种子是农业生产中最重要的生产资料之一，优质的种子可以提高田间作物产量，其中对于种子活力的评估是衡量种子品质的重要指标之一。传统对于种子活力的检测，通过发芽试验、四唑试验（化学法）和电导率法等测定种子的某些生理特性来评价其活力特性，但会对种子造成不可恢复的损伤。相较于传统检测方法，如今快速发展的光学检测手段开始应用于农业检测中，常用近红外光谱、高光谱等无损检测技术检测种子活力。本文针对以上问题采用光声光谱检测技术对大豆种子活力进行快速检测。主要研究内容如下：

（1）基于Comsol多物理场仿真对大豆种子在不同调制频率下光声信号的强度进行仿真研究。通过Comsol仿真光声效应下大豆种子，首先通过测量大豆种子得到其均值几何数据，基于此数据建立大豆几何模型。然后将氦气作为光声池内部填充介质，并对大豆种子热粘性声学相关系数进行补充定义。其次，对光声池内大豆种子及介质进行网格剖分。基于声波在固体和气体边界处传播时，固体表面会产生粘性边界层和热边界层，因此在大豆种子表面进行网格细化。通过在实际实验中麦克风所在位置添加仿真时探针位置，得出不同调制频率光下大豆光声信号声压级变化规律。为后续实验检测不同调制频率下大豆种子光声信号作出仿真验证。

（2）基于光声光谱技术的大豆种子活力分类建模。以3种颜色（黄色、青色、黑色）、3个自然老化年份（2017、2018、2019）的6个品种（古田、滇豆1号、中黄1号、淮阳青大豆、涡阳黑豆、延药豆）为研究对象，对6种大豆种子在（100、200、300、400、500、800和1000Hz）调制频率下进行光声采集，该实验样本均为自然老化大豆种子。首先，通过集合经验模态分解（Ensemble Empirical Mode Decomposition，EEMD）去噪预处理6类大豆种子光声数据。以活力指数将样本划分为无活力、低活力和高活力大豆种子。然后，使用主成分分析(Principal Component Analysis，PCA) 降维特征提取后的光声数据分别建立随机森林（Random Forest，RF）种子活力分类模型、概率神经网络（Probabilistic Neural Network，PNN）种子活力分类模型和深度置信网络（Deep Belief Network，DBN）种子活力分类模型，对单一品种大豆和混合大豆活力进行分类。实验结果表明，针对单一品种的大豆种子活力分类，200Hz下分类结果较优。最优分类模型DBN活力分类准确率达90.91%。针对最优调制频率200Hz下的混合大豆种子活力分类，得出最优活力分类模型PCA-DBN模型，分类准确率达95.83%，实现对大豆种子活力的精确分类。

（3）基于光声光谱技术的大豆种子生活力预测建模。对上述采集和去噪后的各调制频率下的大豆光声数据，首先利用PCA、竞争性自适应重加权算法（Competitive adaptive reweighted sampling， CARS）和连续投影算法（Successive Projections Algorithm，SPA）3种特征提取方法提取光声特征波长。然后，分别通过广义神经网络（Generalized Regression Neural Network，GRNN）、偏最小二乘回归法（Partial Least Squares Regression，PLSR）和支持向量回归（Support Vector Regression，SVR）进行大豆种子生活力预测建模。研究表明，200Hz为最优扫描频率。SPA结合SVR建立的模型为最优生活力预测模型。6个品种独立生活力预测模型的相关系数分别0.9755，0.9549，0.9487，0.9846，0.9815，0.9864。最优频率下混合种子SVR生活力预测模型相关系数为0.7884。为进一步优化混合大豆种子生活力预测模型提出遗传算法优化的支持向量回归模型（GA-SVR），混合大豆种子GA-SVR模型预测相关系数为0.9511，预测精度得到提高。并且为实现对新品种大豆种子生活力精准预测，将其中5类大豆种子和单独1类大豆种子的光声信息通过迁移学习方式使其具有相似的空间分布特性，迁移成分分析（Transfer Componet Analysis，TCA）后的5种大豆种子建立生活力预测模型为新大豆种子生活力做出预测。迁移后所建模型相关系数为0.9571。

