作者：王姣, 孙皓月, 马娉妍, 刘雅军, 张大伟

作者单位：河北建筑工程学院信息工程学院, 河北 张家口 075000

关键词：高光谱;缺素检测;感兴趣区域;卷积神经网络;支持向量机;主成分分析

摘要：

针对目前大多数农作物缺素研究多为传统的实验室化学方法、形态法和施肥法等，检测成本较高且会破坏到植物本身，不宜在实际农业生产中推广使用的问题。研究提出一种高光谱分析结合图像处理的方法，尝试使用新型手持式高光谱相机直接到户外马铃薯示范基地进行实地采拍，通过高光谱图像不同地物的类别光谱曲线差异搭建一个1D-CNN网络结构提取感兴趣区域并进行标签化处理，建立自然环境下的正常、缺氮、缺磷、缺钾四类马铃薯植株高光谱图像数据集。在SVM下使用PCA的特征提取与特征选择方法对缺素马铃薯植株分类识别进行对比实验，结论表明经PCA特征提取降维到20后，缺素植株总识别率由91.7%提高到93.1%，在保证准确识别率的情况下，降维处理可极大提高运行速度，与特征波段选择相比特征提取更适合本次的无损定性研究分析，为高光谱技术监测农作物生长状况问题提供一个新思路。

SVM nutrient element deficit identification of potato plantsbased on hyperspectral analysis
WANG Jiao, SUN Haoyue, MA Pingyan, LIU Yajun, ZHANG Dawei
College of Information Engineering, Hebei University of Architecture, Zhangjiakou 075000, China
Abstract: In view of the problems that most of the current researches on crop nutrient deficiency mainly use traditional laboratory chemical methods, morphological methods and fertilization methods, etc., which have high detection cost and damage to the plants themselves, and are not suitable for promotion and use in actual agricultural production. This paper proposes a hyperspectral analysis and image processing method, attempts to use a new handheld hyperspectral camera directly to the outdoor potato demonstration base for field shooting, builds a 1D-CNN network structure based on the spectral curve differences of different hyperspectral images, extracts the region of interest and carries out label processing. The hyperspectral image data sets of normal, nitrogen, phosphorus and potassium deficiency potato plants in natural environment were established. A comparison experiment was carried out on the classification and recognition of vegetative deficient potato plants based on PCA feature extraction and feature selection method based on SVM. The conclusion showed that the total recognition rate of vegetative deficient plants increased from 91.7% to 93.1% after dimension reduction by PCA feature extraction to 20. Under the condition of ensuring accurate recognition rate, dimension reduction treatment could greatly improve the running speed. Compared with feature band selection, feature extraction is more suitable for this lossless qualitative research，It provides a new idea for monitoring crop growth by hyperspectral technology.
Keywords: hyperspectral;deficiency detection;region of interest;convolutional neural network;support vector machine;principal component analysis
2023, 49(11):141-149  收稿日期: 2023-01-19;收到修改稿日期: 2023-03-15
基金项目: 2022年度市级科技计划自筹经费项目（2221008A）
作者简介: 王姣(1994-),女,河北张北县人,助教,硕士,研究方向为模式识别与智能控制。
参考文献
[1] 王福德, 荣蓉. 农作物营养元素缺乏症状及防治措施[J]. 农村实用科技信息, 2010(183): 33.
[2] 韦德顺, 胡万芬. 重庆市开州区李子营养元素分析及其施肥措施[J]. 南方农业, 2018, 12(32): 34-36.
[3] 沐婵, 钱荣青, 吕艳玲, 等. 蓝莓“灿烂”叶片黄化症的相关营养元素分析[J]. 中国南方果树, 2019, 48(3): 121-123+127.
[4] 王谢, 唐甜, 张建华. 桑树13种不同营养元素单一缺乏时生长期落叶情况的定量分析[J]. 中国蚕业, 2020, 41(3): 1-3.
[5] 李江文, 李治国, 王瑞珍, 等. 草甸草原常见天然牧草营养元素差异性分析[J]. 中国草地学报, 2016, 38(6): 66-70.
[6] 孙俊, 张梅霞, 毛罕平,等. 基于高光谱图像的桑叶农药残留种类鉴别研究[J]. 农业机械学报, 2015, 46(6): 251-256.
[7] ZHANG C, LIU F, HE Y. Identification of coffee bean varieties using hyperspectral imaging: influence of preprocessing methods and pixel-wise spectra analysis[J]. Scientific Reports, 2018, 8(1): 2166.
[8] 冯海宽, 杨福芹, 杨贵军, 等. 基于特征光谱参数的苹果叶片叶绿素含量估算[J]. 农业工程学报, 2018, 34(6): 182-188.
[9] 胡耀华, 平文学, 徐明珠, 等. 高光谱技术诊断马铃薯叶片晚疫病的研究[J]. 光谱学与光谱分析, 2016, 36(2): 515-519.
[10] WEN W K, CHU Z, FENG C, et al. Detection of Sclerotinia Stem Rot on Oilseed Rape (Brassica napus L.) Leaves Using Hyperspectral Imaging[J]. Sensors, 2018, 18(6): 1764-.
[11] XIE C Q, YANG C, HE Y. Hyperspectral imaging for classification of healthy and gray mold diseased tomato leaves with different infection severities[J]. Computers & Electronics in Agriculture, 2017, 135: 154-162.
[12] 田文忠, 赵庆展, 胡浩伟,等. 无人机高光谱载荷性能交叉验证[J]. 中国测试, 2019, 45(11): 131-137.
[13] 杨蕴睿, 郑东文. 一种用于高光谱图像分类的空谱协同编码方法[J]. 中国测试, 2022, 48(12): 162-171.
[14] 孟庆龙, 张艳, 尚静. 基于高光谱成像技术的苹果表面缺陷无损检测[J]. 食品研究与开发, 2019, 40(5): 168-172.
[15] 于成龙. 基于高光谱数据的主要农作物类型信息提取[J]. 东北农业科学, 2019, 44(3): 45-51.