一种基于深度学习的机械臂分拣方法

A robot sorting method based on deep learning

针对分拣过程中视觉系统识别复杂物体时速度慢、对环境变化适应性不足的问题,提出一种基于轻量型卷积神经网络的机械臂快速分拣方法。该方法首先使用基于轻量型卷积神经网络的MobileNet-SSD算法对图像中的目标物体进行检测,获取目标类别和位置信息;然后根据检测输出结果对图像进行预处理和边缘检测,最终得到校正后的定位结果。在PROBOT Anno机械臂平台上进行分拣实验,实验结果表明,相比于传统的图像处理方法,提出的方法能对复杂目标物体实现快速的检测和定位,对于目标形态和环境的多样性来说具有更好的鲁棒性。

A fast robot sorting method based on lightweight convolutional neural network was proposed to improve the recognition speed and environmental adaptability,especially for sorting complex objects.Firstly,the MobileNet-SSD algorithm was used to detect and classify the objects based on lightweight convolutional neural network.Secondly,image preprocessing and edge extraction were used to revise the object locations according to the above object detection results.The sorting experiments on PROBOT Anno robot arm show that the proposed method can achieve fast detection and location for complex objects.Compared with traditional image processing methods,the proposed method is robust to the diversity of target morphology and environment.