Tea bud pose estimation and grading detection network based on improved YOLOv7.

Area of Science:

• Computer Vision
• Agricultural Robotics
• Machine Learning

Background:

• Traditional algorithms struggle with complex backgrounds and variable tea bud growth, limiting tea-picking machinery.
• Existing methods primarily focus on identifying picking points, neglecting crucial pose and grade information, thus hindering harvesting efficiency.

Purpose of the Study:

• To develop a deep neural network for simultaneous tea bud pose estimation and classification.
• To improve the accuracy and efficiency of tea bud recognition for intelligent harvesting.

Main Methods:

• Proposed YOLO-PC, a deep neural network integrating dynamic snake convolution (DSConv) for shape feature extraction.
• Incorporated ELASPP-CSPC attention mechanism for enhanced spatial pooling and EIoU loss for improved localization accuracy.
• Utilized deep learning for simultaneous pose estimation and classification of tea buds.

Main Results:

• Achieved high detection accuracies: 91.5% (one-bud-one-leaf) and 93.2% (one-bud-two-leaf).
• Obtained an average keypoint detection accuracy (Pose_mAP) of 89.7% and a Normalized Mean Error (NME) of 0.047.
• Outperformed YOLOv7-pose by increasing mean average precision by 7.26% and pose accuracy by 9.65%, while reducing parameters by 14.99 M.

Conclusions:

• The proposed YOLO-PC model demonstrates superior performance in tea bud detection, pose estimation, and classification.
• The model offers robust practical support for adaptive and intelligent tea harvesting systems.
• YOLO-PC enhances harvesting efficiency by accurately recognizing and grading tea buds.