Application Of Deep Learning In Real-time Monitoring Of Algae In Water Bodies

Eutrophication is a serious problem threatening the global water environment security.Monitoring the eutrophication degree of water body by the concentration of algae cells has been widely applied.However,the existing monitoring methods are still inefficient or costly,and there is an urgent need for a long-term highfrequency and low-cost monitoring method.In recent years,the rapid development of deep learning automatic recognition algorithm can provide efficient technical support for monitoring.This paper proposes a method of algae recognition and quantification based on deep learning to achieve a more efficient and fast real-time monitoring method.For the collection of algae images,we use an in-situ microscopic camera that can be placed in field,and design and develop a multidimensions microscopic camera that can automatically calculate the concentration of algae cells.In order to improve the quality and recognizable probability of algae image,different denoising and contrast enhancement methods are adopted,and these methods are compared.Finally,the image preprocessing scheme which combines OTSU denoising and histogram stretching contrast enhancement is confirmed.For the algae recognation,Mobile Net and Faster R-CNN architecture were used to identify Platymonas helgolandica var.Tsingtaoensis and Noctiluca scintillans.The results show that the highest recognition rate is 96.71%.Using the recognition results and denoising layers,we extract the area of algae cells and calculate the concentration of algae cells in algae solution.The results of traditional phytoplankton counting frame method proved that the counting result is reliable and accurate,and the average error of five concentration gradients was about 6%.Algae recognition and counting methods based on deep learning can greatly improve the efficiency of monitoring algae activity and eutrophication,and play an important role in monitoring and early warning of algal blooms.