迁移深度学习地基云图自动识别的网络微调学习过程

An Ground Cloud Image Recognition Method Using AlexNet Convolution Neural Network

研究了一种将迁移学习引入到地基云图自动识别深度学习网络中的学习过程,其中深度学习网络采用AlexNet经典网络模型,数据集采用ImageNet样本库进行预训练,学习过程中采用微调操作对网络的权值进行最佳调整。通过对10类地基云图的仿真实验,可以看出,由于云图类别较多,分类任务较难,将迁移学习和微调方法引入到深度学习地基云图自动识别中,是可行和有效的。该方法的有效实施,为深度学习在高精度的地基云图分类以及其他领域图像识别奠定了技术基础。

A learning process is studied in which the migration learning is introduced into the deep learning network of automatic recognition of ground-based cloud map.The deep learning network uses the AlexNet classic network model,the dataset uses the ImageNet sample library for pre training,and the network weight is optimally adjusted by fine-tuning operation in the learning process.Through the simulation experiment of 10 kinds of ground-based cloud images,it can be seen that because there are many kinds of cloud images and the classification task is difficult,it is feasible and effective to introduce the migration learning and fine-tuning method into the automatic recognition of deep learning ground-based cloud images.The effective implementation of this method lays a technical foundation for deep learning in high-precision cloud image classification and image recognition in other fields.