研究表明，通过光声光谱技术可以实现对大豆种子活力和生活力的检测，活力分类模型精度在95%以上。生活力预测模型相关系数在0.95以上。

Seeds are one of the most important means of production in agricultural production. High quality seeds can improve the yield of field crops. The evaluation of seed vitality is one of the important indicators to measure seed quality. In the traditional detection of seed vitality, some physiological characteristics of seeds are measured by germination test, tetrazole test (chemical method) and conductivity method to evaluate their vitality characteristics, but it will cause irreparable damage to seeds. Compared with the traditional detection methods, the rapidly developing optical detection methods have been applied to agricultural detection. Near infrared spectroscopy, hyperspectral and other nondestructive detection technologies are commonly used to detect seed vitality. Aiming at the above problems, this paper uses photoacoustic spectrum detection technology to quickly detect the vitality of soybean seeds. The main research contents are as follows:
(1) Based on COMSOL multi physical field simulation, the intensity of photoacoustic signal of soybean seeds under different modulation frequencies was simulated. The soybean seeds under photoacoustic effect are simulated by COMSOL. Firstly, the mean geometric data are obtained by measuring the soybean seeds, and the soybean geometric model is established based on this data. Then helium was used as the filling medium in the photoacoustic cell, and the thermal viscosity acoustic correlation coefficient of soybean seeds was defined. Secondly, the soybean seeds and medium in the photoacoustic tank are meshed. When the sound wave propagates at the solid and gas boundaries, the solid surface will produce viscous boundary layer and thermal boundary layer, so the mesh refinement is carried out on the surface of soybean seeds. By adding the simulated probe position to the microphone position in the actual experiment, the variation law of soybean photoacoustic signal sound pressure level under different modulation frequency light is obtained. Simulation verification is made for the subsequent experiments to detect the photoacoustic signals of soybean seeds under different modulation frequencies.
(2) Classification and modeling of soybean seed vigor based on photoacoustic spectroscopy. Six varieties (Gutian, diandou No. 1, Zhonghuang No. 1, Huaiyang green soybean, Woyang black bean and Yanyao bean) with three colors (yellow, cyan and black) and three natural aging years (2017, 2018 and 2019) were taken as the research objects. Six kinds of soybean seeds were collected under the modulation frequency of (100, 200, 300, 400, 500, 800 and 1000Hz). All the experimental samples were natural aging soybean seeds. Firstly, six kinds of soybean seed photoacoustic data were preprocessed by ensemble empirical mode decomposition (EEMD). The samples were divided into no vigor, low vigor and high vigor soybean seeds by vigor index. Then, using the photoacoustic data extracted from the dimensionality reduction features of principal component analysis (PCA), the seed vigor classification model of random forest (RF), the seed vigor classification model of probabilistic neural network (PNN) and the seed vigor classification model of deep belief network (DBN) were established to classify the seed vigor of single variety soybean and mixed soybean. The experimental results showed that the classification result of soybean seed vigor under 200Hz was better for a single variety. The accuracy of DBN vitality classification is 90.91%. For the hybrid soybean seed vigor classification under the optimal modulation frequency of 200Hz, the optimal vigor classification model pca-dbn model is obtained, and the classification accuracy is 95.83%.
(3) Prediction and modeling of soybean seed viability based on photoacoustic spectroscopy. Firstly, PCA, competitive adaptive reweighted sampling (cars) and continuous projections algorithm (SPA) are used to extract the photoacoustic characteristic wavelength of soybean photoacoustic data at each modulation frequency after the above acquisition and denoising. Then, the prediction models of soybean seed viability were established by generalized regression neural network (GRNN), partial least squares regression (PLSR) and support vector regression (SVR). The research shows that 200Hz is the optimal scanning frequency. The model established by spa combined with SVR is the optimal vitality prediction model. The correlation coefficients of the independent viability prediction models of the six varieties were 0.9755, 0.9549, 0.9487, 0.9846, 0.9815 and 0.9864 respectively. The correlation coefficient of SVR viability prediction model of mixed seeds at the optimal frequency is 0.7884. In order to further optimize the prediction model of mixed soybean seed viability, a support vector regression model (GA-SVR) optimized by genetic algorithm is proposed. The prediction correlation coefficient of GA-SVR model of mixed soybean seed is 0.9511, and the prediction accuracy is improved. In order to accurately predict the viability of new soybean seeds, the photoacoustic information of five types of soybean seeds and a single type of soybean seeds are made to have similar spatial distribution characteristics through transfer learning. The viability prediction model of five kinds of soybean seeds after transfer component analysis (TCA) is established to predict the viability of new soybean seeds. The correlation coefficient of the model after migration is 0.9571.
The results show that the detection of soybean seed vigor and viability can be realized by photoacoustic spectroscopy, and the accuracy of vigor classification model is more than 95%. The correlation coefficient of vitality prediction model is more than 0.95.

