Design And Implementation Of Citrus Canopy Feature Detection System Based On 3D Reconstruction

The detection of canopy characteristics of citrus trees is a key task in citrus production,breeding,and management,and is closely related to precision orchard management projects such as water and fertilizer irrigation,soil moisture detection,fruit tree breeding,and yield estimation.The precise detection of citrus canopy characteristics is a crucial issue for the current intelligentization and sustainable development of agricultural construction.Traditional methods for detecting citrus canopy characteristics still rely heavily on manual measurement,which is costly in terms of labor,time-consuming,and may result in significant errors.In recent years,many researchers have focused on laser and sonar radar for detection,using depth maps collected by radar to register point clouds and reconstruct three-dimensional models of the target.Although some progress has been made,the drawbacks of laser detection radar,such as large data storage,bulky equipment,difficult portability,and high cost,prevent it from being widely used in citrus canopy detection.Existing vision-based reconstruction technology effectively provides an efficient,practical,and low-cost method for reconstructing three-dimensional digital models of fruit trees.The accuracy,completeness,and reliability of the algorithm still have a significant impact on the quality of canopy characteristic detection results.In the field of 3D reconstruction,many studies have focused on improving the accuracy and completeness of reconstruction models to make their digital models more similar to the source objects.However,in an era of rapid development of the Internet and computers with frequent product updates,many programs and theoretical algorithms are still limited by computer memory and computing power,making it difficult to achieve expected results.As the storage consumption of 3D data is geometrically higher than that of 2D data,the reconstruction algorithm is constrained by hardware equipment.Therefore,in the reconstruction process.In summary,the main research work of this thesis includes:(1)To solve the problem of high GPU memory consumption in 2D feature extraction and cost volume 3D regularization of existing depth learning based multi view geometry algorithms,this thesis proposes a multi view depth estimation algorithm based on OctConv convolution(abbreviated as OCT-MVSNet).OCT-MVSNet algorithm separates high-frequency and low-frequency information of features after 2D feature extraction and regularization of cost volume,and reduces memory overhead by compressing the proportion of low-frequency information.Secondly,in response to the existing problem of color consistency ambiguity,the OCT-MVSNet algorithm introduces feature consistency constraints to reduce errors caused by color consistency and improve the robustness of the model.And the effectiveness of the OCT-MVSNet algorithm in reconstruction accuracy,reconstruction completion,and GPU memory usage was verified in the DTU dataset and Tanks and Temples dataset.(2)Design and implement a citrus canopy feature detection system based on the OCT-MVSNet model using software engineering principles.The system is implemented with microservice architecture,adopts the front end and rear end separation strategy,and completes the development with Vue front-end development framework.The microservice architecture of the Nest microservice solution is used to split the entire system into multiple microservices,which can well solve the problems caused by highly concurrent processes and system iterations and reduce module coupling.The feature detection system that integrates the OCT-MVSNet algorithm includes fruit tree depth map estimation,depth map point cloud registration,point cloud gridding,canopy feature detection,and auxiliary decision-making,achieving the goal of assisting researchers and fruit farmers in water and fertilizer irrigation and early fruit tree yield estimation.Finally,this thesis tests the system from the perspectives of functional testing and performance testing.Verified the accuracy of system analysis and stability of system implementation.