龙 燕,连雅茹,马敏娟 宋怀波,何东健
基于高光谱技术和改进型区间随机蛙跳算法的番茄硬度检测
龙 燕1,2,3,连雅茹1,2,3,马敏娟1,2,3宋怀波1,2,3,何东健1,2,3
(1. 西北农林科技大学机械与电子工程学院,杨凌 712100;2. 农业农村部农业物联网重点实验室,杨凌 712100;3. 陕西省农业信息感知与智能服务重点实验室,杨凌 712100)
为了准确、快速的检测番茄硬度,该文提出了一种基于改进型区间随机蛙跳算法优选高光谱特征波长的番茄硬度检测模型。在获取番茄高光谱图像后,首先对光谱数据进行多元散射校正(multiplicative scatter correction,MSC)和归一化预处理。针对区间随机蛙跳算法(interval random frog,iRF)所需迭代次数大、算法收敛慢等缺点,该文提出了改进型区间随机蛙跳算法(modified interval random frog, miRF),并将其应用于特征波长选择。最后建立偏最小二乘回归模型(partial least squares regression, PLSR)预测番茄的硬度。iRF共选出特征波段100个,算法收敛时间为32.1 min,而miRF共选出特征波长47个,算法收敛仅需1.6 min。同时miRF-PLSR番茄硬度预测精度也更优,测试集相关系数达到了0.968 5,均方根误差为0.004 0kg/mm2。试验结果表明:结合高光谱技术和miRF算法可实现对番茄硬度的快速、无损检测。
光谱分析;算法;模型;高光谱技术;番茄;硬度;特征波长;区间随机蛙跳
番茄富含多种营养成分,不仅有着很高的食用价值,还有一定的的药用价值。近年来,番茄已成为全球栽培最广、消费量最大的蔬菜作物之一,而中国也是世界最大的番茄生产和消费国家之一[1]。番茄在成熟过程中由于水解酶的作用使细胞壁果胶含量下降[2],导致其硬度发生改变。因此,番茄果实硬度是判定其成熟度的重要指标之一。研究番茄硬度的测定方法可为番茄的质量评价以及番茄在储藏、运输过程中的硬度变化提供参考[3]。但传统的硬度检测方法一般采用果实硬度计进行测量,该方法耗费时间比较长且具有破坏性[4],从而难以应用于果品硬度的批量化检测[5-6]。因此开发无损、高效的番茄硬度的检测方法具有重要的意义。
高光谱成像技术具有无损、无污染、高效等优点[7-8],被广泛应用于果品硬度的无损检测[9-10]。郝勇等[11]采用基于小波变换的蒙特卡罗无信息变量消除方法对光谱变量进行筛选,建立偏最小二乘回归模型预测梨的硬。Sun等[12]利用高光谱技术检测甜瓜的糖度和硬度,对比分析了偏最小二乘回归、支持向量机和人工神经网络3种预测模型,试验结果表明偏最小二乘回归模型检测准确度最高。孙静涛等[13]结合高光谱技术和支持向量机检测哈密瓜的硬度和可溶性固形物。
王世芳等[14]在近红外漫反射780~2 500 nm波段利用杠杆率校正结合偏最小二乘回归对果实质地进行定量分析,其中果肉平均硬度预测相关系数为0.761。王世芳,宋海燕等[15]利用6个近红外特征波段建立番茄果肉硬度分析模型检测番茄硬度,相关系数为0.938。王凡等[16]利用可见/近红外全透射光谱(630~1 100 nm)对番茄多品质参数无损检测,对光谱数据进行SG卷积平滑、标准正态变量变换和多元散射校正等预处理后,建立偏最小二乘预测模型,硬度的预测集相关系数可达0.940 5。
高光谱漫反射成像技术可同时探测目标的二维集合空间与一维光谱信息,结合图像和光谱特点,能够获取番茄整体空间光谱信息,在番茄硬度无损检测上具有明显的优势[17]。但现有的基于高光谱技术的番茄硬度检测模型中,特征波段选择方法还需改进,有效光谱信息的提取仍是研究难点。本文以不同成熟时期的番茄为研究对象,利用高光谱漫反射成像获取番茄高光谱图像数据,对传统随机蛙跳算法进行改进,提出了一种基于改进型区间随机蛙跳的特征波长提取算法,建立番茄硬度偏最小二乘回归检测模型,以提供一种快速、准确的番茄硬度无损检测方法。
试验所用的样品为“海星”番茄,均为果形匀称、无缺陷、无损伤的番茄,来自陕西省咸阳市杨凌区某温室大棚。试验前先剔除畸形和表面损伤的样品,再对样品逐个清洁处理并编号,后将其置于温度20 ℃,相对湿度50%的试验条件下1 d。
