基于改进Mask R-CNN特征点精确提取技术研究与应用

Research and application of accurate extraction technology of feature points based on improved Mask R-CNN

针对大型零部件自动对接装配系统使用的传统特征提取算法泛化能力不足的问题,提出一种基于改进Mask R-CNN模型的深度学习特征提取算法。首先采集目标对象的图像数据并进行人工标注,采用数据增强策略扩充数据集;其次构建Mask R-CNN模型,基于迁移学习策略预训练模型,使用人工采集数据集训练模型;然后使用鲸鱼优化算法改进Mask R-CNN模型,优化超参数,进一步提升模型性能;最后对传统特征提取算法和基于Mask R-CNN模型的深度学习特征提取算法进行对比实验,验证了在自动对接装配系统中使用所提算法有更高的准确性和更强的泛化能力。

Aiming at the defect of insufficient generalization ability of the traditional feature extraction algorithm used in the automatic docking assembly system of large parts,a deep learning algorithm based on the improved Mask R-CNN model is used.Firstly,it collects the image data of the target object and manually marks them,and uses the data enhancement strategy to expand the data set.Secondly,it builds the Mask R-CNN model,pre-train the model based on the migration learning strategy,and uses the artificially collected data set to train the model.Thirdly,it uses the whale optimization algorithm to improve the Mask R-CNN model,optimizes hyperparameters,and further improves model performance.Finally,a comparative experiment is carried out between the traditional feature extraction algorithm and the feature extraction algorithm based on the Mask R-CNN model,and it is verified that the deep learning algorithm used in the automatic docking assembly system has higher accuracy and stronger generalization ability.