Research On Plant Leaf Recognition Based On Image Analysis

At present,the popularity of plant science research has not reached a high level.People can't get the scientific name,character,category and other characteristics of the plant quickly and accurately,which brings some difficulties to plant protection and plant disease knowledge research.Therefore,plant recognition is an indispensable task in plant learning and plant research.This paper focuses on the recognition of plant leaf features.Firstly,this paper designs a recognition system based on the characteristics of plant leaves,and tests the accuracy of the system for comparative study.After that,the plant depth recognition system is designed by using convolution neural network of deep learning,which realizes the automatic recognition of plant leaves under complex background and non artificial participation.The main work of this paper is as follows:(1)Building plant leaf data set.In this paper,Plant Village data set,AI Challenger PDR2018 and PPBC(Chinese plant image database)are used as the basis of plant leaf database.By analyzing and comparing the advantages and disadvantages of these three kinds of data and the degree of agreement with this study,13 kinds of data sets with a capacity of about 40000 are finally generated through extraction and fusion.(2)A plant leaf recognition system based on Feature Engineering is designed.Through image preprocessing,feature extraction,feature processing and other operations,the shape and texture features of plant leaves are extracted.Finally,a three-layer BP neural network is designed as a classifier to identify features,and the recognition results are obtained.(3)Based on deep learning convolution neural network,Resnet-inception network model is designed to automatically classify and identify plant species.Firstly,the data set of plant image is expanded to 300000 capacity for deep convolution network training through regularization,noise and image enhancement.Through training VGGNet,Goog Lenet V3,Res Net and Resnet-inception network model designed in this paper,the accuracy of 79.8%,90.4%,89.7% and 92.8% respectively are obtained.By comparing the training accuracy andthe convergence speed of each network,the conclusion is that this improvement improves the accuracy and speeds up the convergence speed.(4)The theory of traceability is used to evaluate the stability of the model.From the heat map after tracing the source,we can clearly see the pixel distribution which contributes a lot to the recognition.Through analyzing the heat map,we can draw the conclusion that the contribution degree of the image decreases with the increase of the edge of the image.(5)Finally,this paper transplanted the designed recognition model to the mobile terminal of mobile phone,and made the plant leaf image recognition app.The plant leaf image to be recognized is extracted through the mobile phone shooting or mobile phone storage,and uploaded to the server for recognition to get the plant type.