Identification Of Ginkgo Leaf Disease Based On Convolutional Neural Network

Ginkgo biloba is an important economic tree,with extremely high medicinal value and ornamental value.It’s unique terpenoid components make it irreplaceable in the medical field,and therefore has great potential for development.For forestry prevention aspect,with the development of Internet and computer technology,digital forestry management and "wisdom forestry" have become a trend.Machine vision has great significance in realizing real-time monitoring of forest areas and promoting the digital management of forestry information.In this context,we will explore how to identify the health status of Ginkgo biloba through electronic mobile devices,so that managers in the forest region can grasp the status of trees,and it is necessary to treat the diseased trees as soon as possible.In this paper,Ginkgo biloba leaves with different degrees of ring vein disease were selected as experimental samples,convolutional neural network was designed to classify five types of leaves.According to the characteristics of leaves which have ring vein disease,an 18-layer convolutional neural network is designed in this paper.It consists of four feature extraction layers and two fully connected layers.The performance of different aspects of the network is tested by adjusting different types of parameters,such as the convolution kernel parameters,learning rate and sample set.A data-enhanced method was used to achieve a sample expansion of five times that of the original sample set,which included a fuzzification sample designed for testing network robustness.By adjusting the convolution kernel size parameter,the feature mapping layer can be added within the range to improve the recognition accuracy.Convolutional neural network can recognize the accuracy of sample sets with five different degrees of diseased leaves.The average rate can reach 94.85%,and the highest can reach 100%.Retrain the well-trained model to get the green score in each category,and set this value as a digital judgment of the true value of the disease level classification.After realizing the hierarchical recognition on the computer,encapsulating the algorithm and using transformation model to design an application by Java language.After the program is installed on the mobile phone,the client can test the disease level of the ginkgo leaf.In addition,two sets of comparison experiments are designed in this paper,which are classical Alex-net and BP neural network based on K-Means clustering feature extraction.The final experimental data show that the convolutional neural network designed in this paper is optimal.