Bands Sensitive Convolutional Network for Hyperspectral Image Classification

Ran, Lingyan; Zhang, Yanning; Wei, Wei; Yang, Tao

doi:10.1145/3007669.3007707

Cited by 10 publications

(11 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Dimensionality reduction can significantly improve results in HSI processing. Ran et al 192 split the spectrum into groups based on correlation, then apply m CNNs in parallel, one for each band group. The CNN output are concatenated and then classified via a two-layer FC-CNN.…”

Section: Dimensionality Reductionmentioning

confidence: 99%

“…3D (depth and shape) analysis [107][108][109][110][111][112][113][114][115] Advanced driver assistance systems [116][117][118][119][120] Animal detection 121 Anomaly detection 122 Automated Target Recognition [123][124][125][126][127][128][129][130][131][132][133][134] Change detection [135][136][137][138][139] Classification Data fusion 191 Dimensionality reduction 192,193 Disaster analysis/assessment 194 Environment and water analysis [195][196][197][198] Geo-information extraction 199 Human detection [200][201][202][203] Image denoising/enhancement 204,…”

Section: References Area Referencesmentioning

confidence: 99%

See 1 more Smart Citation

Comprehensive survey of deep learning in remote sensing: theories, tools, and challenges for the community

Ball

Anderson

Chan

2017

J. Appl. Remote Sens

538

331

View full text Add to dashboard Cite

Abstract. In recent years, deep learning (DL), a rebranding of neural networks (NNs), has risen to the top in numerous areas, namely computer vision (CV), speech recognition, and natural language processing. Whereas remote sensing (RS) possesses a number of unique challenges, primarily related to sensors and applications, inevitably RS draws from many of the same theories as CV, e.g., statistics, fusion, and machine learning, to name a few. This means that the RS community should not only be aware of advancements such as DL, but also be leading researchers in this area. Herein, we provide the most comprehensive survey of state-of-the-art RS DL research. We also review recent new developments in the DL field that can be used in DL for RS. Namely, we focus on theories, tools, and challenges for the RS community. Specifically, we focus on unsolved challenges and opportunities as they relate to (i) inadequate data sets, (ii) human-understandable solutions for modeling physical phenomena, (iii) big data, (iv) nontraditional heterogeneous data sources, (v) DL architectures and learning algorithms for spectral, spatial, and temporal data, (vi) transfer learning, (vii) an improved theoretical understanding of DL systems, (viii) high barriers to entry, and (ix) training and optimizing the DL. © The Authors.Published by SPIE under a Creative Commons Attribution 3.0 Unported License. Distribution or reproduction of this work in whole or in part requires full attribution of the original publication, including its DOI.

show abstract

Section: Dimensionality Reductionmentioning

confidence: 99%

Section: References Area Referencesmentioning

confidence: 99%

Comprehensive survey of deep learning in remote sensing: theories, tools, and challenges for the community

Ball

Anderson

Chan

2017

J. Appl. Remote Sens

538

331

View full text Add to dashboard Cite

show abstract

“…The slicing of the original features greatly reduces the complexity of network design and improves the efficiency of data abstraction. Very recently, Ran et al [ 47 ] propose the band-sensitive network (BsNet) for feature extraction from correlated band groups, with each band group earning a respective classification confidence. The BsNet label prediction is based on all available band group classification confidences.…”

Section: Related Workmentioning

confidence: 99%

A Hyperspectral Image Classification Framework with Spatial Pixel Pair Features

Ran

Zhang

Wei

et al. 2017

Sensors

Self Cite

View full text Add to dashboard Cite

During recent years, convolutional neural network (CNN)-based methods have been widely applied to hyperspectral image (HSI) classification by mostly mining the spectral variabilities. However, the spatial consistency in HSI is rarely discussed except as an extra convolutional channel. Very recently, the development of pixel pair features (PPF) for HSI classification offers a new way of incorporating spatial information. In this paper, we first propose an improved PPF-style feature, the spatial pixel pair feature (SPPF), that better exploits both the spatial/contextual information and spectral information. On top of the new SPPF, we further propose a flexible multi-stream CNN-based classification framework that is compatible with multiple in-stream sub-network designs. The proposed SPPF is different from the original PPF in its paring pixel selection strategy: only pixels immediately adjacent to the central one are eligible, therefore imposing stronger spatial regularization. Additionally, with off-the-shelf classification sub-network designs, the proposed multi-stream, late-fusion CNN-based framework outperforms competing ones without requiring extensive network configuration tuning. Experimental results on three publicly available datasets demonstrate the performance of the proposed SPPF-based HSI classification framework.

show abstract

“…The correlation matrix of the Indian Pines dataset is showed as the figure 3, and where the light color indicates a higher degree of correlation. Neighbouring bands with a correlation coefficient difference smaller than a threshold τ are grouped together [20], and in this paper τ =0.7 . According to the degree of correlation of the 200 bands, they can be divided into four band groups: 1~35, 36~77, 78~102, 103~200.…”

Section: Regroup the Bands Based On The Similarity Between Bandsmentioning

confidence: 99%

“…In this paper, we mainly modify the CNN, so we will compare our algorithm with other CNN algorithm, including the spectral classification [20]and the joint spectral-spatial classification 3D-CNN [14]. 3D-CNN and our algorithm are set as the same network configuration.…”

Section: Figure 4: Correlation Matrix Of Indian Pinesmentioning

confidence: 99%

The Classification of Hyperspectral Images Based on Band-Grouping and Convolutional Neural Network

Fu¹,

Gu²

2017

Proceedings of the 7th International Conference on Computer Engineering and Networks — PoS(CENet2017)

View full text Add to dashboard Cite

For the classification of hyperspectral images, a classification algorithm based on band grouping and three-dimensional convolutional neural network(3D-CNN-BG) is proposed. The algorithm uses the correlation matrix of hyperspectral images to determine the similarity of bands, and the high similarity bands are grouped together. Then, every bands group is extracted spatial-spectral feature using 3D convolutional neural network. Finally, the high-level feature of every 3D-CNN is stacked together trained by the classifier. With band regroup, the feature in the similar group can be extracted fully by the 3D-CNN, which is good for classification. The experiments showed that compared with other hyperspectral images classification algorithms based on CNN, the proposed algorithm converges more faster, has less training parameters, and acquires higher classification accuracy. After training 100 times, the overall classification accuracy on the indian pines data set can reach 97.42%, increased by approximately 2% ~ 5%, which shows the strong practicability of proposed algorithm in hyperspectral application.

show abstract

Bands Sensitive Convolutional Network for Hyperspectral Image Classification

Cited by 10 publications

References 21 publications

Comprehensive survey of deep learning in remote sensing: theories, tools, and challenges for the community

Comprehensive survey of deep learning in remote sensing: theories, tools, and challenges for the community

A Hyperspectral Image Classification Framework with Spatial Pixel Pair Features

The Classification of Hyperspectral Images Based on Band-Grouping and Convolutional Neural Network

Contact Info

Product

Resources

About