农产品品质光谱成像的空间预测规律

由于受到化学检测手段限制，无法获取光谱图像中的农产品品质在各像素位置处的参考值，因此无法直接验证基于光谱图像得到的农产品品质空间预测结果。为探索基于光谱图像的农产品空间品质检测规律，本文采用计算机生成已知空间品质样本，并分别以固定曝光和变曝光方式采集不同灰度等级标准板的光谱图像。定量分析采集系统误差，借助样本区域光谱与区域指标之间的全波段偏最小二乘（ALL-PLS）和遗传特征波长偏最小二乘（GA-PLS）预测函数，研究出区域指标预测准确时样本空间品质指标的预测精度规律，建立光谱成像空间预测准确度模型。实验结果表明：变曝光方式下的数据采集可以提高光谱图像信噪比，在波段两侧极限处尤为明显；应用区域品质数据预测空间品质分布，空间预测误差主要受光谱图像采集噪声影响：即使区域预测准确，空间预测可能已完全失真。通过衡量数据采集系统误差，可以间接评价农产品品质空间预测的准确度。只有在数据采集系统误差在允许范围内时，光谱成像技术才可准确预测农产品品质的空间分布。

Law of Spatial Prediction of Agro-product Qualities Based on Spectral Imaging

Due to the limitation of chemical testing methods, we cannot get reference value of agro-product qualities in every pixel location and are not directly to verify spatial prediction results of agro-product qualities based on spectral images. In order to research spatial prediction laws of agro-product qualities based on spectral images, the computer was used to generate the samples of known spatial distribution and to acquire the spectral images of different gray-level reflectance standard boards by ways of fixed and variable exposures respectively. The error of data acquisition system was analyzed quantitatively. The spatial prediction accuracy model of spectral imaging was built and the prediction accuracy law of spatial qualities was researched as regional parameter values which could be predicted accurately by means of partial least squares(PLS) prediction methods based on full wavelengths and characteristic wavelengths between regional spectral values and regional parameter values. The results show the signal-to-noise ratio of the spectral image can be improved by changing exposure time and it is more obvious on both sides of the spectral band. When these models that can predict regional parameter values accurately are used to predict space parameter values, their spatial prediction accuracy is affected by error in a similar way. The accuracy of spatial prediction results of agro-product qualities can be evaculated by the way of measuring the error of the data acuisition system. Spectral image detection technology can be used to predict spatial parameter values of agro-product qualities when the error of acquisition system is limited in a specified range.