Research On Intelligent Vision For Monitoring Fruit Tree Leaf Health

The fundamental way out for agriculture is modernization.The use of artificial intelligence technology in agriculture as a method of increasing agricultural output while reducing human labor has provided strong technical support for the development of intelligent agriculture.As an important component of smart agriculture,smart orchards must adopt advanced smart agricultural solutions to meet the agricultural demand for fruit production.Agricultural production faces challenges in terms of productivity,environmental impact,and sustainability,and deep learning architecture has strong adaptability,which means it can be used to address these challenges.Unmanned Aerial Vehicle(UAV)remote sensing technology is an observation technology that has developed rapidly in recent years.Using UAVs,aerial images can be quickly captured,required data can be obtained,and real-time processing,modeling,and analysis can be performed.In this regard,the latest research trend is to integrate computer vision with deep learning,use deep learning technology to mine and learn aerial image features,and form intelligent vision-based monitoring management and prediction solutions for specific applications.With the gradual expansion of the citrus industry in China,the cultivation area and yield of citrus are among the best in our cultivated fruit.The application of unmanned aerial vehicle remote sensing technology in citrus intelligent orchard has important practical significance for predicting citrus yield,managing citrus growth status,and improving citrus yield.Therefore,this thesis takes citrus leaf health monitoring as the starting point,and proposes a solution based on intelligent vision to address two important issues in citrus cultivation management: summer shoot monitoring and leaf disease monitoring.The main work of this thesis is as follows:(1)In terms of summer shoot monitoring and management,in order to address the current issues of high manual management costs and low reliability,this thesis proposes a citrus summer shoot management algorithm based on semantic circle detection and pixel segmentation within a circle,which combines UAV aerial images,simplifying summer shoot management into a decision-making problem.In this scheme,the citrus fruit tree in the aerial image is approximated to a circle.Firstly,the citrus fruit tree is detected in the aerial image through semantic circle detection,and then the fruit tree pixels are segmented in the semantic circle to obtain the proportion of the area of the summer shoots in the entire tree.Combining the proportion of the summer shoots and other characteristic information,decisions are made on the management of the summer shoots.The proposed algorithm is experimentally demonstrated,and the experimental results show that the method achieves higher accuracy and better comprehensive performance.(2)In terms of leaf disease monitoring and management,in order to improve the detection accuracy of similar leaf diseases and highlight the characteristics of leaf lesion points,this thesis proposes a citrus leaf disease instance segmentation algorithm based on dual attention mechanism and Path Aggregation Feature Pyramid Network(PAFPN),which has good results in identifying leaf diseases that affect citrus production.Citrus diseases are diverse in nature,and some of them are similar in appearance,making it difficult for fruit farmers to accurately identify them in a timely manner.The proposed scheme in this thesis is an effective instance segmentation network structure for citrus leaf diseases.It addresses the complex background and similar symptoms and appearances of citrus leaf diseases,and can effectively detect diseased leaves even in complex background information and severe interference.At the same time,it has good recognition ability for similar leaf diseases with high recognition difficulty.(3)Based on the above work,a citrus leaf health management system was designed and implemented,which is divided into two subsystems: summer shoot monitoring subsystem and disease monitoring subsystem.The system design,implementation,and testing process are presented in the thesis.