采用神经架构搜索的高光谱图像深度学习分类方法

Hyperspectral Image Classification with Neural Architecture Search-Based Deep Learning Methods

大多数深度网络架构都是人工手动设计的,过程耗费精力且易出错;自动的神经架构搜索和优化学习方法引起了学者的广泛关注;自动架构工程在高光谱图像分类任务中仍然鲜有研究。基于此,提出了一种快速且自动化的深度网络模型构建和生成方法,并将其应用于高光谱图像分类任务。实验表明,相较于传统人工设计的深度卷积网络,该方法的性能更优异。

Most deep network architectures are manually designed by experts,and the procedure is laborious and errorprone.Moreover,more and more scholars pay attention to the automatic neural architecture search and optimization theory.However,the focused technique is still rarely studied in the task of hyperspectral image classification.In this case,we present a fast and automatic method to construct and generate deep learning models,and apply it to the task of hyperspectral image classification.Experiments show that the method has better performance than that of hand-designed deep convolutional networks.