基于RT-DETR-Faster的苹果采摘机器人实时目标检测算法

Real-time Object Detection Algorithm for Apple Picking Robots Based on RT-DETR-Faster

为了解决苹果采摘中目标小,实时性要求高等问题,提出了一种基于RT-DETR的采摘机器人目标检测方法,名为RT-DETR-Faster。首先,采用FasterNet部分卷积替换主干网络的传统卷积,有效提升了模型的运算速度;其次,使用改进的级联注意力编码器替换原始的编码器,使网络更专注于目标区域;最后,引入Faster_Rep融合特征模块,保留更多有效特征并减少计算量。该文在实际的果园图像上进行了实验,结果表明,该文提出的算法与原始的RT-DETR算法相比,FPS提升了34%,帧数达到了47.9,同时准确率更高,适用于苹果采摘机器人的实时目标检测任务。

In order to solve the problems of small targets and high real-time requirements in apple picking,this paper proposes a target detection method for picking robots based on RT-DETR,named RT-DETR-Faster.Firstly,the paper adopts FasterNet partial convolution to replace the traditional convolution of the backbone network,which effectively improves the model’s computation speed.Secondly,the paper uses an improved cascaded attention encoder to replace the original encoder,which makes the network more focused on the target area.Finally,the paper introduces Faster_Rep feature fusion module,which preserves more effective features and reduces the computation cost.Conducted experiments on real orchard images to evaluate the effectiveness of proposed algorithm.The results show that our algorithm achieves a 34%increase in FPS,reaching 47.9 fps,and a higher accuracy compared to the original RT-DETR algorithm,making it suitable for real-time object detection tasks of apple picking robots.