参照中GB8852—1988的相关规定,可根据番茄颜色外观将试验样品粗略划分为“绿熟期”、“变色期”、“红熟前中期”和“红熟后期”。但实际上,番茄由于物理、化学损伤,放置时间久或番茄红素的使用等,通过颜色并不能准确判断番茄硬度。本试验番茄样品共120个,其中“绿熟期”20个、“变色期”20个、“红熟前中期”40个、“红熟后期”40个。
1.可见/进红外成像光谱仪 2.CCD相机 3.光纤卤素灯光源(150 W) 4.电控移动载物台 5.暗箱 6.计算机
番茄硬度采用TA.XT Plus质构仪(Stable Micro systems Ltd公司,英国)进行测定。选择型号为P/5的探头,设置预压速度、下压速度和压后上行速度分别为1、2、10 mm/s,触发力为0.05 N,穿刺深度为20 mm,以探头下压时产生的应力大小作为反映番茄硬度的指标。由于番茄的果皮与成熟度有较大的相关性,本试验选用带皮刺穿法进行番茄的硬度测量[18]。质构仪预热30 min后,将样品番茄竖立于载物台上。从番茄的结构来看,番茄带液汁的腔室占据大部分,选择赤道面带汁液的地方作为测量点更易保证数据一致性。本试验在赤道面带汁液的地方选择4个测量点,如图2中a、b、c、d所示,并将4个测量点的平均值作为番茄硬度测定值。
图3为本文研究所采用的质构仪及硬度测定过程中的应力变化曲线图。表皮未被探头刺穿前,番茄在压力作用下产生一定的变形,当压力值逐渐上升到达峰值点时番茄被刺穿,峰值的应力大小即可视为番茄的表皮硬度(锚1位置)。随后,应力大小迅速下降并相对稳定直到穿刺深度达到设定值(锚2位置),此时测试探头已刺入番茄果肉。本研究选取锚1位置和锚2位置之间的平均值来计算番茄的硬度值。
图2 番茄硬度测量点示意图
1.机械臂 2.探头 3.测试平台 4.控制按钮 5.急停按钮
由于原始光谱曲线两端噪声较大,截取信息量较丰富且平滑的881.71~1 695.11 nm(共246个波段)的光谱数据作为定标和建模。为了消除光谱噪声,提高信噪比,本文采用MSC和归一化对原始光谱数据进行预处理。
图4 番茄图像背景分割和感兴趣区域的提取结果
模型的预测性能好坏,究其根本在于特征波长的选择,当特征波长的选择过少时,会导致部分有用信息的缺失;但当特征波长的选择过多时,会出现波段信息的冗余导致模型精度低。因此,需要一种有效的手段来提取有效波长,提高建模的精度。
区间随机蛙跳算法(interval random frog,iRF)算法是由Yun等提出的一种特征变量选取方法[19]。该算法[20-22]同时具有适者生存和随机搜索的特性,能够按照定义好的策略更新变量子集,当满足迭代次数后,统计每个波段被选择的概率并降序排列,实现了局部信息的传递,最终选择最优波段。该算法主要的运算步骤包括以下5步:
1)随机选取个光谱波段组成初始变量子集V,设定迭代次数。
2)基于初始变量子集,选出候选变量子集V,包含个波段:首先利用V建立PLS模型,计算每个波段的绝对回归系数,并对各波段的绝对回归系数降序排列:
a.若=,则V=V0;
b.若<,前个波段构成候选子集V;
c.若>Q,前个波段构成候选子集V;
3)令V=V(=1~),并利用更新V。重复上述过程直到次迭代结束。
4)计算次迭代后产生的个变量子集中各每个波段被选择概率,并按降序排列。
5)依次联合被选概率排列前10、前11,直到前246个波段,每一组波段均进行交叉验证,分别得到联合均方根误差,均方根误差最小组中的波段即为被选波段。
在原始iRF算法中初始波段子集的产生是随机的,具有很大的不确定性,难以保证初始信息的有效性,导致结果再现性低,这就要求迭代次数必须足够大,以保证算法遍历整个数据集,因此算法的运行时间长、收敛速度慢。为了提高iRF算法的收敛速度和寻优精度,本文对该算法初始变量子集的构造进行改进。连续投影(successive projections algorithm,SPA)算法能够从光谱信息中充分寻找含有最低限度冗余信息的变量组来概括大多数样品的光谱信息,最大程度避免信息的重复[23-24]。因此,本文首先利用SPA算法对特征波段进行初选,将SPA初选结果作为iRF的初始变量子集,从而减少iRF算法的迭代次数。本文将这种利用SPA初选波段作为初始变量子集的区间随机蛙跳算法称为改进型区间随机蛙跳算法。在 SPA算法初选iRF初始变量子集时,SPA特征波段的个数分别设置为10、20、30、40、50、60、70,通过试验可知当SPA选出的波段数为40时,改进型区间随机蛙跳算法(modified interval random frog, miRF)迭代次数最小,对番茄的硬度检测准确度最高。
