当前位置: X-MOL 学术Infrared Phys. Technol. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
CNN based Hyperspectral Image Classification using Un-supervised Band selection and Structure-Preserving Spatial Features
Infrared Physics & Technology ( IF 3.3 ) Pub Date : 2020-11-01 , DOI: 10.1016/j.infrared.2020.103457
Radhesyam Vaddi , Prabukumar Manoharan

Abstract Hyperspectral image (HSI) consists of hundreds of contiguous spectral bands, which can be used in the classification of different objects on earth. The inclusion of both spectral and as well as spatial features is essential for better classification accuracy. Extraction of spectral and spatial information without preserving the intrinsic structure of the data will downscale classification accuracy. To address this issue, we have proposed a method that uses unsupervised band selection called optimal neighboring reconstruction (ONR), which extracts a subset of spectral bands to linearly reconstruct the original data with minimum loss and Structure-Preserving Recursive Filter (SPRF) to extract spatial features. Then we have adopted Convolutional Neural Networks (CNN) with different sets of convolutional, pooling, and fully connected layers for classification of the data. To test the performance of proposed method, experiments are conducted with three benchmarks HSI data sets Indian pines, University of Pavia, and Salinas. These experiments reveal that the proposed method performed better classification accuracy over state-of-art methods in terms of standard metrics like Overall Accuracy (OA), Average Accuracy (AA), and kappa coefficient (k). The proposed method has attained OA's of 99.9%, 98.9%, and 99% for the three datasets, respectively.

中文翻译:

使用无监督波段选择和结构保留空间特征的基于 CNN 的高光谱图像分类

摘要 高光谱图像(HSI)由数百个连续的光谱带组成,可用于地球上不同物体的分类。包含光谱和空间特征对于提高分类精度至关重要。在不保留数据内在结构的情况下提取光谱和空间信息会降低分类精度。为了解决这个问题,我们提出了一种使用无监督频带选择的方法,称为最优相邻重建(ONR),它提取光谱带的子集以最小损失线性重建原始数据,并使用结构保留递归滤波器(SPRF)提取空间特征。然后我们采用了具有不同卷积、池化、和用于数据分类的全连接层。为了测试所提出方法的性能,使用三个基准 HSI 数据集印度松树、帕维亚大学和萨利纳斯进行了实验。这些实验表明,在总体准确度 (OA)、平均准确度 (AA) 和 kappa 系数 (k) 等标准指标方面,所提出的方法比最先进的方法具有更好的分类准确度。所提出的方法在三个数据集上分别达到了 99.9%、98.9% 和 99% 的 OA。这些实验表明,在总体准确度 (OA)、平均准确度 (AA) 和 kappa 系数 (k) 等标准指标方面,所提出的方法比最先进的方法具有更好的分类准确度。所提出的方法在三个数据集上分别达到了 99.9%、98.9% 和 99% 的 OA。这些实验表明,在总体准确度 (OA)、平均准确度 (AA) 和 kappa 系数 (k) 等标准指标方面,所提出的方法比最先进的方法具有更好的分类准确度。所提出的方法在三个数据集上分别达到了 99.9%、98.9% 和 99% 的 OA。
更新日期:2020-11-01
down
wechat
bug