【摘要】 为提高多物体抓取检测网络的抓取检测准确率，提出一种基于改进Cascade R-CNN模型的机械臂抓取检测算法.首先，引入ResNeXt结构能够在不加大网络设计难度的前提下提高了模型的准确率；引入带空洞卷积的空间金字塔池化模块以解决分辨率较低的问题；接着对抓取框回归分支和角度分类分支以分治方法进行优化.其次，针对多物体抓取数据集缺乏的问题，构建多目标抓取数据集(multi-object grasping dataset, MOGD),有效地扩充了多物体抓取检测数据集.最后，基于改进Cascade R-CNN模型设计抓取检测网络，实验结果表明，改进后的算法效率更高，PI-Cascade R-CNN实验准确率为93%,较Cascade R-CNN提升1.5个百分点.更多还原

【Abstract】 In order to improve the grasping detection accuracy of the multi-object grasping detection network, a robotic arm grasping detection algorithm based on the improved Cascade R-CNN model is proposed. Firstly, the introduction of the ResNeXt structure can improve the accuracy of the model without increasing the difficulty of network design. The atrous spatial pyramid pooling module is introduced to solve the problem of low resolution. Then, the grasping box regression branch and the angle classification branch are optimized by the divide and conquer method. Secondly, aiming at the lack of multi-object grasping datasets, a multi-object grasping dataset(MOGD) is constructed, which effectively expands the multi-object grasping detection dataset. Finally, a grasping detection network is designed based on the improved Cascade R-CNN model. The experimental results show that the improved algorithm is more efficient. The experimental accuracy of the PI-Cascade R-CNN is 93 %, which is 1.5 percentage higher than that of the Cascade R-CNN.更多还原