Plant Disease Detection Based On Visible Images And Convolutional Neural Network

Agriculture has been always an essential part of national economy.It is not only related to the development of civil economy,but also the quality of people’s daily life.However,due to the situation of traditional plant disease,detection which overly relies on the experts’ experience,poor efficiency and intelligence have become the limitation of performance of plant disease detection.With the continuous development of computer vision and neural network,and the popularization of smart phone,people have paid more and more attention on the intelligent plant disease detection,which has obtain quick development later.This paper has proposed a plant disease detection method based on visible images and convolutional neural network,which has obtained a good effect for plant disease detection.In this paper,the plant disease detection method based on visible images and convolutional neural network mainly includes four parts: convolutional neural network,transfer learning thought,support vector machine,and data augmentation.Detailed introduction for every part has been presented in this paper.Before the formal introduction of the proposed method,the paper has made detailed introduction of our plant disease dataset,which is used in nearly all the experiments.Meanwhile,this paper has listed a group of results from a kind of traditional method used on our plant disease dataset.From the results we can find the performance of the traditional method is not satisfying,especially on large-scale dataset.Later,the paper starts to introduce the proposed method formally.First of all,the paper has introduced a detection method just based on convolutional neural network.This part content has made detailed and comprehensive description about the structure design and training details of convolutional neural network.At the same time,the classification results of our plant disease dataset by the detection method just based on convolutional neural network has been listed.The reason of over-fitting problems also has been analyzed.Later,this paper has made improvement for proposed method based on transfer learning thought.From this part content,improvement based on transfer learning thought could mitigate the over-fitting problem largely;in the result,the classification accuracy has large increase.Next,we use support vector machine as classifier to replace softmax layer,bringing little increment of classification accuracy and mitigation of over-fitting problems to some degree.At the last part,the paper has introduced important meaning of data augmentation for mitigation of over-fitting problems and improvement of plant disease detection accuracy.Relevant experiments conclusions has been presented.What’s more,this paper has applied the proposed detection method on the dataset collected by ourselves,which is closer to actual situation.The performance is relatively good and we could conclude that this proposed method is robust.Compared with traditional methods,the plant disease detection method based on visible images and convolutional neural network proposed by this paper could produce good detection results.Its intelligence and high accuracy could make good effects as for practical problems.