Acousto-optic tunable filter(AOTF)hyper-spectral imaging system with acousto-optic tunable filter are used to collect the hyper-spectral images of tomato seedlings'various parts,such as main stem,new leaves and old leaves. In this experiment,tomato seedlings were cultivated at 10 different nitrogen-treatment levels. The spectral information from 615 nm to 1000 nm are extracted from the hyper-spectral imaging by template matching preprocessing. Competitive adaptive reweighted sampling (CARS )method is adoptded to select feature. The experimental results show that the performance of nitrogen quantitative model built by the leaves and stems set is the worst,that of the model built by the leaves set is better,and that of the model built only by the old leaves is the best. The best model is established by 5 fold partial least square (PLS )regression method on the base of 10 wavelength points selected by CARS including 703 nm,906 nm and so on. The best principal component number is 9,the determine coefficient R2 is 0. 95 and the root mean square error of cross validation(RMSECV)is 0. 08. Therefore,it is possible to monitor the nitrogen content of tomato seedlings with the AOTF hyper-spectral imaging technology. The old leaves with relative stability of old leaf nitrogen content should be recommended to build quantitative analysis model,and the preferred feature wavelength for nitrogen analysis is 703 nm and 906nm.%使用声光可调谐滤光器(AOTF)型高光谱成像仪采集10种不同氮素处理水平下的西红柿苗各部位(老叶、嫩叶和主茎)的高光谱图像,经模板匹配预处理后提取615~1000 nm高光谱信息,采用竞争自适应重采样(CARS)方法优选西红柿苗氮素含量表达的特征部位和特征波段.实验结果表明:综合番茄苗叶茎的光谱信息建立的氮素定量模型性能最差,由番茄苗老叶和嫩叶建立的氮素定量模型性能较好,由番茄苗老叶建立的氮素定量模型性能最佳,在挑选的703,906 nm等10个波长点的基础上,采用5折—偏最小二乘交互校验方法建立的氮素定量分析模型的决定系数R2达到了0.95,交叉校验标准差(RMSECV)为0.08.因此,采用AOTF高光谱成像技术定量测量番茄苗氮素含量是可行的,应以氮素含量相对稳定的老叶建立定量分析模型,703,906 nm为氮素分析的首选特征波长.
展开▼
