An Improved YOLOv5 Model for Detecting Citrus Shoots in Complex Orchard Environments

The development of citrus shoots is of great significance for the growth, flowering, fruiting, and pest and disease control of citrus trees. In order to achieve accurate recognition of citrus shoots in complex orchard environments, an improved YOLOv5 model, named YOLOv5-RepVGG-SE, was proposed. Firstly, in order to enhance the feature extraction capability, the Rep VGG modules are used to replace the Conv modules in the backbone networks of the YOLOv5 model. Second, the Squeeze-Excitation (SE) attention modules are added to the neck networks of the YOLOv5 model to improve the feature fusion capability. A dataset of citrus shoots was established to evaluate the effectiveness of the improved YOLOv5 model. The experimental results show that the improved YOLOv5 model achieves a precision ( $P$ ) of 81.1 %, a recall ( $R$ ) of 76.8%, and a mean average precision (mAP) of 71.0% at a threshold of 0.5. These values are 0.6%, 9.20/0, and 9.4% higher, respectively, than those of the YOLOv5 model. This research can serve as a reference for the development of smart devices in citrus orchards.