Hyperspectral Imagery Classification Based on Semi-Supervised Broad Learning System

Yi Kong,C Chen,Yuhu Cheng,Xuesong Wang

doi:10.3390/rs10050685

Abstract

Recently, deep learning-based methods have drawn increasing attention in hyperspectral imagery (HSI) classification, due to their strong nonlinear mapping capability. However, these methods suffer from a time-consuming training process because of many network parameters. In this paper, the concept of broad learning is introduced into HSI classification. Firstly, to make full use of abundant spectral and spatial information of hyperspectral imagery, hierarchical guidance filtering is performing on the original HSI to get its spectral-spatial representation. Then, the class-probability structure is incorporated into the broad learning model to obtain a semi-supervised broad learning version, so that limited labeled samples and many unlabeled samples can be utilized simultaneously. Finally, the connecting weights of broad structure can be easily computed through the ridge regression approximation. Experimental results on three popular hyperspectral imagery datasets demonstrate that the proposed method can achieve better performance than deep learning-based methods and conventional classifiers.

Highlights

Hyperspectral imagery (HSI) captured by hyperspectral sensors has high spectral and spatial resolution, has a strong capability to distinguish surface objects [1]
The flowchart of HSI classification based on saubpoveervmisetehdodBsL. S (SBLS) is shown in Figure 1 and includes three steps: (1) the osrtiegpisnT: ah(le1)flHtohwSe Icohradigraitntoaafl HHaSSrIeI dclapatsrasoiaficrceeastpsioreondcebsabsseyeddbhoyniehSrieBarrLacSrhcihiscicsaahllogwgunuidiiadnnacFneigcfiuelrteefr1iinltagenr(dHiniGngFcl)u(tHdoeGgsetFth)trheeeto get the spectral-spspaetcitarlale-sxppatriealsseixopnresosfioHn SofI;H(S2I); t(h2)ethpesepuseduodolalabbeellss ooff uunnlalbaebleedlesdamsapmlespaleres oabrteainoebdtaviina ed via CP structure;CaPnsdtru(3ct)uSreB; aLnSd i(s3)tSrBaLinS eisdtrabiyneldabbyellaebdelesdamsamplpelessaannddccoorrrerspeosnpdoinngdlianbgelsl,aabsewlesl,l assuwnlealblelaesd unlabeled samples asnadmpcloersraensdpcoonrrdesipnogndpisnegupdseoudlaoblaeblesl.s
To evaluate the performance of the proposed SBLS on HSI classification, we investigate the following nine methods for comparison

Summary

Introduction

Hyperspectral imagery (HSI) captured by hyperspectral sensors has high spectral and spatial resolution, has a strong capability to distinguish surface objects [1]. Compared with conventional CNN, R-VCANet has a simpler structure and fewer network parameters, which demand fewer labeled training samples. Several experiments on hyperspectral datasets demonstrated that the classification accuracy of R-VCANet is higher than other deep learning methods such as R-PCANet and NSSNet. DL methods require complicated structural adjustment and a vast computation of network training. DL methods require complicated structural adjustment and a vast computation of network training Aiming at such problems, Chen and Liu [20] proposed a novel broad learning system (BLS) to offer an optional learning approach. (2) The class-probability structure is introduced into BLS for an extended semi-supervised BLS to make use of limited numbers of labeled samples as well as vast The mainCcoonnstidriebriungtiothniss, Sohfatoh, eist apl.a[2p8e]rprienscelnuteddea: c(l1a)ssT-poroobuabrilkitnyo(CwPl)esdtrguect,utrhe,iws hisichthceanfierxsptretrssiathl ewhere BLS is appliedreilnatiHIonnSsbIuemctwlmaesaesrniyf,eiaaccaHhtSisoaI mnclpatslaessiafikncsda.teiToanchhemcelpatrhssoovdpioaissapecrdloapsSsoB-speLrdoSbbacabasienlidtygomentaathrsiiexgm.hi-esrupHerSvIisceldasBsLiSfi(cSaBtLioS)n. accuracy and fasterTthreaminaiinngcosnpteriebdut.io(2n)s Tofhtehicslpaasps-epr rinocbluadbei:li(t1y) sTtoruoucrtukrneowisleidngter,otdhuisciesdthienfitorstBtLriSalfworhearne extended semi-supeBrLvSisiesdapBpLlieSdtion mHSaIkcelausssifiecoatfiolnimtaistkesd. nTuhempbroeprsosoedf lSaBbLeSlecadnsgaemt hpiglehseraHs SwI ecllalssaisficvataiostn unlabeled samples. accuracy and faster training speed. (2) The class-probability structure is introduced into BLS for an extended semi-supervised BLS to make use of limited numbers of labeled samples as well as vast

HSI Classification Based On SBLS

Class-Probability Structure

Comparative Experiments

Discussion

Conclusions