[1] 刘梅芳,樊琦.中国大豆消费、生产和进口现状及存在的问题[J]. 粮食科技与经济,2021,46(06):28-35. DOI:10.16465/j.gste.cn431252ts.20210606.

[2] 石慧. 大豆在美国的引种推广及本土化研究[D].南京农业大学,2018. DOI:10.27244/d.cnki.gnjnu.2018.001050.

[3] 李文彬,于秀琪,刘春宇,谭勇,苗馨卉,宋少忠.农作物种子活力检测方法研究进展[J].农业灾害研究,2021,11(06):7-8+10.

[4] Feng L, Zhu S, Liu F, et al. Hyperspectral imaging for seed quality and safety inspection: A review [J]. Plant methods, 2019, 15(1): 1-25.

[5] Tran V N, Cavanagh A K. Structural aspects of dormancy [M]//Germination and reserve mobilization. Academic Press, 1984: 1-44.

[6] ElMasry G, Mandour N, Wagner M H, et al. Utilization of computer vision and multispectral imaging techniques for classification of cowpea (Vigna unguiculata) seeds[J]. Plant Methods, 2019, 15(1): 1-16.

[7] Xia Y, Xu Y, Li J, et al. Recent advances in emerging techniques for non-destructive detection of seed viability: A review [J]. Artificial Intelligence in Agriculture, 2019, 1: 35-47.

[8] 田忠亮.浅谈活力检测在我国种子质量检测体系中的重要性[J].南方农业,2019,13(09):140-141.DOI:10.19415/j.cnki.1673-890x.2019.09.073.

[9] Kamboj U, Kaushal N, Mishra S, et al. Application of selective near infrared spectroscopy for qualitative and quantitative prediction of water adulteration in milk[J]. Materials Today: Proceedings, 2020, 24: 2449-2456.

[10] 杨渊婷,高光伟.探析近红外光谱分析技术在食品检测中的应用[J].科技资讯,2019,17(13):75+77.DOI:10.16661/j.cnki.1672-3791.2019.13.075.

[11] Al-Amery M, Geneve R L, Sanches M F, et al. Near-infrared spectroscopy used to predict soybean seed germination and vigour [J]. Seed Science Research, 2018, 28(3): 245-252.

[12] Wang Y L, Peng Y K, Zhuang Q B, Zhao X L. Feasibility analysis of NIR for detecting sweet corn seeds vigor [J]. Journal of Cereal Science. 2020, 93.

[13] Chen J, Li M, Pan T, et al. Rapid and non-destructive analysis for the identification of multi-grain rice seeds with near-infrared spectroscopy [J]. Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy, 2019, 219: 179-185.

[14] 金文玲,曹乃亮,朱明东,陈伟,张佩光,赵庆磊,梁静秋,余应弘,吕金光,阚瑞峰.基于近红外超连续激光光谱的水稻种子活力无损分级检测研究[J].中国光学,2020,13(05):1032-1043.

[15] Tigabu M, Daneshvar A, Wu P, et al. Rapid and non-destructive evaluation of seed quality of Chinese fir by near infrared spectroscopy and multivariate discriminant analysis [J]. New Forests, 2020, 51(3): 395-408.

[16] SILVA I. Raman Spectroscopy [M]//Between Making And Knowing: Tools In The History Of Materials Research. 2020: 435-441.

[17] 韩宇,宋少忠,张佳环,谭勇,刘春宇,周云全,曲冠男,韩艳丽,张京,胡玉,孟维实,刘焕军,张一翔,李佳一. 基于拉曼光谱的大豆细菌性病害标志物研究[J].光谱学与光谱分析,2022,42(02):459-463.

