85 40.11 32.85 55.22
0.75 7.91 7.00 4.24 7.05 61.54 54.16 33.01 54.88
1.00 12.54 11.45 13.77 12.04 54.77 50.38 50.13 52.57
1.25 18.16 18.87 19.57 19.85 50.71 53.14 54.63 55.41
1.50 24.74 26.04 26.28 27.98 48.03 53.37 51.01 54.32

图12

改进SIFT算法全景拼接结果图像 a: 8月1日, 发根期。b: 8月31日, 生长期。c: 9月8日, 衰亡期。"

[1] 赵蕾, 刘润慧, 张高煜, 唐清芸, 王子建, 魏萌, 王国栋, 李玉祥. 不同灌水量对滴灌水稻叶片光合特性及根系内源激素的影响. 中国农业大学学报, 2023, 28(1): 12-26.
Zhao L, Liu R H, Zhang G Y, Tang Q Y, Wang Z J, Wei M, Wang G D, Li Y X. Effects of different irrigation amounts on the photosynthetic characteristics of rice leaves and root endogenous hormones under mulching drip irrigation. J China Agric Univ, 2023, 28(1): 12-26 (in Chinese with English abstract).
[2] Rich S M, Christopher J, Richards R, Watt M. Root phenotypes of young wheat plants grown in controlled environments show inconsistent correlation with mature root traits in the field. J Exp Bot, 2020, 71: 4751-4762.
doi: 10.1093/jxb/eraa201 pmid: 32347952
[3] Primka E J, Adams T S, Buck A S, Eissenstat D M. Shifts in root dynamics along a hillslope in a mixed, mesic temperate forest. Plant Soil, 2022, 477: 707-723.
[4] 程强, 刘雨欣, 杨涵青, 许新宇, 范继泽, 颜小飞, 杜太生. 面向作物干旱胁迫诊断的表型成像技术研究进展. 农业工程学报, 2024, 40(20): 1-11.
Cheng Q, Liu Y X, Yang H Q, Xu X Y, Fan J Z, Yan X F, Du T S. Research progress on the phenotype imaging technology for diagnosis of crop drought stress. Trans CSAE, 2024, 40(20): 1-11 (in Chinese with English abstract).
[5] 李龙, 李超男, 毛新国, 王景一, 景蕊莲. 作物根系表型鉴定评价方法的现状与展望. 中国农业科学, 2022, 55: 425-437.
doi: 10.3864/j.issn.0578-1752.2022.03.001
Li L, Li C N, Mao X G, Wang J Y, Jing R L. Advances and perspectives of approaches to phenotyping crop root system. Sci Agric Sin, 2022, 55: 425-437 (in Chinese with English abstract).
doi: 10.3864/j.issn.0578-1752.2022.03.001
[6] 张翠梅, 师尚礼, 吴芳. 干旱胁迫对不同抗旱性苜蓿品种根系生长及生理特性影响. 中国农业科学, 2018, 51: 868-882.
doi: 10.3864/j.issn.0578-1752.2018.05.006
Zhang C M, Shi S L, Wu F. Effects of drought stress on root and physiological responses of different drought-tolerant alfalfa varieties. Sci Agric Sin, 2018, 51: 868-882 (in Chinese with English abstract).
doi: 10.3864/j.issn.0578-1752.2018.05.006
[7] 艾栋, 刘青丽, 常乃杰, 闫芳芳, 边立丽, 李斌, 李志宏, 冯文强, 张宗锦, 陈曦, 等. 菌丝营养钵栽培对烤烟根系生长的影响. 植物营养与肥料学报, 2022, 28: 181-190.
Ai D, Liu Q L, Chang N J, Yan F F, Bian L L, Li B, Li Z H, Feng W Q, Zhang Z J, Chen X, et al. Effects of mycelium-straw bowl on the rhizospheric condition and root growth of flue-cured tobacco. J Plant Nutr Fert, 2022, 28: 181-190 (in Chinese with English abstract).
[8] 王春辉, 祝鹏飞, 束良佐, 朱继荣, 于红梅, 詹雨珊, 袁梅. 分根区交替灌溉和氮形态影响土壤硝态氮的迁移利用. 农业工程学报, 2014, 30(11): 92-101.
Wang C H, Zhu P F, Shu L Z, Zhu J R, Yu H M, Zhan Y S, Yuan M. Effects of alternate partial root-zone irrigation and nitrogen forms on utilization and movement of nitrate in soil. Trans CSAE, 2014, 30(11): 92-101 (in Chinese with English abstract).
[9] Shen C, Liu L T, Zhu L X, Kang J, Wang N, Shao L M. High-throughput in situ root image segmentation based on the improved DeepLabv3+ method. Front Plant Sci, 2020, 11: 576791.
[10] 付莉娇, 李雪琴, 范继辉, 鲁旭阳, 鄢燕. 藏北高寒草原典型植物根际土壤细菌群落结构多样性及根系特征分析. 草地学报, 2022, 30: 1131-1140.
doi: 10.11733/j.issn.1007-0435.2022.05.013
Fu L J, Li X Q, Fan J H, Lu X Y, Yan Y. Analysis of rhizosphere soil bacterial community structure diversity and root characteristics of typical plants in alpine steppe of Northern Tibet. Acta Agrest Sin, 2022, 30: 1131-1140 (in Chinese with English abstract).
[11] 郑一力, 张振翔, 邢达, 刘卫平. 基于微根管图像的作物根系分割和表型信息提取. 农业工程学报, 2024, 40(18): 110-119.
Zheng Y L, Zhang Z X, Xing D, Liu W P. Crop root system segmentation and phenotypic information extraction based on images of minirhizotron. Trans CSAE, 2024, 40(18): 110-119 (in Chinese with English abstract).
[12] Peters B, Blume-Werry G, Gillert A, Schwieger S, von Lukas U F, Kreyling J. As good as human experts in detecting plant roots in minirhizotron images but efficient and reproducible: the convolutional neural network “RootDetector”. Sci Rep, 2023, 13: 1399.
