基于深度迁移学习的煤岩显微组分识别算法

Coal rock microfraction identification method based on deep transfer learning

针对VGG-16网络模型对小数量的煤岩显微组分图像数据集识别准确度低、模型参数量过多的问题,提出一种基于迁移学习的小样本煤岩显微组分图像识别算法。算法首先改进了VGG-16分类网络模型,将Res2Net模块与VGG-16相结合,并用深度可分离卷积层代替VGG-16和Res2Net模块的原始卷积层,以此作为深度迁移识别网络预训练模型;然后迁移预训练模型中部分网络结构和参数,并结合优化的分类器模块完成网络的学习与优化。实验结果表明,在样本数据不充足的条件下,基于深度迁移学习识别网络模型识别准确率为96.33%,模型参数量为11.79 M。与其它网络模型相比,该方法在小数量的煤岩显微组分图像识别中具有明显的优越性。

To solve the problems of low accuracy of the VGG-16 network model in recognizing small number of coal rock microfraction images and excessive number of model parameters,an algorithm for recognizing small sample coal rock microfraction images based on transfer learning is proposed.Firstly,the VGG-16 network model is improved.The Res2Net module is combined with VGG-16 and the original convolutional layers of the VGG-16 and Res2Net modules are replaced with depth-separable convolutional layers as a pre-training model for the deep transfer recognition network.Then some of the network structures and parameters in the pre-training model are transferred and combined with the optimized classification module to complete the learning and optimization of the network.The experimental results show that,under the condition of insufficient sample data,the recognition accuracy of the deep migration learning based recognition network model is 96.33%and the number of model parameters is 11.79 M.Compared with other network models,the proposed method has obvious advantages in the recognition of small number of microscopic components of coal rocks.