基于高光谱的微藻生物膜生长预测模型研究

Study on prediction model of microalgae biofilm development based on hyperspectral imaging

为了快速无损地获取微藻生物膜生长信息,本文利用高光谱成像技术在线原位监测了不同时期微藻生物膜的特征光谱,提取了25个光谱特征参数;利用相关系数对光谱特征参数与生物膜量进行了分析,提取了相关系数绝对值较大的四个特征参数:(SDr-SDy)/(SDr-SDy)、SDr、OSAVI、NDVI;同时对比分析了单一特征的曲线拟合模型、BP神经网络模型及多特征参数曲线拟合模型对生物膜生长预测所需时间及精度。结果表明:单一特征拟合模型预测精度的SDr参数为5.12%,预测时间为0.018s;BP神经网络模型预测的精度为2.68%,预测时间为0.873s;多特征参数曲线拟合模型的预测精度提高到3.01%,预测时间缩短至0.024s。实验结果及理论分析表明多光谱特征参数拟合模型对生物膜量的预测较好,可为微藻生物膜高效培养提供参考。

In this work, to obtain the information of microalgae biofilm development in real time, quickly and nondestructively, the hyperspectral imaging is used to monitor the characteristic spectrum of microalgae biofihn at different culture times. We adopt 25 spectral characteristic parameters and analyze coefficient of association between the spectral characteristic parameters and biofilm biomass;we also choose four characteristic parameters with high absolute values from the 25 spectral characteristic parameters as follows ： （SDr-SDy） / （ SDr-SDy）, SDr, OSAVI, NDVI. In addition, we comparatively analyze the prediction time and accuracy of biofilm growth by using the curve fitting model of single feature, BP neural network model and multi-characteristic parameters curve fitting model. We discovered that the SDr parameter in single feature fitting model prediction accuracy is 5.12%, prediction time is 0.018s; prediction accuracy of BP neural network model is 2.68% and prediction time is 0. 873s; prediction accuracy of multi-characteristic parameters curve fitting model reaches to 3.01% and prediction time of the model reduces to 0.019s. The facts show that multi-spectral feature fusion model has advantages in prediction biofilm growth, the experimental results can provide a reference for efficient cultivation of microalgae biofilm.