| Smart agriculture is the development direction of future agriculture.The rapid development of technologies such as the "Internet of Things","big data" and "artificial intelligence" is leading the transformation of modern agriculture to information and digitalization.Agricultural Internet of Things technology is widely used in monitoring systems for agricultural environments such as greenhouses and orchards because of its great advantages in cost,practicality,and convenience.Computer vision technology is widely used in manufacturing,agriculture,medical,security and other industries.In recent years,deep learning has continuously improved the performance of tasks such as image classification and target detection in computer vision.Applying computer vision technology to the orchard monitoring platform can automatically understand and analyze the orchard image information,and realize a more comprehensive perception of crop growth information.In order to realize the intelligent monitoring of fruit growth information in the orchard,an intelligent monitoring system for litchi orchards has been designed and built in this article.Based on computer vision and deep learning technology,this paper focused on researching fruit detection methods and fruit maturity discrimination algorithms in orchards,and developed a web-based remote monitoring platform for litchi orchards.The main work of this article is as follows:(1)Firstly,images of litchi fruit in various growing periods in the orchard were collected,and the images were annotated and divided to construct a litchi image data set.This dataset provides data support for the study of litchi fruit detection tasks based on deep learning.(2)Based on the YOLOv3 algorithm,this paper improves the target detection model to make the algorithm more suitable for the detection of litchi fruits in orchards.The k-means++ clustering is used to determine 9 bounding box priors,which is used to train the litchi detection model.This paper proposes the YOLOv3-darknet44 network model with a larger output scale and fewer network layers than YOLOv3.It has been verified that the improved model has significantly improved the accuracy and recall rate of litchi fruit detection under the large visual scene,and the detection speed is also faster.(3)For single litchis classification task,this paper builds a convolutional neural network model to identify fruits of different maturity and diseased fruits.Following the empirical rules of manual recognition,the algorithm mainly divides litchi fruits into four categories: unripe fruits,70%～90% mature fruits,90%～100% mature fruits,and diseased fruits.After dividing the data set,a data augmentation was taken to the train set.The model is traind and tuned by using Pytorch.After testing,the classification model has an accuracy rate of 100% for identifying immature litchis,97% for 70%～90% fruits,97% for90%～100% fruits and 92.3% for diseased fruits.The average accuracy on the entire test set is 96.58%.(4)An intelligent monitoring system for litchi orchards is designed and built by combining the Internet of Things and computer vision technology.In this system,the number and maturity of fruits are automatically detected by deploying a deep learning model on a smart camera,and environmental data are collected by wireless sensor devices.The framework of a remote monitoring platform for litchi orchard based on Web is built in this paper,and the functional modules that make up the platform are described.The back-end database structure of the monitoring platform is designed.According to the actual production environment of litchi cooperative orchard,wireless sensor hardware equipment was arranged on site to collect environmental data of the orchard.A web-based litchi production platform of IOT has been established. |
PDF Full Text Request