深度学习的车间零件分拣机器人目标识别方法

Object Recognition Method for Workshop Parts Sorting Robot Based on Deep Learning

针对目前方法识别机器人目标时,由于未能详细分析分拣机器人运动规律,导致该方法开展目标识别时,存在平均置信度低、识别效果差等问题,提出深度学习的车间零件分拣机器人目标识别方法。该方法通过分析分拣机器人运动规律,提取待检测目标的特征向量值;结合深度学习理论建立目标识别模型,并寻找模型最佳参数;建立待检测目标的相关测试集放入模型中训练,基于模型输出结果,完成机器人的目标识别。实验结果表明,运用该方法识别目标时,其在特征提取后,分拣目标数量为1000个时,识别准确率达到了97.5%以上,识别耗时在100s以下,平均置信度约为0.8,有效提高了平均置信度、降低了识别时间,识别效果好。

When the current method is used to identify robot targets,it fails to analyze the movement law of sorting robots in detail,leading to problems such as low average confidence and poor recognition effect in target recognition.A deep learning method for workshop parts sorting robot target recognition was proposed.The method extracted the feature vector value of the target to be detected by analyzing the results of the motion law of the sorting robot;the deep learning theory to establish a target recognition model was combined,and the best parameters of the model was found;established the relevant test set of the target to be detected and puted it into the model for training,and the target recognition of the robot based on the model output was completed.The experimental results show that the recognition accuracy reaches more than 97.5%,the recognition time is less than 100s,and the average confidence is about O.8,which effectively improves the average confidence and reduces the recognition time,and the recognition effect is good when the target number is 1000 after feature extraction.