Three-Dimensional ResNeXt Network Using Feature Fusion and Label Smoothing for Hyperspectral Image Classification.

Peida Wu,Ziguan Cui,Feng Liu,Zongliang Gan

doi:10.3390/s20061652

Peida Wu, Ziguan Cui + Show 2 more

Open Access

https://doi.org/10.3390/s20061652

Copy DOI

Abstract

In recent years, deep learning methods have been widely used in the hyperspectral image (HSI) classification tasks. Among them, spectral-spatial combined methods based on the three-dimensional (3-D) convolution have shown good performance. However, because of the three-dimensional convolution, increasing network depth will result in a dramatic rise in the number of parameters. In addition, the previous methods do not make full use of spectral information. They mostly use the data after dimensionality reduction directly as the input of networks, which result in poor classification ability in some categories with small numbers of samples. To address the above two issues, in this paper, we designed an end-to-end 3D-ResNeXt network which adopts feature fusion and label smoothing strategy further. On the one hand, the residual connections and split-transform-merge strategy can alleviate the declining-accuracy phenomenon and decrease the number of parameters. We can adjust the hyperparameter cardinality instead of the network depth to extract more discriminative features of HSIs and improve the classification accuracy. On the other hand, in order to improve the classification accuracies of classes with small numbers of samples, we enrich the input of the 3D-ResNeXt spectral-spatial feature learning network by additional spectral feature learning, and finally use a loss function modified by label smoothing strategy to solve the imbalance of classes. The experimental results on three popular HSI datasets demonstrate the superiority of our proposed network and an effective improvement in the accuracies especially for the classes with small numbers of training samples.

Highlights

With the rapid development of hyperspectral imaging technology, the increasing number and the higher quality of available hyperspectral data make hyperspectral image processing a critical technique in numerous practical applications, such as vegetation ecology [1], atmosphere science [1], geology and mineral resources [2,3], ocean research [4], and precision agriculture [5]
To evaluate the hyperspectral image (HSI) classification performance of the proposed model, we compare the proposed model with recent representative HSI classification models which have been introduced in Section 1, such as support vector machine (SVM) [12], rank-1 feedforward neural network (FNN) [14], 3D-convolutional neural network (CNN) [21], spectral-spatial residual network (SSRN) [26], and 3D-residual network (ResNet) [34]
The experiment found that the cardinality has an obvious influence on the overall accuracies (OA), showing a wavelike trend, whereas the parameter quantity, training time, and test time all HSI

Summary

Introduction

With the rapid development of hyperspectral imaging technology, the increasing number and the higher quality of available hyperspectral data make hyperspectral image processing a critical technique in numerous practical applications, such as vegetation ecology [1], atmosphere science [1], geology and mineral resources [2,3], ocean research [4], and precision agriculture [5]. Traditional machine learning (ML) based pixelwise HSI classification methods mainly consist of two steps: feature engineering and classifier training [7]. Feature engineering methods are used to reduce the high dimensionality of HSI pixels, and extract the most representative features or select informative spectral bands [8]. These selected features from the first step are trained in the classifiers through nonlinear transformation [9]. Among those classifiers, support vector machine (SVM) is the most widely used one for HSI classification tasks [10,11,12,13].

Results

Discussion

Conclusion