CAJ | 학술논문

为了实现对造纸原料的快速准确判别，收集了5种共40份原料品种的近红外光谱数据。通过M A F和一阶导数方法进行光谱数据预处理，用主成分方法对光谱数据进行压缩降维，分别利用Fisher算法和BP人工神经网络来建立原料近红外光谱判别模型，并对两种判别模型进行比较。结果表明：两种模型都能较好地进行造纸原料的近红外判别，且BP人工神经网络比Fisher判别函数在容错性上表现得更为优越，建立的模型用于种类判别时表现得更为稳健。
위료실현대조지원료적쾌속준학판별，수집료5충공40빈원료품충적근홍외광보수거。통과M A F화일계도수방법진행광보수거예처리，용주성분방법대광보수거진행압축강유，분별이용Fisher산법화BP인공신경망락래건립원료근홍외광보판별모형，병대량충판별모형진행비교。결과표명：량충모형도능교호지진행조지원료적근홍외판별，차BP인공신경망락비Fisher판별함수재용착성상표현득경위우월，건립적모형용우충류판별시표현득경위은건。
Forty examples of ifve kinds of papermaking raw material near infrared spectroscopy are prepared to discriminate the kind of papermaking raw material quickly and accurately. The NIRS of examples are pretreated by moving average filter and first derivative. Fisher algorithm and BP-ANN are introduced to establish the raw material discriminant model after using principle components analysis to compress spectral data and the two discriminant models are compared. The results indicate that two discriminant models can discriminate the kind of raw material precisely and BP artiifcial neural network is even more advantageous than the Fisher discriminant function in the performance of fault tolerance. What’s more, BP artiifcial neural network discriminant model is more robust than Fisher discriminant model in discriminating the kind of papermaking raw material.