Study On The Identification Of Tea Diseases Based On Transfer Learning

Tea is an important part of Chinese agricultural crops,but with the continuous expansion of China’s tea plant area,the problem of tea diseases has become increasingly prominent.According to the analysis of 32 tea exporting countries and regions in the world by the European Union,Chinese tea agricultural residues is one of the countries with the highest exceeding standard rate in the world.In order to reduce the loss of tea caused by diseases,and effectively reduce the use of pesticides and the impact on the environment.This paper will start from the identification of tea diseases.It provides a feasible scheme for automatic diagnosis of tea,promotes the development of intelligent diagnosis of tea,and improves the quality and yield of tea.At present,there is little research on tea diseases at home and abroad,the main reason is the lack of a large number of practical data samples.and in the field of image recognition,the traditional machine vision image recognition methods need to extract features manually,the accuracy is low;and the recognition and classification research based on convolution neural network needs a large number of sample data,but the actual tea disease sample collection is extremely difficult at the same time,for deep convolution neural network training from scratch,it needs a lot of time.Therefore,in view of the above problems,and on the basis of the research on the domestic and foreign tea disease control measures,this paper studies based on the migration of the model,and the specific contents are as follows:(1)In view of the fact that the image of tea disease is complex and the characteristics of the disease are not significant,this paper proposes a series of processing for the data of tea disease,including image segmentation and image enhancement of the disease spots to improve the local characteristics.(2)In view of the small number of tea data samples,the traditional machine learning needs to extract features manually,but the actual image data is complex and the computer cost is high,so it is difficult to be applied in practice.Therefore,this paper proposes a transfer learning method,and through fine-tuning,modify the full connection layer to adapt to the target domain recognition task.(3)For labeled data,the similarity between the source domain and the target domain is not high.In order to further improve the adaptability of the migration model to the target domain,an improved densenet deep migration model is proposed.An adjustment module is added to improve the feature recognition ability and adaptability of the model to reduce the content difference between the source domain and the target domain.The experimental results show that this improved method has remarkable effect.