Hyperspectral imaging feature of catalase activity in eggplant leaves stressed by grey mold was researched. Hyperspectral imagings of healthy, slight, moderate and heavy infected eggplant leaves were obtained by hyperspectral imaging system across the wavelength region of 400-100nm and diffuse spectral response of objects from hyperspectral imaging was extracted by ENVI software. Then, different preprocessing methods were used to improve the signal noise ratio (SNR) including smoothing, median filter and normalization et al. The models of hyperspectral imaging response and catalase activity were built by the partial least squares regression (PLSR), least squares support vector machines (LS-SVM) and BP neural network (BPNN). The first two latent variables suggested by PLSR model can qualitatively distinguish healthy, slight, moderate and heavy infected eggplant leaves, the coefficient of determination (R2) of BPNN model built by the nine latent variables recommended by PLSR model is 0.8930 and the root mean square error of prediction (RMSEP) is 2.17×103. It demonstrated that catalase activity in eggplant leaves can be effectively detected and disease degree of eggplant leaves stressed by grey mold can also be effectively distinguished by the hyperspectral imaging technique.%对灰霉病胁迫下茄子叶片过氧化氢酶(CAT)活性的高光谱图像特征进行了研究.采用380～1 030 nm范围的高光谱图像摄像仪获取健康、轻度、中度、严重染病茄子叶片的高光谱图像信息,基于ENVI软件处理平台提取高光谱图像中对象的漫反射光谱响应特性,并采用平滑、中值滤波、归一化法等预处理方法提高光谱的信噪比.然后采用偏最小二乘回归(PLSR)、最小二乘支持向量机(LS-SVM)和BP神经网络算法来建立叶片高光谱响应特征与CAT活性之间的关系模型.在PLSR模型中,前2个隐含变量能够实现健康、轻度、中度、严重染病茄子叶片的直观定性区分,而基于PLSR模型推荐的9个隐含变量建立的BP神经网络模型的预测集决定系数R2为0.8930,均方根误差为2.17×103.表明基于高光谱图像特性可以实现灰霉病胁迫下茄子病害程度的有效区分,同时证明基于高光谱图像特性的茄子叶片CAT活性的定量检测是可行的.
展开▼