[18] 朱培培,田芳明,谭峰,马文宝,严陈慧子.寒地粳稻种子的拉曼光谱鉴别方法研究[J].中国粮油学报,2021,36(07):169-174.

[19] Alvarez J, Martinez-Ramirez S, Martinez E. Magneto-primed triticale seeds studied by micro-raman spectra [J]. Plants, 2021, 10(6): 1083.

[20] Kecoglu I, Sirkeci M, Sen A, et al. Quantification of salt stress in wheat leaves by Raman spectroscopy and machine learning [J]. bioRxiv, 2022.

[21] Zhu M, Huang D, Hu X J, et al. Application of hyperspectral technology in detection of agricultural products and food: A Review[J]. Food Science & Nutrition, 2020, 8(10): 5206-5214.

[22] Zhang T, Wei W, Zhao B, Wang R, et al. A Reliable Methodology for Determining Seed Viability by Using Hyperspectral Data from Two Sides of Wheat Seeds. Sensors. 2018, 18, 813.

[23] Pang L, Wang J, Men S, et al. Hyperspectral imaging coupled with multivariate methods for seed vitality estimation and forecast for Quercus variabilis [J]. Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy, 2021, 245: 118888.

[24] Pang L, Men S, Yan L, et al. Rapid vitality estimation and prediction of corn seeds based on spectra and images using deep learning and hyperspectral imaging techniques[J]. IEEE Access, 2020, 8: 123026-123036.

[25] Baek I, Kim M S, Cho B K, et al. Selection of optimal hyperspectral wavebands for detection of discolored, diseased rice seeds [J]. Applied Sciences, 2019, 9(5): 1027.

[26] Zhu S, Chao M, Zhang J, et al. Identification of soybean seed varieties based on hyperspectral imaging technology [J]. Sensors, 2019, 19(23): 5225.

[27] Yang Y, Chen J, He Y, et al. Assessment of the vigor of rice seeds by near-infrared hyperspectral imaging combined with transfer learning [J]. RSC Advances, 2020, 10(72): 44149-44158.

[28] Ravansari R, Wilson S C, Tighe M. Portable X-ray fluorescence for environmental assessment of soils: Not just a point and shoot method [J]. Environment International, 2020, 134: 105250.

[29] Wei L, GUO Y, Dai D, et al. Rice germination rate detection based on fluorescent spectrometry and deep belief network [J]. Spectroscopy and Spectral Analysis, 2018, 38(4): 1303.

[30] Bamrah R K, Vijayan P, Karunakaran C, et al. Evaluation of x‐ray fluorescence spectroscopy as a tool for nutrient analysis of pea seeds [J]. Crop Science, 2019, 59(6): 2689-2700.

[31] Barboza da Silva C, Oliveira N M, de Carvalho M E A, et al. Autofluorescence-spectral imaging as an innovative method for rapid, non-destructive and reliable assessing of soybean seed quality[J]. Scientific reports, 2021, 11(1): 1-12.

[32] 李菲菲.X射线荧光光谱法检测粮食中镉元素的应用探讨[J]. 粮食与饲料工业, 2020(02):21-25.

[33] 李翠玲,姜凯,马伟,王秀.基于荧光光谱的辣椒种子品种识别[J]. 蔬菜, 2020(09):9-13.

[34] Liu Y, Xu Z, Hu J, et al. Research on quality of agricultural products by terahertz spectroscopy [J]. Laser & Optoelectronics Progress, 2021, 58(1): 0100005.

[35] 陈云锋.转基因水稻种子识别和检测方法研究[D].浙江大学,2020.DOI:10.27461/d.cnki.gzjdx.2020.004337.

[36] 吴静珠,李晓琪,刘翠玲,余乐,孙晓荣,孙丽娟.基于THz-TDS反射成像技术的玉米种子活力无损检测研究[J].光谱学与光谱分析,2020,40(09):2840-2844.

[37] Gong Z, Deng D, Sun X, et al. Non-destructive detection of moisture content for Ginkgo biloba fruit with terahertz spectrum and image: A preliminary study [J]. Infrared Physics & Technology, 2022, 120: 103997.

[38] Zhang J, Yang Y, Feng X, et al. Identification of bacterial blight resistant rice seeds using terahertz imaging and hyperspectral imaging combined with convolutional neural network [J]. Frontiers in Plant Science, 2020, 11: 821.

