【摘要】 为实现岩石种类的智能化识别,对橄榄岩、玄武岩、大理岩、片麻岩、砾岩、石灰岩图像进行数据扩增处理,在Faster R-CNN深度学习目标检测框架下,采用简化VGG16为基础特征提取网络,对岩石图像进行特征提取和特征学习,进而训练成岩石种类区分识别模型。通过验证,模型对6张单类岩石图像识别均正确且分类概率均大于96%。对多类岩石混合图像区分识别效果良好,大部分分类概率超过80%且定位准确。模型能够很好地将图像特征相似、有遮挡的岩石种类区分识别出来,证明模型的鲁棒性和泛化能力较强。更多还原

【Abstract】 In order to realize the intelligent identification of rock types,the images of peridotite,basalt,marble,gneiss,conglomerate and limestone were enlarged and processed.Under the framework of Faster R-CNN depth learning target detection,the feature extraction and feature learning of rock images were carried out based on the simplified VGG16 feature extraction network,and then the discrimination and recognition model of rock types were trained.By verification,the 6 single rock images are correctly identified and the probability of classification is greater than 96%.The recognition result of multi-class rock mixed images is good,and most of the classification probability is more than 80% and the location is accurate.The model can well distinguish the rock types with similar image features and occlusion.It proves that the model has strong robustness and generalization ability.更多还原