CAJ | 학술논문

电子鼻检测的原始特征的数据量很大，一般在进行降低维数的处理前需要对原始特征进行合理的选择。选用最大值( Max)、全段数据平均值( Mean)、响应曲线最大曲率( kmax )、响应曲线的全段积分值( IV)作为4种不同特征参数对感染早疫病病害的番茄苗进行区分效果研究，结果表明在进行PCA和LDA区分时，利用全段数据平均值和响应曲线的全段积分值作为特征参数的效果较好，其次为最大值方法，最差的是响应曲线最大曲率方法；利用BP神经网络( BPNN)和遗传算法BP神经网络( GABPNN)两种识别模式进行预测时，利用全段数据平均值和响应曲线的全段积分值作为特征参数的训练集和预测集的正确率较好，其次为最大值方法，预测结果最差的是响应曲线最大曲率方法。
전자비검측적원시특정적수거량흔대，일반재진행강저유수적처리전수요대원시특정진행합리적선택。선용최대치( Max)、전단수거평균치( Mean)、향응곡선최대곡솔( kmax )、향응곡선적전단적분치( IV)작위4충불동특정삼수대감염조역병병해적번가묘진행구분효과연구，결과표명재진행PCA화LDA구분시，이용전단수거평균치화향응곡선적전단적분치작위특정삼수적효과교호，기차위최대치방법，최차적시향응곡선최대곡솔방법；이용BP신경망락( BPNN)화유전산법BP신경망락( GABPNN)량충식별모식진행예측시，이용전단수거평균치화향응곡선적전단적분치작위특정삼수적훈련집화예측집적정학솔교호，기차위최대치방법，예측결과최차적시향응곡선최대곡솔방법。
In order to improve the calculation efficiency,we should minimize the number of features and try not to lose useful information for classification under the premise of the original features. This paper use the maximum value, average value of all data,maximum curvature of response curve and the intact integral value of response curve as the feature parameters for distinguishing infection early blight disease of tomato seedlings. The results show that the distin-guished effect using average value of all data and the intact integral value of response curve as the feature parameters is the best,and that using maximum value is better. The worse is using maximum curvature of response curve.