Recent advances in airborne and spaceborne hyperspectral imaging technology have provided end users with rich spectral, spatial, temporal information. They made a plethora of applications feasible for the analysis large areas Earth?s surface. However, significant number factors-such as high dimensions size data, lack training samples, mixed pixels, light-scattering mechanisms acquisition process, different atmospheric geometric distortions-make such data inherently nonlinear complex, which...

Linear spectral unmixing is a commonly accepted approach to mixed-pixel classification in hyperspectral imagery. This involves two steps. First, find spectrally unique signatures of pure ground components, usually known as endmembers, and, second, express mixed pixels linear combinations endmember materials. Over the past years, several algorithms have been developed for autonomous and supervised extraction from data. Due lack data quantitative approaches substantiate new algorithms,...

Spectral mixture analysis provides an efficient mechanism for the interpretation and classification of remotely sensed multidimensional imagery. It aims to identify a set reference signatures (also known as endmembers) that can be used model reflectance spectrum at each pixel original image. Thus, modeling is carried out linear combination finite number ground components. Although spectral models have proved appropriate purpose large hyperspectral dataset subpixel analysis, few methods are...

Convolutional neural networks (CNNs) exhibit good performance in image processing tasks, pointing themselves as the current state-of-the-art of deep learning methods. However, intrinsic complexity remotely sensed hyperspectral images still limits many CNN models. The high dimensionality HSI data, together with underlying redundancy and noise, often makes standard approaches unable to generalize discriminative spectral-spatial features. Moreover, deeper architectures also find challenges when...

This work describes sequences of extended morphological transformations for filtering and classification high-dimensional remotely sensed hyperspectral datasets. The proposed approaches are based on the generalization concepts from mathematical morphology theory to multichannel imagery. A new vector organization scheme is described, fundamental operations defined by extension. Extended transformations, characterized simultaneously considering spatial spectral information contained in...

Convolutional neural networks (CNNs) have recently exhibited an excellent performance in hyperspectral image classification tasks. However, the straightforward CNN-based network architecture still finds obstacles when effectively exploiting relationships between imaging (HSI) features spectral-spatial domain, which is a key factor to deal with high level of complexity present remotely sensed HSI data. Despite fact that deeper architectures try mitigate these limitations, they also find...

Hyperspectral imaging is a widely used technique in remote sensing which an spectrometer collects hundreds of images (at different wavelength channels) for the same area on surface earth. In last two decades, several methods (unsupervised, supervised, and semisupervised) have been proposed to deal with hyperspectral image classification problem. Supervised techniques generally more popular, despite fact that it difficult collect labeled samples real scenarios. particular, deep neural...

This paper proposes a new method, called multilayer stacked covariance pooling (MSCP), for remote sensing scene classification. The innovative contribution of the proposed method is that it able to naturally combine feature maps, obtained by pretrained convolutional neural network (CNN) models. Specifically, MSCP-based classification framework consists following three steps. First, CNN model used extract maps. Then, maps are together, and matrix calculated features. Each entry resulting...

The classification of hyperspectral images (HSIs) using convolutional neural networks (CNNs) has recently drawn significant attention. However, it is important to address the potential overfitting problems that CNN-based methods suffer when dealing with HSIs. Unlike common natural images, HSIs are essentially three-order tensors which contain two spatial dimensions and one spectral dimension. As a result, exploiting both information very for HSI classification. This paper proposes new...

Deep neural networks (DNNs), including convolutional (CNNs) and residual (ResNets) models, are able to learn abstract representations from the input data by considering a deep hierarchy of layers that perform advanced feature extraction. The combination these models with visual attention techniques can assist identification most representative parts standpoint, obtained through more detailed filtering features extracted operational network. This is significant interest for analyzing remotely...

This paper proposes a novel end-to-end learning model, called skip-connected covariance (SCCov) network, for remote sensing scene classification (RSSC). The innovative contribution of this is to embed two modules into the traditional convolutional neural network (CNN) i.e., skip connections and pooling. advantages newly developed SCCov are twofold. First, by means connections, multi-resolution feature maps produced CNN combined together, which provides important benefits address presence...

In this paper, a novel local covariance matrix (CM) representation method is proposed to fully characterize the correlation among different spectral bands and spatial-contextual information in scene when conducting feature extraction (FE) from hyperspectral images (HSIs). Specifically, our first projects HSI into subspace, using maximum noise fraction method. Then, for each test pixel its most similar neighboring pixels (within spatial window) are clustered cosine distance measurement. The...

Super-resolution (SR) brings an excellent opportunity to improve a wide range of different remote sensing applications. SR techniques are concerned about increasing the image resolution while providing finer spatial details than those captured by original acquisition instrument. Therefore, particularly useful cope with demand imaging applications requiring fine resolution. Even though machine learning paradigms have been successfully applied in SR, more research is required process without...

Hyperspectral imaging (HSI) is a competitive remote sensing technique in several fields, from Earth observation to health, robotic vision, and quality control. Each HSI scene contains hundreds of (narrow) contiguous spectral bands. The amount data generated by devices often both solution problem for given application. Extracting information cubes complex computationally demanding problem. To tackle this challenge, convolutional neural networks (CNNs) have been widely applied classification....

Remotely sensed hyperspectral imaging instruments are capable of collecting hundreds images corresponding to different wavelength channels for the same area on surface Earth. For instance, NASA is continuously gathering high-dimensional image data with such as Jet Propulsion Laboratory's Airborne Visible-Infrared Imaging Spectrometer (AVIRIS). This advanced sensor Earth observation records visible and near-infrared spectrum reflected light using more than 200 spectral bands, thus producing a...

Denoising is a common preprocessing step prior to the analysis and interpretation of hyperspectral images (HSIs). However, vast majority methods typically adopted for HSI denoising exploit architectures originally developed grayscale or RGB images, exhibiting limitations when processing high-dimensional data cubes. In particular, traditional do not take into account high spectral correlation between adjacent bands in HSIs, which leads unsatisfactory performance as rich information present...

Convolutional neural networks (CNNs) have become a powerful tool for remotely sensed hyperspectral image (HSI) classification due to their great generalization ability and high accuracy.However, owing the huge amount of parameters that need be learned complex nature HSI data itself, these approaches must deal with important problem overfitting, which can lead inadequate loss accuracy.In order mitigate this problem, in letter we adopt random occlusion, recently developed augmentation (DA)...

The availability of diverse data captured over the same region makes it possible to develop multisensor fusion techniques further improve discrimination ability classifiers. In this paper, a new sparse and low-rank technique is proposed for hyperspectral light detection ranging (LiDAR)-derived features. consists two main steps. First, extinction profiles are used extract spatial elevation information from LiDAR data, respectively. Then, utilized estimate fused features extracted ones that...

As an unsupervised learning tool, autoencoder has been widely applied in many fields. In this letter, we propose a new robust unmixing algorithm that is based on stacked nonnegative sparse autoencoders (NNSAEs) for hyperspectral data with outliers and low signal-to-noise ratio. The proposed network contains two main steps. the first step, series of NNSAE used to detect data. second final performed achieve endmember signatures abundance fractions. By taking advantage from autoencoding,...

Deep neural networks (DNNs) have emerged as a relevant tool for the classification of remotely sensed hyperspectral images (HSIs), with convolutional (CNNs) being current state-of-the-art in many tasks. However, deep CNNs present several limitations context HSI supervised classification. Although models are able to extract better and more abstract features, number parameters that must be fine-tuned requires large amount training data (using small learning rates) order avoid overfitting...