基于Shufflenetv2-YOLOv5的番茄采摘机器人目标检测方法

Target detection method for tomato picking robot based on Shufflenetv2-YOLOv5

为了使目标检测算法更好地嵌入到番茄采摘机器人中,使其可以在温室环境中快速准确地识别检测成熟的番茄,提出一种基于Shufflenetv2-YOLOv5的番茄采摘机器人目标检测方法。该方法以YOLOv5为基础,以轻量级网络Shufflenetv2为主干网络,减少参数量和计算量,加入ECA注意力机制,并用FReLU激活函数替换原来网络结构中的激活函数,实现像素级的空间建模能力,进一步提高检测精度,增加该模型的鲁棒性。试验结果表明,改进后模型的算法精确率P、召回率R和平均精度均值mAP分别提升了2.2%、1.9%、3.2%,参数量降低约54%,计算量降低约62.6%,模型大小减少约53.1%,FPS也有了一定的提升,能够满足高速度、高精度检测的要求。

In order to better embed the object detection algorithm into tomato picking robots and enable them to quickly and accurately recognize and detect mature tomatoes in greenhouse environments,a tomato picking robot object detection method based on Shufflenetv2-YOLOv5 is proposed.This method is based on YOLOv5 and uses the lightweight network Shufflenetv2 as the backbone network to reduce parameter and computational complexity.Secondly,ECA attention mechanism is added and FReLU activation function is used to replace the activation function in the original network structure,achieving pixel level spatial modeling ability,further improving detection accuracy,and increasing the robustness of the model.The experimental results show that the improved model's algorithm accuracy P,recall R,and average average accuracy mean mAP have been improved by 2.2%,1.9%,and 3.2%,respectively.The number of parameters has been reduced by about 54%,the computational complexity has been reduced by about 62.6%,the model size has been reduced by about 53.1%,and FPS has also been improved to a certain extent.It can meet the requirements of high-speed and high-precision detection,providing an effective solution for tomato picking robots.