人工神经网络参数调整对生物光谱识别的影响

The Effect of Adjusting the ANN′s Parameters on the Identifing of Biology Spectrum

本文介绍了一种利用反向传播人工神经网络 (BP ANN)对生物可见光谱进行识别的方法。利用自组的光纤探头式光谱仪对苹果的疤痕和烂痕微区表面进行部分可见光范围 (5 0 0～ 730nm)的光谱测量 ,采用具有单隐层的BP ANN对光谱数据进行分析 ,以便对生物表面性质实现自动识别。本文着重就神经网络的输出值范围、训练方式、隐层数及不同程度的噪音信号等对神经网络识别精度的影响做了较为详尽的讨论。

A method to identify the visible spectrum of micro areas on the biological surface with artificial neural net (BP-ANN) was introduced in this paper. The visible spectra (from 500 nm to 730 nm) of the micro areas with some rotten or scars on the surface of the apples were measured with fiber sensor spectrometer. A kind of ANN with a single hidden layer was created to identify the characters on the surface automatically. The effects of different ranges of output, different training functions, different number of single hidden layers, and different noise levels on the ANN were also studied.