基于卷积神经网络算法的微藻识别及生物量预测

Recognition and Biomass Prediction of Microalgae Based on Convolutional Neural Network Algorithms

微藻培养的生物量产率直接决定了其能源利用效率。传统的生物量测定需要依靠离线人工检测分析手段,不可避免地产生巨大的人力浪费和时间成本。基于ResNet、MobileNet以及EfficientNet三种深度卷积神经网络模型,将图像分析与微藻培养相结合,提出了一种能够识别藻种类别,同时直接通过图像信息拟合微藻图像-浓度的非线性映射关系并精确预测微藻生物量的检测方法。研究表明,三种模型对三种实验藻种(小球藻、红藻以及螺旋藻)的分类识别准确率均超过99%。其中,红藻得益于其颜色特征,具有最佳的预测表现。而ResNet对藻生物量预测性能最优,三种藻生物量在该模型下的预测决定系数R2分别为0.766 4、0.962 8和0.921 5。该方法基本满足了微藻培养过程中藻生物量的监测需求,为微藻能源化的工业过程监测提供了一种具有潜力的技术方案。

The biomass yield of microalgae culture directly determines its energy utilization efficiency.Traditional biomass measurement relies on offline manual detection and analysis,inevitably resulting in significant labor wastage and time costs.A detection method that combines image analysis with microalgae cultivation is introduced.Utilizing three deep convolutional neural network models—ResNet,MobileNet,and EfficientNet—the method enables the online identification of algae species and directly fits a nonlinear mapping between microalgae images and concentration to predict microalgal biomass accurately.The study demonstrates that the classification accuracy of these three models for three experimental algae species(Chlorella,Rhodophyta,and Spirulina)exceeds 99%.Rhodophyta,owing to its color characteristics,exhibits the best predictive performance.ResNet showed the optimal performance in predicting algal biomass,with the determination coefficients(R2)for the biomass of the three algae being 0.7664,0.9628,and 0.9215,respectively.This method essentially meets the monitoring requirements for algal biomass during microalgae cultivation and provides a highly promising technical solution for the industrial process monitoring of algae energy conversion.