传统活力检测方法存在操作复杂、耗时长、可重复性差、对种子造成损伤且不可逆等不足，基于此，提出一种基于光声光谱结合最小二乘支持向量机回归(LS-SVR)的稻种活力快速、无损检测方法。在温度为45 ℃、相对湿度为90%的条件下，对南粳46(粳稻)和内5 优8015(杂交稻)进行高温高湿人工老化处理，依次老化0，24，48，72，96 h，获得不同活力的稻种；采集2类稻种光声光谱数据，总计100份，其中校正集样本60个，预测集样本40个；采用小波包对原始光谱数据进行预处理，通过协方差分析和主成分分析(PCA)对光谱进行降维；分别通过偏最小二乘回归(PLSR)、反向传播神经网络(BPNN)和LS-SVR 建立稻种活力预测模型。其中，采用协方差分析结合LS-SVR 建立的模型性能最优，该模型不仅适用于单一稻种，而且适用于不同种类稻种活力的预测。研究表明，采用光声光谱技术结合LS-SVR 对稻种活力进行测定是可行的，且所建模型在稻种活力预测方面具有较好的预测精度，为便携式水稻活力光声光谱仪的研制提供了理论依据。

Considering that the traditional method for vigor test is complex, time- consuming, with poor reproducibility, irreversible, and likely to cause damage to the seed, a rapid and non-destructive testing method of rice seed vigor based on photoacoustic spectroscopy technology and least squares support vector regression (LSSVR) is proposed. Under the condition of temperature of 45 ℃ and relative humidity of 90%, Nanjing46 (japonica rice) and Nei5you 8015 (hybrid rice) rice seeds are artificially aged for 0, 24, 48, 72, 96 h to get rice seeds with different vigor. Spectral data of 100 samples for the two types of rice seeds are collected and divided into a calibration set (60 samples) and a prediction set (40 samples). Wavelet packet analysis is used in spectral preprocessing. Dimensionality reduction of the spectrum data is realized by covariance analysis and principal component analysis (PCA). Prediction models of rice seed vigor are established respectively by partial least squares regression (PLSR), back propagation neural network (BPNN) and LS-SVR. The results show that the optimal model is constructed by covariance analysis and LS-SVR, and the model is not only suitable for single rice species, but also for different types of rice seed in vigor forecast. The experiment shows that it is feasible that rice seed vigor is detected rapidly and non-destructively by photoacoustic spectroscopy technology and LS-SVR. The model has better prediction accuracy in vigor detection, and provides a theoretical basis for the development of portable rice seed vigor spectrometers.