基于高光谱成像技术的菜青虫生命状态检测研究

采用近红外高光谱成像技术对菜青虫的存活与死亡状态进行了研究，通过提取菜青虫不同状态的光谱信息，建立判别分析模型。以不同预处理方法对所提取的951.5～1 649.2 nm光谱进行预处理，并建立偏最小二乘判别分析(partial least square-discriminant analysis, PLS-DA)模型对菜青虫的生死状态进行判别分析，判别正确率接近或达到100%。用移动平均(moving average，MA)5点平滑光谱分别采用连续投影算法(successive projections algorithm, SPA)以及加权回归系数(weighted regression coefficient，B_w)分别选取了17和20个特征波长进行生与死状态的判别。基于特征波长建立了PLS-DA, K最邻近节点算法(K-nearest neighbor，KNN)，BP神经网络(back propagation neural network，BPNN)以及支持向量机(support vector machine，SVM)模型，判别正确率接近100%。结果表明采用近红外高光谱成像技术对菜青虫生命状态的研究是可行的，为作物虫害的快速诊断提供了新方法。

Study on the Live State of Pieris Rapaes Using Near Infrared Hypserspectral Imaging Technology

Near-infrared hypserspectral imaging technology was applied for the discrimination of a variety of life states, the judgment of being alive or death. Discrimination models were built based on spectral data of pieris rapaes acquired during different life states. The wavelengths from 951.5 to 1 649.2 nm were used for analysis after the removal of spectral region with obvious noises at the beginning and the end. And the spectra data of 951.5～1 649.2 nm were preprocessed by different pretreatment methods. To discriminate the state of being alive or death of pieris rapaes, discrimination models were built based on the spectral data processed by different pretreatment methods. Results showed that the discriminant accuracy can approach or attain 100%. Thus the method was proved to be useful for the discrimination of the state of being alive or death of pieris rapaes. After the spectral data were preprocessed by moving average (MA) algorithm, 17 characteristic wavelengths were extracted based on weighted regression coefficient (B_w) and 20 were extracted based on successive projections algorithm (SPA) to identify the state of being alive or death of pieris rapaes. Four classification methods based on characteristic wavelengths, including partial least squares-discriminant analysis (PLS-DA), K-nearest neighbor algorithm (KNN), back propagation neural network (BPNN) and support vector machine (SVM) were used to build discriminant models for identifying the state of being alive or death of pieris rapaes. The discriminant accuracy all can approach or attain 100%.