Research Of Aquatic Plant Species Recognition Based On Convolution Neural Network Model And System Development

Aquatic plants have great potential as biological resources to remove various pollutants in water.The efficiency of species identification is related to the cultivation of aquatic plants and the treatment of water pollution.Although the traditional manual discrimination can get results,it is inefficient and has some subjective factors.Therefore,an efficient method is presented to replace the traditional manual discrimination.A set of on-line identification system for four categories of aquatic plants is developed based on the classification and recognition system of aquatic plants.The main contents are as follows:(1)Image database is established and data preprocessing.The data set was collected from the China Aquatic Plant Database and the field photography of Liaocheng National Dongchang Lake Wetland Park.By referring to the data and combining with the local agricultural documents,the judging standard of the four classifications of aquatic plants was established.According to this standard the aquatic plants can be divided into four categories,such as hygrophytes,emergent plants,floating leaf plants and submerged plants.Data preprocessing is carried out from two aspects of normalization and data enhancement to provide data support for our thesis.(2)The convolution neural network is constructed to identify for four categories of aquatic plants.Google Net model is first introduced and its Inception structure is optimized,where the experiment shows that the optimization of Google Net model training can obtain97% accuracy and 93% overall recognition rate.Aiming at the problems of low recognition rate and large model volume,we improve the structure of Res Net50 network by using small convolution kernel stack instead of large convolution kernel and an Inception structure is added to further optimize the model as well.The experiment shows that the accuracy and overall recognition rate of the model are maintained at 97% and 96% respectively,and the convergence speed is significantly improved.(3)Based on WeChat applet technology,an online identification system is built to realize the whole process from uploading pictures to displaying results.The system uses Apache as the server,using WXML,WXSS and Java Script three main technologies to complete the front-end page of the system,using PHP’s built-in method to call python files inside the server,python program loads the neural network model generated by Res Net50 optimization algorithm,and sends the identification result back to the front-end page through asynchronous communication technology.In this thesis,the deep learning technology is introduced into the We Chat applet platform,and online recognition system for identifying four categories of aquatic plants is developed.The system has a high recognition accuracy and good stability,and realizes the purpose of identifying aquatic plants accurately.