Research Of Plant Recognition Method Based On Improved Capsule Neural Network

Plant identification is crucial for maintaining the diversity of plant species,understanding the growth characteristics and geographical distribution of plants,and rational development and utilization of plant resources.Traditional plant classification and identification work is usually based on the morphological characteristics of flowers,leaves,stems,fruits and other organ parts of plants,that is,color,texture,morphology and other characteristics are crucial to plant recognition.At present,with the rapid development of image recognition technology,automatic recognition of plant images through deep learning technology has become a research focus in the field of plant recognition,and has achieved good results,and there is a certain space for research.As one of the representatives of deep learning technology,convolutional neural network has also made rapid development.As a special deep neural network,convolutional neural network contains a special convolutional layer,which can reduce the number of parameters required for training to a certain extent,because its connection mode is local connection and weight sharing can be carried out.However,the descending sampling layer can improve the robustness of the network,greatly reduce the input dimension and complexity of the network,and effectively suppress the overfitting problem.This paper starts with the basic concepts and algorithms of convolutional neural networks,studies in depth the convolutional neural network theory and related issues,and proposes a self-attention mechanism based capsule neural network for the classification and recognition of plant images.The main work of the paper is as follows:(1)First,the basic concepts and principles of the convolutional neural network and self-attention mechanism are summarized and summarized,the basic structure and network parameters are explained in detail,the advantages are demonstrated,and the relevant content of the Dense block and the experiments are introduced.Commonly used evaluation indicators.(2)secondly,this paper studies the application of capsule network,a new depth architecture model,to plant species identification,and proposes an improved dense capsule network model,which reduces the network parameter scale and ADAPTS to the training and learning requirements of large-scale picture input and small sample data set.(3)Finally,the network model proposed in this paper compared with several kinds of classical identification methods,the data set is applied to two kinds of flowers and leaves the data set to experiment,such as the effectiveness of the proposed algorithm framework is verified,according to the final results can be concluded that under the same experimental environment and under the condition of the same data set,in this paper,higher identification accuracy,the proposed algorithm and parameters of the smaller,this paper demonstrates that the proposed algorithm is effective.To sum up,this paper studies the classification and recognition of plant image deeply,puts forward corresponding solutions,and then conducts sufficient experimental verification.The proposed algorithm model opens a new idea for the follow-up research work,which is convenient for reference and reference,as well as necessary supplement for species identification and other classification and recognition tasks.