基于改进CNN的高光谱图像分类算法研究

打开文本图片集

Hyperspectral Image Classification Based on Improved Convolutional NeuralNetwork

LIU Fanghua WANG Huiqing （Zhengzhou Universityof Light Industryt， Zhengzhou 45Oooo，China）

Abstract：[Purposes] Hyperspectral remote sensing technology，which integrates imaging and spectral techniques to capture multi-dimensional information，has been widely used in the field of remote sensing. To address the issues of abundant spectral information redundancy and low classification accuracy in hyperspectral remote sensing images，this paper proposes an improved convolutional neural network （CNN）-based algorithm for hyperspectral image classification.[Methods] Batch normalization layers were incorporated after the convolutional layers in both the 3D-CNN and the multi-scale 3D-CNN （MS3D-CNN） to stabilize the output of each convolutional layer through normalization.Additionally，the activation function was replaced with Leaky ReLU，which assigns a non-zero slope to all negative values-avoiding complete deactivation when the input is negative.[Findings] Experimental results show thatboth improved models achieve higher classificationaccuracy on the Indian Pines and Pavia University datasets.[Conclusions] The two enhanced algorithms provide effective technical support for hyperspectral image classification and demonstrate significant practical potential for improving clasification accuracy.

Keywords： remote sensing technology; 3D convolutional neural network; hyperspectral image classification

0 引言

高光谱图像（HyperspectralImage，HSI）是三维图像，其最重要的性质是空谱合一。（剩余6147字）