Research on Tea Bud Detection Method Based on Improved YOLOv8

The tea industry holds significant economic and social importance in China; however, traditional manual detection methods are inefficient and costly. To accurately estimate tea production and determine the optimal harvest time, this paper proposes a tea bud detection and counting model based on an improved YOLOv8. This model addresses the limitations of the YOLOv8 framework by adding deformable convolution modules to the backbone network, enhancing small object detection capabilities. It also integrates the CBAM attention module in the neck structure to focus on key features and replaces the original CIOU loss function with HIOU, dynamically adjusting component weights at different training stages. This adjustment allows small objects to regress more accurately to their true bounding boxes, further improving the detection accuracy of tender tea buds. Experimental results show that the improved YOLOv8 model achieves higher precision and recall in tea bud detection tasks, demonstrating its superior performance. The outcomes of this research are of great significance for advancing intelligent harvesting and quality control in the tea production sector.