Deep Learning for Land Cover Classification Using Only a Few Bands

Chiman Kwan,Jiang Li,Daniel Perez,Bulent Ayhan,Yan Lu,Antonio Plaza,Bence Budavari,Sergio Bernabe

doi:10.3390/rs12122000

Abstract

There is an emerging interest in using hyperspectral data for land cover classification. The motivation behind using hyperspectral data is the notion that increasing the number of narrowband spectral channels would provide richer spectral information and thus help improve the land cover classification performance. Although hyperspectral data with hundreds of channels provide detailed spectral signatures, the curse of dimensionality might lead to degradation in the land cover classification performance. Moreover, in some practical applications, hyperspectral data may not be available due to cost, data storage, or bandwidth issues, and RGB and near infrared (NIR) could be the only image bands available for land cover classification. Light detection and ranging (LiDAR) data is another type of data to assist land cover classification especially if the land covers of interest have different heights. In this paper, we examined the performance of two Convolutional Neural Network (CNN)-based deep learning algorithms for land cover classification using only four bands (RGB+NIR) and five bands (RGB+NIR+LiDAR), where these limited number of image bands were augmented using Extended Multi-attribute Profiles (EMAP). The deep learning algorithms were applied to a well-known dataset used in the 2013 IEEE Geoscience and Remote Sensing Society (GRSS) Data Fusion Contest. With EMAP augmentation, the two deep learning algorithms were observed to achieve better land cover classification performance using only four bands as compared to that using all 144 hyperspectral bands.

Highlights

We summarized the evaluation of land cover classification performance of two Convolutional Neural Network (CNN)-based deep learning algorithms when only a few bands, namely RGB+near infrared (NIR) bands and RGB+NIR+light detection and ranging (LiDAR) bands, were available, and when Extended Multi-attribute Profiles (EMAP) was applied to those limited bands to generate the augmented bands
For classification performance evaluation of the deep learning methods with EMAP-augmented 3b.a1n. dOsuarnCduastlol motihzeedr bCaNnNd cRoemsubltisnations, we used overall accuracy, OA, which is the ratio of the sum of correTcatlbylecl2assshiofiweds opuixreclsusfrtoommiazleldclCasNsNes mtootdheelt[o4t8a]l cnluasmsibfiecraotifopnixreeslsulitns tfhoer tthesettdesattad.aItnasaedtdwitiitohnsitxo dOifAfe,rwenetaslestosgoefnimeraagteedbaavnedrsa.gTehaecsceusriaxcdy,ifAfeAre, nvtaslueteswofhiimchacgoerbreasnpdosnadrse:to(at)hReGaBve+rNaIgRe,lenslRdls-GskanwBcoc+iuwtNhrnIaERcKM+ieaLAspi,DPpwAa, h1(R1iKc×(h)5c4iobs=eaa4ffinl4sdc)os,ie)(k,dnn(t)co5[)w5541nE4].MEasMA‘bPAa-Pala-uangucmgedmenaetncecdteudbraabcnaydn’s.d(TsRh(GReBGl+aBNst+INpReI+RrLfoibDramAndaRnsbcaaeunmgdmsetaerunicgtemwdaetnsotte4hd4e t3o. 5R5esbualntds s with EMAP), (e) 144 hyperspectral bands, and (f) 144 hyperspectral bands+LiDAR (145 bands)
We investigated the classification performance of two deep learning methods for land cover classification, where only four bands (RGB+NIR) and five bands (RGB+NIR+LiDAR) were used in terms of EMAP augmentation

Summary

Introduction

Hyperspectral data have been used for chemical agent detection and classification [1,2], small target detection [3,4], fire damage assessment [5,6], anomaly detection [7,8,9,10,11,12,13], border monitoring [14], change detection [15,16,17,18], and mineral map abundance estimation on Mars [19,20]. There is an increasing interest in adapting deep learning methods for land cover classification after several breakthroughs have been reported in a variety of computer vision tasks such as image classification, object detection and tracking, and semantic segmentation. The authors of [28] described an unsupervised feature extraction framework, with the name “patch-to-patch convolutional neural network (PToP CNN)”, which was used for land cover classification using hyperspectral and LiDAR data. The performances of deep learning models were evaluated against traditional approaches, and the authors concluded that the deep-learning-based methods provided an end-to-end solution and showed better performance than the traditional pixel-based methods by utilizing both spatial and spectral information, whereas traditional pixel-based methods resulted in salt-and-pepper-type noisy land cover map estimates. A number of other works have shown that semantic segmentation classification with deep learning methods are quite promising in land cover classification [32,33,34,35]

Methods

Results

Conclusion