消费级近红外相机的水稻叶片叶绿素(SPAD)分布预测

便捷可靠的作物营养诊断是作物科学施肥管理的基础，也是精准农业的核心。叶绿素含量是作物氮营养含量的重要指标。以水稻叶片为研究对象，用改造后的普通单反相机搭载滤波片的方式拍摄叶片的可见光和中心波长为650，680，720，760，850和950 nm多个波段的近红外图像，获取不同波段的相对反射率值，通过可见光与多个近红外波段结合的回归分析与比较，筛选出精度较高且稳定的模型。经过对比相机三个成像通道，R通道与叶绿素含量(SPAD值)的相关性要高于B和G通道。实验结果表明，植被指数GVI最能反映作物的生长状况，近红外波段760 nm对SPAD值的预测效果最好，最小二乘支持向量机法结合多个植被指数建模的预测精度R2为0.831 4，取得了较为理想的效果。同时使用高光谱成像仪采集水稻叶片的高光谱影像，对比消费级近红外相机成像方式下与高光谱成像方式下得到的植被指数多因子预测模型精度，两者相当。实验证明消费级近红外相机能够获得与高光谱成像仪相近的叶绿素含量估测结果。

Research on the Chlorophyll Content (SPAD) Distribution Based on the Consumer-Grade Modified Near-Infrared Camera

Convenient and reliable crop nutrition diagnosis methods is basis of scientific crop fertilizer management and the core of the precision agriculture, and chlorophyll content is an important index of crop nitrogen nutrition content. In this research, the research object was rice leaf, and visible image and the center wavelength of 650, 680, 720, 760, 850 and 950 nm near infrared image were captured by transformed ordinary camera and filters. Then the relative reflectance values of different wave band were acquired. After regression analysis with visible-band and near-infrared band combined, the high precision and stable models were selected. Compared with the three imaging channels of camera, the correlation between chlorophyll content (SPAD value) and R channel was higher than B, G channels. Results showed that in the comparison of vegetation indexes, GVI can best reflect growth status of crops, and 760 nm has become the best near-infrared band in SPAD prediction. The model prediction accuracy R2 of the least squares support vector machine method combined with multiple vegetation index was 0.831 4, while ideal result had been achieved. Meanwhile, hyperspectral image of rice leaf was captured by hyperspectral imager. Compared the two imaging modalities, the multi factor prediction model based on vegetation index has the same precision. Experiments proved that consumer-grade near infrared camera could gain similar estimation result of chlorophyll content as hyperspectral imager.