为确保模型预测结果的有效性,先利用“二审”回收算子法剔除异常样本7个,并采用Kennard-Stone算法将番茄样本划分为训练集(80%)和测试集(20%),分别用于建模和预测,表1所示为番茄样本硬度信息表。本文采用偏最小二乘回归法建立番茄硬度的预测模型。该方法能直接反映光谱数据对化学指标的预测相关性,是常用的光谱定量分析建模方法[25]。模型的优劣主要由测试集的相关系数和均方根误差评定,训练集的均方根误差作为辅助评价指标。测试集相关系数越高且均方根误差越小,表示模型的效果越好,同时训练集的均方根误差亦是越小越好。
表1 番茄样本硬度信息表
经过SPA算法筛选共产生40个光谱变量作为miRF的初始变量子集,并设置窗口大小为1,最大主成分数为4,迭代次数500,利用miRF进行特征波段的提取,并与传统iRF算法进行比较。表2统计了iRF和miRF的运行时间和所需的迭代次数,miRF算法实现收敛仅需1.6 min,迭代500次;而iRF算法迭代10 000次后才达到收敛,算法运行时间为32.1 min,是miRF算法的20倍左右。试验结果表明miRF算法在高效性方面具有很大的优势。
表2 iRF和miRF算法的比较
图5a为各波段被选择为特征波段的概率,其中横坐标为波段数,纵坐标为该波段被选择的概率。图5b为联合均方根误差的计算结果,最小均方根误差所在位置为第38个区间,所含波段数为47,均方根误差为0.005 3 kg/mm2。
注:图中正方形标出了最小均方根误差
本文利用iRF算法和miRF算法分别提取了100个和47个光谱特征波段,统计图如图6所示。可以看出,2种方法提取的特征波段所在范围大概一致,多集中在1 582~1 655 nm范围内,其次是1 160~1 190 nm和1 353~1 383 nm范围内,说明这些区域是对番茄硬度敏感的区域。
本文利用PLSR建立番茄硬度预测模型。为证明本文算法miRF-PLSR的有效性,将其分别与SPA-PLSR,iRF-PLSR的番茄硬度预测模型比较。
本文所用计算机型号为90GKCTO1WW,主频为3 GHz,内存为8 GB,软件平台为Matlab2018a。SPA-PLSR共选出特征波段40个,占全波段的16.26%;iRF-PLSR共选出特征波段100个,占全波段的40.65%,模型运行时间为32.1 min;miRF-PLSR共选出特征波段47个,占全波段的19.1%,模型运行时间为1.6 min。由此可见,miRF算法不仅有效降低了模型的复杂度,而且大大减少了算法运行时间。
注:图中正方形标出的为被选波段
硬度预测值与实测值散点图如图7所示,图中横坐标为硬度实测值,纵坐标为硬度预测值。本文根据测试集的相关系数R和均方根误差RMSEP对试验结果进行评价,同时利用训练集的均方根误差作为辅助评价指标。表3为3种番茄硬度预测模型结果比较。由表3可以看出,SPA-PLSR的拟合精度较低,测试集相关系数和均方根误差分别为0.803 9和0.007 7 kg/mm2。iRF-PLSR模型测试集相关系数和均方根误差分别为0.936 6和0.004 4 kg/mm2。本文miRF-PLSR模型的拟合效果最好,训练集的均方根误差为0.004 1 kg/mm2,测试集相关系数和均方根误差分别为0.968 5和0.004 0 kg/mm2。
注:图中星号表示训练集样本,实心方块代表测试集样本
表3 3种模型结果比较
本文利用高光谱技术对番茄硬度进行无损检测,提出了miRF特征波长提取方法,并通过与SPA、iRF算法进行比较,分析了它们对番茄硬度预测模型时效性和准确度的影响,证明了miRF算法的可行性和优越性。
利用SPA算法建立的PLSR模型预测精度较低(R和RMSEP分别为0.803 9和0.007 7 kg/mm2),主要原因是算法选择的特征波长较少,导致部分有用信息的缺失,影响了模型的精度。iRF算法根据每个波段被作为特征波段的概率进行选择,能在一定程度上改善SPA算法导致的光谱信息缺失问题,所建模型的预测精度得到了明显的优化,R为0.936 6,比SPA-PLSR模型提高了0.132 7,RMSEP为0.004 4 kg/mm2。但其初始子集随机产生,导致算法所需迭代次数较多(10 000次),算法收敛速度慢(32.1 min)。
