In order to address the issues of low recognition rates and slow detection speeds for cotton apical buds under complex environmental conditions, an improved YOLOv5s object detection model is proposed. Initially, data of cotton apical buds in complex cotton field environments were collected. Subsequently, a lightweight Hd-ShuffleNetv2 network module was integrated into the backbone of the model to reduce the number of model parameters and accelerate detection speed. Additionally, NLMA and BotNeT attention mechanism modules were incorporated into the neck of the model to enhance feature extraction capabilities for cotton apical buds, thereby improving the model's recognition accuracy. Finally, the EIoU loss function was employed to tackle recognition challenges in cases where the apical buds were partially occluded, further increasing the success rate of identification. In order to verify the practical effectiveness of the improved object detection model, tests were conducted on cotton apical bud samples. The test results indicated that the mean average precision of the improved YOLOv5s model reached 91%, 1 percentage point increase over the original YOLOv5s model, with an enhanced detection confidence. The improved object detection model meets the detection requirements for cotton laser topping machines in the field, and provides robust technical support for further research in cotton laser topping technology.

针对在复杂环境下棉花顶芽识别率低、检测速度慢的问题，提出一种改进的YOLOv5s目标检测模型。首先收集在复杂棉田环境下棉花顶芽数据，其次在模型的主干网络中加入Hd-ShffleNetv2轻量化网络模块，以减少模型参数量，并加快模型的检测速度。同时在颈部中加入NLMA与BotNeT注意力机制模块，增加对棉花顶芽的特征提取能力，从而提高模型的识别精度。最后，采用EIoU损失函数来解决在顶芽部分遮挡情况下的识别问题，进一步提高识别成功率。为验证改进的目标检测模型的实际效果，对棉花顶芽样本进行测试。测试结果表明，改进的YOLOv5s模型的平均检测精度达到91%，较比原始的YOLOv5s模型提升1个百分点，模型的检测置信度也有所提升。改进的目标检测模型满足棉花激光打顶机在棉田中的检测需求，为棉花激光打顶技术的进一步研究提供有力的技术支撑。