Research On Plant Seeding Image Recognition Based On Convolution Neural Network

In botany research and agricultural production,plant seedling recognition and classification and crop information monitoring and collection are the most basic and important botany research and application work.Traditional manual plant taxonomy generally requires individuals to collect and organize samples manually to obtain data and feature capture.While the data is lack of objectivity,not only the processing process is complex,but also the processing effect is poor,and there are many limitations.In recent years,with the rapid development of artificial intelligence such as machine learning,pattern recognition and digital image processing,human beings have solved some problems in actual agricultural production efficiently and accurately in many scenes.The vision algorithm in machine learning is generally based on the complex structure,large amount of calculation and high hardware requirements.The training realizes a reliable calculation model,which meets various practical needs in many application scenarios.Under the background of the lack of research on plant seedling image recognition and the rapid development of machine learning,this paper proposes a method of plant seedling image recognition based on convolutional neural network,which can give full play to the advantages of convolutional neural network to improve the reliability of recognition model,and explore the model structure suitable for solving this problem based on VGG model.At the same time,a plant seedling recognition method based on VGG model is proposed.By comparing the traditional model and analyzing the experimental results,the shortcomings of the traditional model in the plant seedling image recognition ability are obtained.The network structure and the improved method of parameter training are explored,and the improved points are compared in experiments to demonstrate the impact on the plant seedling image recognition.Then,based on VGG,we try to improve the model,design an improved network structure to improve the network performance,and get a reliable convolution neural network model for plant seedling image recognition,On the basis of the improved model,the convolution neural network structure with generalization ability in plant seedling image recognition data set is studied by migration learning and residual network.Mining the advantages and disadvantages of transfer learning in plant seedling image recognition,through resnet50 model for fine-tuning experiment and parameter transfer learning,compared with ab initio training mode and transfer learning training mode,explore and analyze the impact of training mode on plant seedling image recognition results.This paper demonstrates the feasibility and reliability of the image recognition model of convolutional neural network obtained by migration learning in the field of agriculture,as well as the natural advantages of solving the problem of insufficient training of convolutional neural network when the data set capacity is too low.With the help of migration learning,a method for image recognition of plant seedlings in agricultural application scenarios is obtained.