基于深度学习的藻类混凝去除率预测方法

针对目前水厂应对藻类爆发时混凝剂投加量难以确定的问题，提出一种基于深度学习的藻类混凝去除率预测方法，利用DenseNet卷积神经网络和絮体图像对藻类混凝去除率进行预测，从而对投药量进行调整 . 具体做法是，在实验室条件下对高藻水进行混凝处理，记录混凝处理后的絮体图像和对应的去除率 . 以去除率区间为标签构建絮体图像数据集，利用此数据集对 DenseNet-121模型进行训练 . 结果显示，训练后的模型对测试集的预测准确度达到了 89.5%，与 VGG 和 ResNet模型相比，利用 DenseNet模型对本文建立的数据集进行识别的精确度更高，且在识别去除率在 60%~90% 区间的絮体图像相较于其他两种模型具有明显优势. 同时通过对数据集外的铜绿微囊藻絮体图像进行识别，验证了模型具有良好的泛化性.

Prediction Method of Algae Coagulation Removal Rate Based on Deep Learning

Aiming at the problem that it is difficult to determine the dosage of coagulant for algae outbreak in waterworks, a prediction method of algae coagulation removal rate based on deep learning is proposed. DenseNet convolution neural network and floc image are used to predict the algae coagulation removal rate, so as to adjust the dosage.The specific method is to coagulate the high algae water under laboratory conditions, and record the floc image and the corresponding removal rate range after coagulation. The floc image data set was constructed with the removal rate interval as a label, and the DenseNet-121 model was trained with this data set. The results show that the prediction accuracy of the trained model for the test set reaches 89.5%. Compared with VGG and RESNET convolution neural network, the DenseNet convolution neural network has higher recognition accuracy on the data set established in this paper and has obvious advantages over the other two models in identifying floc images with a removal rate of 60%~90%. At the same time, the recognition of Microcystis aeruginosa floc image outside the data set verifies the good generalization of the model.