本文提出的miRF算法对iRF初始变量子集进行有效的构造,融合了SPA算法和传统iRF算法的优势来提取光谱的有效信息。因此miRF算法的所需迭代次数减小到500次,收敛时间仅需1.6 min。miRF算法中,SPA筛选产生的初始变量子集,包含了40个光谱变量的,大部分波段分布在对番茄硬度敏感的区域内。但iRF随机产生的变量子集中,波段的分布毫无规律,大部分波段游离在敏感区域外,这就导致算法收敛时间过长、模型精度降低。传统的iRF算法最终选择的波段数是100个,而改进的miRF算法最终选择的波段数是47个,相对于iRF算法减少了53个,有效地消除了光谱中的冗余信息。且iRF的最小联合均方根误差为0.006 0 kg/mm2,而miRF的最小联合均方根误差为0.005 3 kg/mm2,减小了 0.000 7。因此改进后的模型预测精度更好(RMSEC为0.004 1 kg/mm2,R和RMSEP分别为0.968 5和0.004 0 kg/mm2)。
综上所述,本文提出的特征波长提取方法在一定程度上克服了光谱信息缺失或冗余的问题,提高了番茄硬度检测的时效性和准确度。对其他果蔬的硬度无损检测也具有一定的参考价值。
该文利用高光谱技术对番茄硬度进行无损检测,提出了改进型区间随机蛙跳算法(modified interval Random Frog, miRF)提取特征波长,并通过与连续投影(successive projections algorithm,SPA)算法、区间随机蛙跳算法(interval Random Frog,iRF)进行比较,分析了它们对番茄硬度预测的速度和精度的影响,主要结论如下:
1)证明了iRF算法提取特征波长在番茄硬度预测模型中的可行性和优越性。iRF-PLSR模型的测试集相关系数为0.936 6,比SPA-PLSR模型提高了0.132 7,同时测试集的均方根误差减小到0.004 4 kg/mm2。
2)为克服传统iRF算法收敛时间过长和模型实用性差的不足,本文从初始变量子集构造选择方面对iRF进行改进,建立miRF-PLSR番茄硬度预测模型。miRF在特征波长选择上的有效性,使得该模型在时效性上优于传统的iRF-PLSR算法,算法收敛时间由32.1 min降低至1.6 min。预测效果也更好,训练集的均方根误差为0.004 1 kg/mm2,测试集的相关系数和均方根误差分别为0.968 5和0.004 0 kg/mm2。该研究为番茄硬度无损检测提供了新思路,也为番茄自动采收、自动分级设备的开发提供理论依据。
[1] 霍建勇. 中国番茄产业现状及安全防范[J]. 蔬菜,2016(6):1-4. Huo Jianyong. China's tomato industry status and safety precautions[J].Vegetables, 2016(6): 1-4. (in Chinese with English abstract)
[2] 杨生保,唐亚萍,杨涛,等. 加工型番茄果实硬度特异材料的果实特性及果肉组织特征[J]. 农业工程学报,2017,33(18):285-290. Yang Shengbao, Tang Yaping, Yang Tao, et al. Fruit characteristic and flesh tissue feature of special firmness type processing tomato cultivar[J]. Transactions of the Chinese Society of Agricultural Engineering(Transactions of the CSAE), 2017, 33(18): 285-290. (in Chinese with English abstract)
[3] Sohrabi Mohammad Mahdi, Ahmadi Ebrahim, Monavar Hosna Mohammadi. Nondestructive analysis of packaged grape tomatoes quality using PCA and PLS regression by means of fiber optic spectroscopy during storage[J]. Journal of Food Measurement and Characterization, 2018, 12(2): 949-966.
