Abstract:
Completing mulberry tree branch recognition in the complex natural environment is a critical role in implementing the intelligence of the mulberry leaf picking machine. For the challenge of many changes in lighting conditions, mulberry leaf shading, and mulberry tree branching in practical applications, this paper proposes a deep learning-based method for mulberry tree branch recognition in complex natural environments. Firstly, rotation, mirror flip, color enhancement, and homomorphic filtering were used to extend the image datasets, which could improve the robustness of the model. Then, we obtained the required coordinates of mulberry branches in the photos through Resnet50 target detection network model and camera calibration, and found that the model had the best recognition effect when the learning rate was set to 0.001 and the number of iterations was set to 600. The method has good adaptability to different lighting conditions in complex natural environments, and is able to identify and obtain coordinate information for mulberry branches with multiple branches and those shaded by mulberry leaves, with an accuracy rate of 87.42%, which can meet the actual working requirements.