【摘要】 为实现农业复杂环境中红花的快速准确识别，提出了一种基于深度学习方法的改进YOLOv5s红花目标检测算法。在YOLOv5s基础上融入适配GPU的轻量Ghost模块，获得复杂度更低、网络推理速度更快的基线模型，将CBAM注意力机制嵌入基线模型，增强了小目标物在高频特征中的表现力，并通过建立一种基于边界框宽和高差值的Focal-EIoU损失函数，提高红花在不同遮挡情况下的识别率。最后，在并联式红花采摘机器人上开展红花识别试验，验证改进算法的可行性和可靠性。结果表明：改进后的YOLOv5s模型相较于原始模型在mAP值上提高了1.94个百分点，模型参数量和单幅图像检测速度分别为3.52 MB和0.06 s/幅，红花采摘机器人视觉系统的平均识别成功率可达89.92%。更多还原

【Abstract】 In order to realize rapid and accurate recognition of flesh safflower in complex agricultural environment, a new method based on improved YOLOv5s was proposed. Based on YOLOv5s, a GPU-adapted lightweight Ghost module is integrated to obtain a baseline model with lower complexity and faster network reasoning speed. CBAM attention mechanism is embedded into the baseline model to improve the performance of small objects in high frequency features. A Focal-EIoU loss function based on border width and height difference was established to improve the recognition rate of safflower under different occlusion conditions.Finally, experiments on a parallel safflower picking robot are carried out to verify the feasibility and reliability of the improved algorithm. The experimental results show that the mAP value of the improved Yolov5s model is improved by 1.94 percentage points compared with the original model. The parameters of the model and the detection speed of a single image are 3.52 MB and 0.06 s/amplitude respectively, the recognition success rate of robot vision system for picking safflower can reach 89.92%.更多还原