[4] Xie Chuanqi, Chu Bingquan, He Yong. Prediction of banana color and firmness using a novel wavelengths selection method of hyperspectral imaging[J]. Food Chemistry, 2018, 245: 132-140.
[5] 郭文川,董金磊. 高光谱成像结合人工神经网络无损检测桃的硬度[J]. 光学精密工程,2015,23(6):1530-1537. Guo Wenchuan, Dong Jinlei. Nondestructive detection on firmness of peaches based on hyperspectral imaging and artificial neural networks[J]. Optics and Precision Engineering, 2015, 23(6): 1530-1537. (in Chinese with English abstract)
[6] 胡孟晗,董庆利,刘宝林. 基于高光谱反射、透射和交互作用成像模式的蓝莓硬度和弹性模量预测的比较[J]. 光谱学与光谱分析,2016,36(11):3651-3656. Hu Menghan, Dong Qingli, Liu Baolin. Comparison of predicting blueberry firmness and elastic modulus with hyperspectral reflectance, transmittance and interactance imaging modes[J]. Spectroscopy and Spectral Analysis, 2016, 36(11): 3651-3656. (in Chinese with English abstract)
[7] Sun Jun, Tang Kai, Wu Xiaohong, et al. Nondestructive identification of green tea varieties based on hyperspectral imaging technology[J/OL]. Journal of Food Process Engineering, 2018, 41(5): e12800.
[8] Wang Nannan, Sun Dawen, Yang Yichao, et al. Recent Advances in the application of hyperspectral imaging for evaluating fruit quality[J]. Food Analytical Methods, 2016, 9(1): 178-191.
[9] Li Bo, Cobo-Medina Magdalena, Lecourt Julien, et al. Application of hyperspectral imaging for nondestructive measurement of plum quality attributes[J]. Postharvest Biology and Technology, 2018, 141: 8-15.
[10] 李瑞,傅隆生. 基于高光谱图像的蓝莓糖度和硬度无损测量[J]. 农业工程学报,2017,33(增刊1):362-366. Li Rui, Fu Longsheng. Nondestructive measurement of firmness and sugar content of blueberries based on hyperspectral imaging[J]. Transactions of the Chinese Society of Agricultural Engineering(Transactions of the CSAE), 2017, 33(Supp1): 362-366. (in Chinese with English abstract)
[11] 郝勇,孙旭东,潘圆媛,等. 蒙特卡罗无信息变量消除方法用于近红外光谱预测果品硬度和表面色泽的研究[J]. 光谱学与光谱分析,2011,31(5):1225-1229. Hao Yong, Sun Xudong, Pan Yuanyuan, et al. Detection of firmness and surface color of pear by near infrared spectroscopy based on monte carlo uninformative variables elimination method[J]. Spectroscopy and spectral analysis, 2011, 31(5): 1225-1229.(in Chinese with English abstract)
[12] Sun Meijun, Zhang Dong, Liu Li, et al. How to predict the sugariness and hardness of melons: A near-infrared hyperspectral imaging method[J]. Food Chemistry, 2017, 218: 413-421.