doi: 10.1038/s41598-023-28400-x pmid: 36697423
[13] 吴茜, 张伟欣, 张玲玲, 孙传亮, 刘乃森, 岳延滨, 曹静, 梁万杰, 葛道阔, 唐普传, 等. 植物根系表型信息获取技术研究进展. 江苏农业科学, 2021, 49(5): 31-37.
Wu Q, Zhang W X, Zhang L L, Sun C L, Liu N S, Yue Y B, Cao J, Liang W J, Ge D K, Tang P C, et al. Research progress on acquisition of plant root phenotype information. Jiangsu Agric Sci, 2021, 49(5): 31-37 (in Chinese).
[14] 颜小飞, 王韵博, 宋晓波, 向梓薇, 杜太生, 程强. 植物根系特征与根区土壤水分高通量监测管道机器人研制. 农业工程学报, 2024, 40(6): 192-202.
Yan X F, Wang Y B, Song X B, Xiang Z W, Du T S, Cheng Q. Development of a pipeline robot for high-throughput monitoring plant root characteristics and soil moisture in root zone. Trans CSAE, 2024, 40(6): 192-202 (in Chinese with English abstract).
[15] Pu Y R, Chen C L. A comparison of cross-correlation-based and phase-correlation-based image registration algorithms for optical coherence tomographic angiography. Chin Opt Lett, 2024, 22: 071101.
[16] 刘媛媛, 何铭, 王跃勇, 孙宇, 高雪冰. 基于优化SIFT算法的农田航拍全景图像快速拼接. 农业工程学报, 2023, 39(1): 117-125.
Liu Y Y, He M, Wang Y Y, Sun Y, Gao X B. Fast stitching for the farmland aerial panoramic images based on optimized SIFT algorithm. Trans CSAE, 2023, 39(1): 117-125 (in Chinese with English abstract).
[17] 潘维东, 李安虎, 刘兴盛. 基于区域优化的图像拼接技术研究及应用进展. 激光与光电子学进展, 2024, 61(18): 1-20.
Pan W D, Li A H, Liu X S. Progress in research and application of image stitching technology based on regional optimization. Laser Optoelectr Prog, 2024, 61(18): 1-20 (in Chinese with English abstract).
[18] Nie L, Lin C Y, Liao K, Liu S C, Zhao Y. Unsupervised deep image stitching: reconstructing stitched features to images. IEEE Trans Image Proc, 2021, 30: 6184-6197.
[19] Lowe D G. Distinctive image features from scale-invariant keypoints. Int J Comput Vis, 2004, 60: 91-110.
[20] Bay H, Tuytelaars T, Van Gool L. SURF:speeded up robust Features. Computer Vision-ECCV 2006. Berlin, Heidelberg: Springer Berlin Heidelberg, 2006. pp 404-417.
[21] Rublee E, Rabaud V, Konolige K, Bradski G. ORB: an efficient alternative to SIFT or SURF. 2011 International Conference on Computer Vision. Barcelona, Spain: IEEE, 2011. pp 2564-2571.
[22] Tang Z T, Zhang Z M, Feng J J, Chen W, Zhu K, Yang W T, Han W J. A fast image stitching algorithm based on texture classification and improved SIFT. IEEE Access, 2024, 12: 124183-124208.
[23] Zhang W N, Zhao Y Q. An improved SIFT algorithm for registration between SAR and optical images. Sci Rep, 2023, 13: 6346.
doi: 10.1038/s41598-023-33532-1 pmid: 37072451
[24] Ye R C, Qian Y Q, Huang X M. RT-CBAM: refined transformer combined with convolutional block attention module for underwater image restoration. Sensors, 2024, 24: 5893.
[25] Zhu J Y, Park T, Isola P, Efros A A. Unpaired image-to-image translation using cycle-consistent adversarial networks. 2017 IEEE International Conference on Computer Vision (ICCV). Venice, Italy: IEEE, 2017. pp 2242-2251.
[26] 张峰, 黄仕鑫, 花强, 董春茹. 基于Depth-wise卷积和视觉Transformer的图像分类模型. 计算机科学, 2024, 51(2): 196-204.
doi: 10.11896/jsjkx.221100234
Zhang F, Huang S X, Hua Q, Dong C R. Novel image classification model based on depth-wise convolution neural network and visual transformer. Comput Sci, 2024, 51(2): 196-204 (in Chinese with English abstract).
[27] Li C N, Li L, Reynolds M P, Wang J Y, Chang X P, Mao X G, Jing R L. Recognizing the hidden half in wheat: root system attributes associated with drought tolerance. J Exp Bot, 2021, 72: 5117-5133.
[28] Ke Y, Sukthankar R. PCA-SIFT: a more distinctive representation for local image descriptors. Proceedings of the 2004 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2004. Washington, DC, USA: IEEE, 2004. II.
[29] Friedman J H, Bentley J L, Finkel R A. An algorithm for finding best matches in logarithmic expected time. ACM Trans Math Softw, 1977, 3: 209-226.
No related articles found!
Viewed
Full text


Abstract

Cited
  Shared   
  Discussed   
No Suggested Reading articles found!
京ICP备15008789号-2