【摘要】 穗颈瘟的发生会导致稻米产量降低和品质下降,对穗颈部发生侵染但并未引起倒伏的水稻(D)、穗颈部侵染严重已发生倒伏的水稻(L)和正常水稻(H)进行准确地识别和区分是采取病虫害防治措施和灾害评估的基础。本研究选用水稻黄熟期田间冠层测定的27个H的冠层样本、9个D的冠层样本和10个L的冠层样本数据,并以这些样本数据的红边斜率、红边面积、绿波峰值和绿峰面积等4个高光谱变量作为分析数据,分别运用概率神经网络和系统聚类法进行分类识别,识别精度可以分别高达到93.5%和91.3%,显然,在进行分类识别时概率神经网络这种新方法优于传统的系统聚类法。研究表明,概率神经网络具有更为强大的分类功能,应用于穗颈瘟的高光谱识别,可以实现对D、L和H精确分类,能够补充和完善传统的肉眼观测。更多还原

【Abstract】 The incidence of rice panicles blast led to the decrease of yield and quality,precise differentiation of disease severe rice that their necks were contagious(D),lodging rice that their panicles were severely contagious(L) and healthy rice(H) was the basis of disease and pest prevention measures and damage assessment.This study adopted 27 samples of H’s canopy,9 samples of D’s canopy and 10 samples of L’s canopy,and their red edge,red edge area,green peak and green peak area parameters were treated.Samples were analyzed by probabilistic neural network(PNN)and the hierarchical cluster,and the discrimination accuracy of D,L and H were separately as high as 93.5% and 91.3%.PNN was better than the hierarchical cluster in classification and differentiation.The study demonstrated that PNN had a great classified function.Differentiation of rice panicles blast by hyperspectral remote sensing based on PNN,and this method was feasible to precisely differentiate D and L from Hs,and supplement or perfect the conventional visual survey.更多还原