Abstract:The high dimensionality of hyperspectral images (HSIs) with significant levels of redundancy and noise is a pervasive problem that creates a heavy burden and many inconveniences in terms of accuracy and efficiency for all the subsequent applications. To address this problem, we proposed an adaptive-boundary adjustment-based groupwise band-categorization (BC) framework that classifies bands by the information contained within each band and redundant/noisy components without compromising the original content from the raw HSI. Since HSI labels are difficult to collect, an unsupervised spectral-spatial adaptive band-noise factor-based formulation is devised for BC based on the measurement of boundary movement/adjustment and band correlation. Robust clustering and boundary-adjustment-based methods involve two key issues: band correlation and band preference. Our proposed method contemplates band correlation and band synergy, and band preference is defined by the discriminative information contained in each band. According to different information-distribution levels, an adaptive mechanism is proposed to retain the spectral correlations in spectral dimensions and match the actual structure in spatial dimensions. The proposed method is based on the observation that clustering and boundary adjustment techniques are sensitive to noise and redundancy, which can effectively contribute to estimates and help classify the bands. Comprehensive experimental results on challenging airborne visible/infrared-imaging spectrometer datasets from Indian Pines, Moffett Field, and Salinas demonstrate the significant advancement in these two applications and show the reliability and effectiveness of the proposed framework without compromising the informative bands. Because of its usefulness, BC can be successfully applied to many practical applications of hyperspectral remote sensing. Effective application of the proposed framework is demonstrated by applying it on HSI classification and also for estimating the noise. The ability of the method to characterize the bands along with the noise estimation of HSIs is of significant benefit for subsequent remote-sensing techniques. (C) 2016 Society of Photo-Optical Instrumentation Engineers (SPIE)

Reducing Boundary Effects in a Kernel-Based Classifier

Boundary Detection and Sample Reduction for One-Class Support Vector Machines

A Spectral-Texture Kernel-Based Classification Method for Hyperspectral Images.

No tricks no bluff, focusing on localizing crisp boundaries in image media

Combining the FCM Classifier with Various Kernels to Handle Non-linearity of Class Boundaries

Beyond Difference: Learning Optimal Metric for Boundary Detection

Ab(3)C: Adaptive Boundary-Based Band-Categorization of Hyperspectral Images

Improved Robust Kernel Subspace for Object-Based Registration and Change Detection

Spectral-Spatial Boundary Detection in Hyperspectral Images

Binary Classification Algorithm with Class-Dependent Reject Cost

Region Labeling Algorithm Based on Boundary Tracking for Binary Image

A New Robust Approach for Remote Sensing Image Regional Classification

Land Cover Classification from Hyperspectral Images via Weighted Spatial-Spectral Kernel Collaborative Representation with Tikhonov Regularization

Boundary-constrained multi-scale segmentation method for remote sensing images

E-CRF: Embedded Conditional Random Field for Boundary-caused Class Weights Confusion in Semantic Segmentation

Improved camouflaged detection in the large-scale images and videos with minimum boundary contrast in detection technique

Boundary Attention: Learning curves, corners, junctions and grouping

Land cover classification algorithm based on multi-modal collaboration and boundary-guided fusion

Integration of Classification Methods for Improvement of Land-Cover Map Accuracy

Reducing the effects of misregistration on pixel-level change detection

Learning to predict crisp boundaries