[13] 孙静涛,马本学,董娟,等. 高光谱技术结合特征波长筛选和支持向量机的哈密瓜成熟度判别研究[J]. 光谱学与光谱分析,2017,37(7):2184-2191. Sun Jingtao, Ma Benxue, Dong Juan, et al. Study on maturity discrimination of hami melon with hyperspectral imaging technology combined with characteristic wavelengths selection methods and SVM[J]. Spectroscopy and Spectral Analysis, 2017, 37(7): 2184-2191. (in Chinese with English abstract)
[14] 王世芳,宋海燕,张志勇,等. 基于近红外光谱的冷藏期番茄果实质地定量分析模型[J]. 农产品加工,2017(3):16-19. Wang Shifang, Song Haiyan, Zhang Zhiyong, et al. The quantitative analysis model of fruit texture of tomato in cold storage period based on near infrared spectroscopy[J]. Farm Products Processing. 2017 (3): 16-19. (in Chinese with English abstract)
[15] 王世芳,宋海燕,张志勇,等. 基于近红外光谱的常温贮藏期番茄果肉硬度动力学模型[J]. 食品与发酵工业,2017,43(9):83-86. Wang Shifang, Song Haiyan, Zhang Zhiyong, et al. Kinetic model of tomato hardness during storage at room temperature by near-infrared spectroscopy[J]. Food and Fermentation Industries. 2017, 43(9): 83-86. (in Chinese with English abstract)
[16] 王凡,李永玉,彭彦昆,等. 便携式番茄多品质参数可见/近红外检测装置研发[J]. 农业工程学报,2017,33(19):295-300. Wang Fan, Li Yongyu, Peng Yankun, et al. Development of portable device for simultaneous detection on multi-quality attributes of tomato by visible and near-infrared[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2017, 33(19): 295-300. (in Chinese with English abstract)
[17] 张若宇. 番茄可溶性固形物和硬度的高光谱成像检测[D]. 杭州:浙江大学,2014. Zhang Ruoyu. Detection of Soluble Solids Content and Firmness of Tomato Using Hyperspectral Imaging[D]. Hangzhou: Zhejiang University, 2014. (in Chinese with English abstract)
[18] 王虹,周心智,杨丽,等. 质构仪测定番茄硬度方法的比较[J]. 南方农业,2009,3(6):s40-43. Wang Hong, Zhou Xinzhi, Yang Li, et al. Comparison of methods for determining tomato hardness by texture analyzer[J]. South China Agriculture, 2009, 3(6): 40-43. (in Chinese with English abstract)
[19] Yun Yonghuan, Li Hongdong, E. Wood Leslie R., et al. An efficient method of wavelength interval selection based on random frog for multivariate spectral calibration[J]. Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy, 2013, 111: 31-36.
[20] Wang Hongbo, Ren Xuena, Tu Xuyan. Bee and frog co-Evolution algorithm and its application[J]. Applied Soft Computing, 2017, 56: 182-198.
[21] Sharma Tarun Kumar, Pant Millie. Identification of noise in multi noise plant using enhanced version of shuffled frog leaping algorithm[J]. International Journal of System Assurance Engineering and Management, 2018, 9(1): 43-51.
[22] Sharma Tarun Kumar, Pant Millie. Opposition based learning ingrained shuffled frog-leaping algorithm[J]. Journal of Computational Science, 2017, 21:307-315.
[23] Pan Leiqing, Sun Ye, Xiao Hui, et al. Hyperspectral imaging with different illumination patterns for the hollowness classification of white radish[J]. Postharvest Biology and Technology, 2017, 126: 40-49.
[24] Cheng Junhu, Sun Dawen. Rapid and non-invasive detection of fish microbial spoilage by visible and near infrared hyperspectral imaging and multivariate analysis[J]. LWT-Food Science and Technology, 2015, 62(2): 1060-1068.
[24] Zheng Xiaochun, Li Yongyu, Wei Wensong, et al. Detection of adulteration with duck meat in minced lamb meat by using visible near-infrared hyperspectral imaging[J]. Meat Science, 2019, 149: 55-62.
