Maojian Tea Bud Detection Method in Different Scene Based on Improved YOLOv5s

Cheng Zhenzhen; Cheng Yifan; Fang Tingting; Gong Shoufu

doi:10.14083/j.issn.1001-4942.2025.09.018

Abstract：Accurate detection of tea buds is of great significance to the production and processing of tea.Aiming at the problems of insignificant small target features and serious interference from complex background in the detection of Maojian tea buds,a detection method based on improved YOLOv5s was proposed in this study.Firstly,the SE_CSP module combining the Squeeze-and-Excitation(SE)attention mechanism with the Cross Stage Partial networks(CSP)structure was designed and introduced into the backbone network of YOLOv5s.Secondly,the BiFPN(Bidirectional Feature Pyramid Network)module was introduced into the middle layer structure of the network to enhance the model's ability to extract small target features and the bidi-rectional fusion effect of multi-scale features,so as to adapt to the target detection needs in complex scenes.The improved YOLOv5s algorithm was verified using the tea bud datasets from different scenes,and compara-tively analyzed with multiple algorithms(Faster R-CNN,MobileNetV+SSD and YOLOv5s).The results showed that the model proposed in this study improved the precision,recall and mean average precision(MAP)by 3.8,6.5 and 5.8 percentage points respectively compared with the original YOLOv5s model.The improved YOLOv5s algorithm performed well in the accuracy of identifying tea buds in various scenes,signifi-cantly reducing the missed detection rate and false detection rate,which could provide technical supports for the automated and intelligent development of the tea industry.