CAJ | 학술논문

为了实现彩色扫描仪的光谱特征化，采用一种GA修正的BP神经网络与PCA相结合的方法对其进行研究。首先，通过主成分分析，对训练样本的光谱反射率进行降维，以RGB信号和降维后的光谱数据作为输入、输出变量进行GA-BP神经网络的建模，对任意RGB信号都可以通过模型得到其低维光谱信号；再通过主成分分析重构光谱反射率，由此实现RGB信号对光谱反射率的重构，即实现扫描仪的光谱特征化。实验结果表明，GA的优化有效地改善了BP神经网络的极值问题，提高了模型的预测精度，PCA在不影响模型精度的同时提高了模型的效率。由此说明，所提出的模型能够满足扫描仪光谱特征化的需求。
위료실현채색소묘의적광보특정화，채용일충GA수정적BP신경망락여PCA상결합적방법대기진행연구。수선，통과주성분분석，대훈련양본적광보반사솔진행강유，이RGB신호화강유후적광보수거작위수입、수출변량진행GA-BP신경망락적건모，대임의RGB신호도가이통과모형득도기저유광보신호；재통과주성분분석중구광보반사솔，유차실현RGB신호대광보반사솔적중구，즉실현소묘의적광보특정화。실험결과표명，GA적우화유효지개선료BP신경망락적겁치문제，제고료모형적예측정도，PCA재불영향모형정도적동시제고료모형적효솔。유차설명，소제출적모형능구만족소묘의광보특정화적수구。
To achieve spectral characterization of color scanners, a spectral characterization model based on GA-BP and PCA was proposed. Firstly, the dimension of spectral reflectance was reduced by PCA. The GA-BP neural network model was built with input of variables of RGB signal and output of variables of low dimensional spectrum signal. Any low dimensional spectrum signal could be got by this model with any input RGB signal, while the spectral reflectance could be reconstructed by PCA. The spectral characteristics of color scanners were achieved. Experimental results show that the extremum problem of BP neural network could be effectively improved by GA. PCA could improve the operating efficiency of the model under the circumstances of maintaining accuracy. This implied it was a high-precision color scanner characte-ristic model.