基于卷积神经网络的高光谱图像分类方法

Classification of hyperspectral images based on convolutional neural networks

为解决在样本有限的情况下高光谱图像分类精度不高的问题,提出一种基于卷积神经网络的高光谱遥感图像分类方法。引入滤波、增加虚拟样本、标准化等预处理技术,使分类模型对地物样本种类和数量的敏感度降低;通过对梯度下降法和学习率计算方法进行优化,降低计算复杂度和计算时间;设计符合高光谱数据特点的网络结构,提高分类方法的泛化性。实验结果表明,与传统分类方法进行比较,该方法有较高的分类精度。

To address the problem of classification accuracy of hyperspectral images under limited samples,a method for hyperspectral remote sensing image classification based on convolutional neural network was proposed.By filtering,adding virtual samples and standardizing,the sensitivity of classification model to the classification of objects and quantity of classification objects was reduced.The computational complexity and computation time were reduced by optimizing the gradient descent method and the learning rate calculation method.By setting up a suitable network structure,the generalization was improved.Experimental results show that compared with the traditional classification methods,the method using convolutional neural network has more competitive results in computational efficiency and classification accuracy.