Research On Fresh Tea Type Recognition Based On Deep Learning

As part of the development of modern agriculture,intelligent tea identification has always been the focus of attention.With the advent of the era of artificial intelligence and the development of information technology,how tea detection departments and tea companies use computer technology to identify fresh tea varieties,quickly and accurately distinguish the true and false tea,and at the same time save the cost of tea detection,the focus is on Improve the accuracy of tea identification.Therefore,it is extremely important to improve the accuracy of tea recognition through computer technology.In recent years,tea recognition has been a research hotspot in the field of computer vision.This article combines learning function and neuroscience,and is creatively applied to tea recognition.It improves the accuracy and timeliness of tea recognition by improving and optimizing deep learning algorithms and models.The main work of this article is as follows:(1)The algorithm research and models in the field of computer vision in recent years are summarized and summarized.In the early research in the field of deep learning,the inverse multiplication algorithm of artificial neural network was used for feature recognition,and then there were methods such as support vector machine,lifting,and maximum entropy.Summarizes the development of automatic coding,ultimate Boltzmann machine,deep belief network and convolutional neural network on the model.(2)Completed tea data collection and preprocessing.This article uses high-definition cameras to sample 15 tea varieties in natural environment.After sampling,through pre-processing such as light changes,horizontal mirror flip,random cropping,data standardization,etc.,through a series of pre-processing The data of 15 varieties of tea is greatly expanded through data standardization,and the final data volume is about 5,200 image data.(3)Carried out research on tea recognition rate based on deep learning.The watershed function is used to scan the data of 15 varieties of tea data sets,and the scanned data is divided by the grabcut algorithm,and finally the divided data is divided into the training set,the verification set,and the test set for network model training and testing use.For the data set segmented by the final scan,when the shallow 8-layer CNN neural network model is used for training,it does not perform particularly well in terms of accuracy or time.The recognition rate of all varieties of tea is about 80%.Therefore,for the expanded database,our research ideas are mainly to use the deep learning framework to improve the idea of the CNN network model,deepen the network level,adjust the network parameters,and make the layer-by-layer autonomous learning deeper.The experimental results show that by adding the transfer learning initialization method,adjusting the deep network structure and using the Momentum gradient descent algorithm,the recognition accuracy of the same tea data set is significantly improved compared to the shallow network,and the recognition rate of all varieties of tea is generally improved.To more than 90%,the highest individual recognition rate reaches 99.23%.This paper aims at the problem of relatively low computer recognition rate in the process of tea recognition.Through the implementation of a series of optimization methods,the optimized network has deeper,wider and better performance,thereby solving the shortcomings of the traditional shallow network model The accuracy rate of tea identification is provided,which provides a theoretical basis for improving the accuracy of tea quality identification and tea true and false identification,and also serves as a theoretical reference for the intelligent classification of later tea varieties.