[39] Yang G, Wang Q, Liu C, et al. Rapid and visual detection of the main chemical compositions in maize seeds based on Raman hyperspectral imaging [J]. Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy, 2018, 200: 186-194.

[40] He X, Feng X, Sun D, et al. Rapid and nondestructive measurement of rice seed vitality of different years using near-infrared hyperspectral imaging [J]. Molecules, 2019, 24(12): 2227.

[41] Zhang Y, Guo W. Moisture content detection of maize seed based on visible/near‐infrared and near‐infrared hyperspectral imaging technology[J]. International Journal of Food Science & Technology, 2020, 55(2): 631-640.

[42] Li Y, Sun J, Wu X, et al. Detection of viability of soybean seed based on fluorescence hyperspectra and CARS‐SVM‐AdaBoost model [J]. Journal of Food Processing and Preservation, 2019, 43(12): e14238.

[43] Hernández-Aguilar C, Domínguez-Pacheco A, Cruz-Orea A, et al. Photoacoustic spectroscopy in the optical characterization of foodstuff: a review[J]. Journal of Spectroscopy, 2019, 2019.

[44] Popa C, Bratu A M, Petrus M, et al. The analysis of lead phytotoxicity in seeds using CO2 laser photoacoustic spectroscopy [J]. Molecules, 2020, 25(7): 1637.

[45] Hernandez-Aguilar C, Dominguez-Pacheco A, Tenango M P, et al. Characterization of bean seeds, germination, and phenolic compounds of seedlings by UV-C radiation [J]. Journal of Plant Growth Regulation, 2021, 40(2): 642-655.

[46] 王新宇,牛鹏帅,卢伟,等. 基于光声光谱深度扫描的单粒玉米种子发芽率无损检测方法[J]. 华南农业大学学报,2020,41(6):119-125.

[47] Lv G, Du C, Ma F, et al. Rapid and nondestructive detection of pesticide residues by depth-profiling Fourier transform infrared photoacoustic spectroscopy [J]. ACS omega, 2018, 3(3): 3548-3553.

[48] Valencia-Hernandez J E, Hernandez-Aguilar C, Dominguez-Pacheco A, et al. Optical absorption spectra of germinated seedlings from seeds exposed to vehicle polluting gases [J]. SN Applied Sciences, 2020, 2(4): 1-10.

[49] Popa C, Bratu A M, Petrus M, et al. The analysis of lead phytotoxicity in seeds using CO2 laser photoacoustic spectroscopy [J]. Molecules, 2020, 25(7): 1637.

[50] Popa C, Petrus M. Heavy metals impact at plants using photoacoustic spectroscopy technology with tunable CO2 laser in the quantification of gaseous molecules [J]. Microchemical Journal, 2017, 134: 390-399.

[51] Rosencwaig A. Photoacoustic spectroscopy[M]//Advances in electronics and electron physics. Academic Press, 1978, 46: 207-311.

[52] DAVIS D M. Fourier transform infrared photoacoustic spectroscopy[D]. Oklahoma State University, 1983.

[53] Palzer S. Photoacoustic-based gas sensing: A review[J]. Sensors, 2020, 20(9): 2745.

[54] Bageshwar D V, Pawar A S, Khanvilkar V V, et al. Photoacoustic spectroscopy and its applications–a tutorial review[J]. Eurasian Journal of Analytical Chemistry, 2017, 5(2): 187-203.

[55] 程刚,曹亚南,田兴,曹渊,刘锟.光声池几何形状对光声光谱检测性能的影响[J].光谱学与光谱分析,2020,40(08):2345-2351.

[56] 赵南,刘阳,赵宁阳,马珑格,李媚聪,焦晶晶,丁琛斐.光声光谱检测系统中光声池模型的建立与优化[J].光子学报,2021,50(07):246-256.

[57] 李泽昊,杨春勇,唐梓豪,彭苗苗,倪文军,郭连波,侯金,陈少平.曲体束腰型光声池的设计及性能分析[J].中国激光,2021,48(01):186-195.

[58] Lv F, Wang X, Zhang M, et al. Determination and analysis for parameters of shape, size, physical and mechanical properties of soybean grains[C]//International Conference on Discrete Element Methods. Springer, Singapore, 2016: 1277-1286.

