Research And Implementation Of Plant Classification And Recognition Method Based On Residual Network

In the development of botany research and agricultural production,plant species identification and plant information collection have various applications.Research based on plant species and characteristics.By this study,we can explore multiple values of plants and protect biodiversity.A digital management system can be developed to provide service for scientific and rational reasonable of plant species information.This paper proposes a plant image recognition method based on residual network.Our method focuses on the image recognition of medicinal plants on the grassland,using the convolutional neural network model to learn a targeted and effective feature extraction algorithm in specific tasks.This design includes data preprocessing method,feature extraction network,classification and recognition methods research,hardware and software design.First,it analyzes and studies the traditional plant recognition situation and proposes a plant image classification and recognition method based on convolutional neural network.Construct plant image dataset,including public image dataset and grassland medicinal plant image dataset.Preprocessed the image according to the characteristics and enhance dataset.Set up a comparison experiment to select the basic network model,then analyze the influence of different network models and neural network depth on plant image recognition,finally we use Residual Network,which is “Res Net34” as the basic network model.The features of plant species image recognition are highly similar,and there is little difference between plant species and genera.To be able to locate and identify a more discriminative area,improve recognition accuracy.Using weakly supervised attention learning.In this paper a network is proposed for plant species image recognition,it based on Res Net34.The Transfer Learning is used to accelerate network training and improve performance.A plant image classification and recognition model based on the residual network Res Net34 is proposed,and the effectiveness of this method is verified through experiments.It achieves 96.8% training accuracy and 86.3%verification accuracy on the plant image data set.The Transfer Learning is used to accelerate network training and improve performance,the accuracy of network training reached 98.2%.Design and implement an interactive operating system for image recognition of medicinal plant species,then use GUI interactive interface to complete image recognition and data management operations.