应用遗传算法-主成分分析-反向传播神经网络的近红外光谱识别树种效果

以风车木(Conbretum imberbe)和非洲小叶紫檀(Pterocarpus tinctorius Welw)为研究对象,应用LabSpec光谱仪采集光谱样本进行主成分分析(PCA),并运用遗传算法(GA)对主成分进行寻优,分别以未经GA寻优的主成分和经GA寻优的主成分作为反向传播(BP)神经网络输入量,测试了BP神经网络识别两种树种的效果。结果表明:寻优前,获得高识别率的主成分区间较窄,仅有5种情况识别效果理想,此种情况不利于主成分数的恰当选择;寻优后,获得高识别率的主成分区间较宽,从前6到前17有12种情况可供选择,此种情况更利于主成分的合理选择;寻优后的识别率比寻优前高,且稳定性较好。利用近红外光谱,依据GA-PCA-BP神经网络方法识别树种是一种理想的方法。

Using Near Infrared Spectrum to Identify Tree Species by GA-PCA-BP Neural Network

With Conbretum imberbe and Pterocarpus tinctorius Welw,the spectral samples were collected by labspec spectrometer for principal component analysis,and genetic algorithm was used for optimization.The principal components that were not optimized by GA and the principal components optimized were both taken as the input of Back-Propagation(BP)neural network,respectively,to test the effect of BP neural network in identifying two species.The results show that before optimization,the principal component interval with high recognition rate was narrow at this time,and the recognition effect was ideal in only 5 cases.However,this situation is not conducive to the proper selection of PCN.After optimization,the interval of principal components with high recognition rate is wider.There are 12 choices from the first 6 to the first 17 occasions,which is obviously more conducive to the reasonable choice of principal components.The recognition rate after optimization is higher than that before optimizing,so the stability is better.Therefore,near infrared spectroscopy with GA-PCA-BP neural network is an ideal method for tree species identification.