[59] Huang Z, Zhu H, Yan R, et al. Simulation and prediction of radio frequency heating in dry soybeans[J]. Biosystems Engineering, 2015, 129: 34-47.

[60] Kamizake N K K, Yamashita F, Prudencio S H. Physical alterations of soybean during accelerated and natural aging[J]. Food research international, 2014, 55: 55-61.

[61] 何龙生,李啸天,赵光武,龚淑菲,俞兵桥,王建华,贾良权,祁亨年,曹栋栋.加速老化法在常规水稻种子活力测定中的应用[J].江苏农业科学,2019,47(07):61-64.DOI:10.15889/j.issn.1002-1302.2019.07.015.

[62] Priestley D A, Leopold A C. Lipid changes during natural aging of soybean seeds[J]. Physiologia Plantarum, 1983, 59(3): 467-470.

[63] Mangena P, Mokwala P W. The influence of seed viability on the germination and in vitro multiple shoot regeneration of soybean (Glycine max L.)[J]. Agriculture, 2019, 9(2): 35.

[64] Sheidaei S, Hamidi A, Sadeghi H, et al. Evaluation of initial seed quality and storage conditions on biochemical and physiological changes of soybean seeds[J]. Iranian Journal of Seed Science and Technology, 2020, 9(2): 101-118.

[65] Ebone L A, Caverzan A, Silveira D C, et al. Biochemical profile of the soybean seed embryonic axis and its changes during accelerated aging[J]. Biology, 2020, 9(8): 186.

[66] Xu M, Sun J, Zhou X, et al. Research on nondestructive identification of grape varieties based on EEMD‐DWT and hyperspectral image[J]. Journal of Food Science, 2021, 86(5): 2011-2023.

[67] Gadekallu T R, Rajput D S, Reddy M, et al. A novel PCA–whale optimization-based deep neural network model for classification of tomato plant diseases using GPU[J]. Journal of Real-Time Image Processing, 2021, 18(4): 1383-1396.

[68] Zhang J, Dai L, Cheng F. Corn seed variety classification based on hyperspectral reflectance imaging and deep convolutional neural network[J]. Journal of Food Measurement and Characterization, 2021, 15(1): 484-494.

[69] Denisko D, Hoffman M M. Classification and interaction in random forests[J]. Proceedings of the National Academy of Sciences, 2018, 115(8): 1690-1692.

[70] Liu Z Y, Qi J G, Wang N N, et al. Hyperspectral discrimination of foliar biotic damages in rice using principal component analysis and probabilistic neural network[J]. Precision Agriculture, 2018, 19(6): 973-991.

[71] Li Z, Yang S, An X, et al. Research on Signal Modulation Recognition Method Based on Deep Belief Network[C]//2020 International Conference on Communications, Information System and Computer Engineering (CISCE). IEEE, 2020: 59-62.

[72] Xu X, Ren M, Cao J, et al. Spectroscopic diagnosis of zinc contaminated soils based on competitive adaptive reweighted sampling algorithm and an improved support vector machine[J]. Spectroscopy Letters, 2020, 53(2): 86-99.

[73] 赵静远,熊智新,宁井铭,谢德红.小波变换结合连续投影算法优化茶叶中咖啡碱的近红外分析模型[J].分析科学学报,2021,37(05):611-617.

[74] Cheng J H, Sun D W. Partial least squares regression (PLSR) applied to NIR and HSI spectral data modeling to predict chemical properties of fish muscle[J]. Food engineering reviews, 2017, 9(1): 36-49.

[75] Li W, Yang X, Li H, et al. Hybrid forecasting approach based on GRNN neural network and SVR machine for electricity demand forecasting[J]. Energies, 2017, 10(1): 44.

[76] Xin N, Gu X, Wu H, et al. Application of genetic algorithm‐support vector regression (GA‐SVR) for quantitative analysis of herbal medicines[J]. Journal of Chemometrics, 2012, 26(7): 353-360.

[77] Weiss K R, Khoshgoftaar T M. An investigation of transfer learning and traditional machine learning algorithms[C]//2016 IEEE 28th International Conference on Tools with Artificial Intelligence (ICTAI). IEEE, 2016: 283-290.