Plant Image Sets Classification Using Deep Learning

Plant taxonomy is the basic science that accurately describes the plant,names,classifies,exploits the kinship between groups,and the evolutionary development.With the rapid development of pattern recognition technology in plant image classification task,it has a positive effect on the development of plant taxonomy,and it has brought great help to the scientific research of agriculture.The key to the image classification algorithm based on image set is how to model the image set and how to measure the similarity between the image sets,compared with the traditional image recognition algorithm based on single or a few images.In order to better classify the plant images,it is necessary to study the classification technology of the plant leaf image set as the research object.In this paper,the plant leaf is used as the research object,the nonlinear reconstruction model,the SPCANet model,the KmeansNet model and the depth model were used to describe the granularity of the plant image were introduced in detail.The main works in this paper are as follows:(1)A plant image set identification approach is proposed based on non-linear reconstruction models.This approach initializes the parameters of model by performing unsupervised pre-training using Gaussian Restricted Boltzmann Machines(GRBMs).Then,the pre-initialized model is separately trained for images of each plant set and class-specific models are learnt.At last,based on the minimum reconstruction error from the learnt class-specific models,majority voting strategy is used for classification.Besides,a method of feature extraction is used based on k-means.(2)A classification method of plant image set based on SPCANet model is proposed.Firstly,the SPCANet modle is used to extract the feature of the plant image,and then the linear SVM is used to classify it.Finally,the category of the test set is determined according to the voting strategy.The model is design based on the CNN(convolutional neural network).The model consists of convolution filter layer,nonlinear layer and feature extraction layer.The convolution kernel of the convolution layer is obtained by PCA algorithm which is defferent from the traditional deep learning network.This greatly shortens the network's training time and parameter settings.(3)A classification method of plant image set based on KmeansNet model is proposed.The method is a variant of the SPCANet model,except that the convolution kernel of the convolution layer is obtained by the Kmeans algorithm.(4)Using the depth learning framework Caffe to classify large-scale plant images.The idea of introducing granularity classification provides a new idea for the classification of large-scale plant images.In the background of large data,the use of Caffenet model of large-scale classification capabilities,by fine-tuning the Caffenet worknet in order to achieve the plant images were classified according to the door,grade,head,branch and genus.