Research And Application Of Accurate Recognition Algorithm For Algae Image

Water is indispensable to life,and water quality is obviously an extremely important measure of ecological conditions,and algae in water is an important component of water quality.The study of algae information in water samples is of great significance for water pollution and sewage treatment.However,limited by the fact that most algae need a microscope to observe this situation,the recognition of algae mainly relies on manual detection,and computer vision,deep learning and other fields have not been really applied to the recognition of algae.Aiming at the backward status of algal recognition,this thesis develops an accurate algal image recognition algorithm,and applies it to design a system that can accurately identify algal image.The main contents and contributions are as follows:1.The existing image segmentation algorithm based on concave registration method is improved to make it more accurately applied to the image segmentation of adhesive algae.In pits two matching segmentation synechia algae image segmentation by means of introducing morphological operations before the mild adhesion algae image,thus effectively reduced the number of must pits;Then,based on the process logic of concave point matching,the feature of alga microscopic image is optimized,so that the global concave point retrieval strategy is transformed into a local retrieval strategy.Through this change,the accuracy is not affected and the computational burden of the algorithm is reduced,so as to improve the overall operation efficiency of the algorithm.After that,a segmentation algorithm based on distance transformation function is proposed to solve the problem of isolated concave points which can not be solved by the concave point matching algorithm.Finally,according to the characteristics of algal adhesion image,a special three-adhesion segmentation algorithm is proposed,which can effectively improve the integrity rate of individual algal cells under the premise of maintaining the overall segmentation accuracy,and the extra cost of time can be neglected.2.A network model rn-alga-based for Algae micro-image recognition was designed.Based on the resnet-50 network model,this network model improved the pooling method for algae-algae-6 characteristics,enabling the pooling layer to adapt to the input image size and to fix the output dimension to match the fully connected layer.The accuracy of the experimental results reached more than 88%.3.The migration performance of RN-Algae micro-image recognition network was verified by comparative experiments,and the results showed that the network had good migration learning performance.As the algal database is expanded,new models that can quickly converge and recognize more algae can be trained.