Detection of tomato hardness based on hyperspectral technology and modified interval random frog algorithm
Long Yan1,2.3, Lian Yaru1,2,3, Ma Minjuan1,2,3, Song Huaibo1,2,3, He Dongjian1,2,3
(1.7121002.7121003.712100)
Tomato has become the most cultivated and consumed vegetable crop in the world, and China has become one of the largest producers and consumers of tomatoes in the world. The pectin content in cell wall of tomato during ripening is closely related to fruit hardness, which is one of the important indicators to determine the maturity and reflect the quality of tomato. The requirement of tomato maturity classification and evaluation promotes the development of non-destructive, fast and accurate detection methods of tomato hardness.Hyperspectral imaging integrates spectroscopy and imaging technology in an analysis system, which transfers tomato maturity assessment from subjective, manual classification and measurement methods. Hyperspectral imaging has been widely used in the rapid acquisition of information to classify, detect or identify the quality of various fruits. A novel method for tomato hardness detection based on hyperspectral imaging and modified interval Random Frog was proposed in this paper. Firstly, hyperspectral images of 120 tomato samples in different mature periods were captured by hyperspectral imaging system covering near-infrared region (865.11nm-1 711.71nm). And the hardness data of tomato was obtained by texture analyzer. Secondly, the spectral data were pretreated by multiplicative scatter correction (MSC) and normalized preprocessing to eliminate noise and improve signal-to-noise ratio. The validity of the characteristic wavelength plays a crucial role in the prediction performance of the model. Therefore, we need an effective method to extract the effective wavelength to improve the accuracy of the model. Interval random frog (iRF) algorithm considers all possible spectral wavelengths and ranks all the wavelengths based on selected probability. But one of the disadvantages of this method is large number of iterations and slow convergence. In view of above disadvantages, the traditional iRF algorithm was optimized in terms of constructing initial variable subset method. A modified interval Random Frog (miRF)was proposed to extract the characteristic wavelength effectively. Finally, a prediction model was developed based on partial least squares regression (PLSR) method to detect tomato hardness. The results indicated that the convergence efficiency and accuracy of miRF has a significantly improvement compared with the iRF method. The iRF has selected 100 feature bands, accounting for 40.65% of the full band, and its runtime was 32.1min. miRF has selected 47 feature bands, accounting for 19.1% of the full band, and its runtime was 1.6 min. It can be seen that miRF greatly reduces the running time of the algorithm. The characteristic wavelengths selected by iRF and miRF methods were mainly distributed in 1 582 nm-1 655 nm, followed by 1 160 nm-1 190 nm and 1 353 nm-1 383 nm, indicating that above regions were sensitive bands to tomato hardness. In order to prove the effectiveness of the proposed algorithm, the results of miRF-PLSR were compared with those of iRF-PLSR and SPA-PLSR. The prediction set correlation coefficients (R) of the SPA-PLSR model and the iRF-PLSR model were 0.803 9 and 0.936 6 respectively. And theRof miRF-PLSR model was 0.968 5. The root mean square error (RMSEP) of the SPA-PLSR model and the iRF-PLSR model were 0.007 7 kg/mm2and 0.004 4 kg/mm2respectively. And the RMSEP of miRF-PLSR model was 0.004 0 kg/mm2. The experiments results show that the miRF-PLSR model has the best prediction results in all models.
spectrum analysis; algorithms; models; hyperspectral technology; tomato; hardness; characteristic wavelength; miRF(modified interval Random Frog)
10.11975/j.issn.1002-6819.2019.13.032
S37;TP391
A
1002-6819(2019)-13-0270-07
2019-01-21
2019-05-29
陕西省农业科技创新与攻关(2016NY-157);中央高校基本科研业务费专项(2452016077)
龙 燕,副教授,主要从事农产品无损检测技术、生物图像与计算机视觉方面的研究。Email:longyan@nwsuaf.edu.cn
龙 燕,连雅茹,马敏娟,宋怀波,何东键.基于高光谱技术和改进型区间随机蛙跳算法的番茄硬度检测[J]. 农业工程学报,2019,35(13):270-276. doi:10.11975/j.issn.1002-6819.2019.13.032 http://www.tcsae.org
Long Yan, Lian Yaru, Ma Minjuan, Song Huaibo, He Dongjian.Detection of tomato hardness based on hyperspectral technology and modified interval random frog algorithm[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2019, 35(13): 270-276. (in Chinese with English abstract) doi:10.11975/j.issn.1002-6819.2019.13.032 http://www.tcsae.org