In this paper, a novel framework, MixerSENet, is introduced for hyperspectral image (HSI) classification, designed to address the challenges of computational efficiency and limited labeled data. The proposed model processes hyperspectral image patches while maintaining consistent size and resolution throughout the network, effectively decoupling the mixing of spatial and channel dimensions. Notably, MixerSENet is lightweight and computationally efficient, requiring fewer parameters compared to traditional models, making it suitable for resource-constrained environments. A squeeze and excitation block is incorporated into the model to refine feature extraction, enhancing the network’s ability to capture more informative features. Experimental results on two benchmark datasets demonstrate that MixerSENet achieves superior performance, reaching an overall accuracy (OA) of 82.47% on Houston13 dataset and 96.70% on the Qingyun dataset, outperforming state-of-the-art methods including 3D-CNN, HybridKAN, HSIFormer, SimPoolFormer, and MorphMamba. Furthermore, a detailed analysis of computational efficiency shows that MixerSENet achieves a favorable balance between accuracy and efficiency, with only 53,146 parameters and an low inference time, confirming its practicality for real-world applications. At publication, source code will be publicly available at https://github.com/mqalkhatib/MixerSENet.

Recently, convolutional neural networks (CNNs) have become popular for image classification due to their effectiveness in computer vision tasks. Now, researchers are exploring the potential of vision transformers (ViTs) in remote sensing and Earth observation. However, traditional real-valued networks often overlook important phase information in complex-valued (CV) data such as polarimetric synthetic aperture radar (PolSAR) data. To address this, new CV deep architectures have emerged. HybridCVNet, a novel hybrid network, blends CV convolutional neural network (CV-CNN) and CV vision transformer (CV-ViT) techniques. It efficiently combines CV 3-D and 2-D CNNs as feature extractors, enhancing PolSAR image classification by extracting complementary information and effectively leveraging interdependencies within the data. Experimental results from widely used PolSAR datasets show HybridCVNet outperforms other methods, achieving an overall accuracy (OA) of 97.39% on the Flevoland dataset and showing promise even with just a 1% sampling ratio, with a Kappa value of 0.972 on the San Francisco dataset. The source code is accessible throughhttps://github.com/mqalkhatib/HybridCVNet.

Hyperspectral Imagery (HSI) has great importance in industrial remote sensing applications, such as geological exploration and soil mapping. HSI has high spectral resolution, which gives each object a unique spectral response, making them easily identifiable. Nonetheless, their spatial resolution is compromised due to sensor limitation, which hinders utilizing HSI to their full potential. This paper deals with the spatial enhancement of HSI using Single Image Super Resolution (SISR) approaches. One of the main challenges in this area of research is preserving the spectral signature of HSI while improving the spatial resolution simultaneously. To tackle this challenge, we propose a 3D Wide Residual Convolutional Neural Network (3D-WRCNN) model that effectively utilizes the principle of wide activation to enhance feature propagation throughout the network. Residual connections are also deployed to boost image reconstruction and information sharing between the layers to reduce overfitting. Furthermore, this study incorporates and demonstrates the usage of Bayesian-optimized hybrid loss function to further improve the performance of the 3D-WRCNN. The quantitative and qualitative evaluation indicate that the proposed approach prevails over other state-of-the-art approaches. The implementation of the proposed model is provided in this repository: https://github.com/NourO93/SISR_Library

Hyperspectral Single Image Super Resolution is an important field of research due to the low spatial resolution of Hyperspectral Images (HSI) that limits their usability. Deep Convolutional Neural Networks (DCNNs) have been commonly used for SISR tasks, however, a large dataset is typically needed for training. Because of HSI data scarcity, training DCNNs for HSI-SISR becomes a challenging task. In this study, HSI data scarcity problem is tackled from an SISR perspective via Data Augmentation (DA). Several DA techniques are reviewed in this context, and a new DA technique called CutMixBlur is introduced. The best techniques are decided based on maximizing Peak Signal-to-Noise Ratio (PSNR) and Structure Similarity Index Measurement (SSIM) while minimizing Spectral Angle Mapper (SAM). The aim is to enhance HSI spatially without distorting their unique spectral signature. Experiments on Pavia University and Indian Pines datasets show that CutMixBlur boosts all quality metrics. Additionally, applying DA techniques by randomly stacking their effect significantly enhances the performance of SISR DCNNs, particularly the 3D-SRCNN.

Accurate flood mapping plays a critical role in disaster management, allowing for effective response and mitigation efforts. Thus, researchers seek to boost the accuracy of flood mapping algorithms, especially in terms of generalization capability and minimizing False Positive and False Negative detection. This paper presents a robust flood mapping algorithm from SAR images via Deep Convolutional Neural Network (DCNN) that follows encoder-decoder scheme. By introducing Bidirectional Convolutional LSTM (ConvLSTM) layers into its architecture, the proposed Temporal-Spatial Encoder-Decoder Network (TSEDN) network is able to extract temporal information and produce more accurate change maps. The training and testing are carried using OMBRIA dataset, which is known to be challenging to train. The proposed network is evaluated and compared to other state-of-the-art approaches in terms of Overall Accuracy (OA), Precision, Recall, and mean Intersection over Union (mIoU).

This paper proposes a novel multi-branch feature fusion network for PolSAR image classification and interpretation. It is built using Complex-valued Convolutional Neural Networks (CV-CNNs). The proposed approach utilizes extraction of polarimetric features at each branch to achieve high classification accuracy. Moreover, Squeeze and Excitation (SE) is also introduced within the model’s architecture. SE block improves channel interdependencies with almost no additional computational cost. The proposed approach is tested and evaluated using Flevoland benchmark dataset. Experiments demonstrate the effectiveness of the proposed attention based shallow to deep CV-CNN model for PolSAR image classification in terms of Kappa Coefficient (k), Overall Accuracy (OA), and Average Accuracy (AA) metrics.

Hyperspectral Imagery (HSI) classification is an important research area in remote sensing community due to its high efficiency in accurately analyzing ground features by assigning a class label to each pixel. This paper explores the use of Band Subset selection (BSS) methods as Dimensionality Reduction (DR) pre-processing stage for HSI classification, and compares them to Principal Component Analysis (PCA) approach. BSS is the problem of selecting the most independent bands in HSI cube. Classification is then performed using a proposed multi-branch HydraNet model that combines 1D, 2D, and 3D convolution. HydraNet is trained and tested using the benchmark Pavia University dataset, and the results are evaluated using Kappa and Overall Accuracy. Experimental results show positive indications of the network’s performance, especially when compared to other state-of-the-art CNN networks.

Single Image Super Resolution (SISR) refers to the spatial enhancement of an image from a single Low Resolution (LR) observation. This topic is of particular interest to remote sensing community, especially in the area of Hyperspectral Imagery (HSI) due to their high spectral resolution but limited spatial resolution. Enhancing the spatial resolution of HSI is a pre-requisite that boosts the accuracy of other image processing tasks, such as object detection and classification. This paper deals with SISR of HSI through the 3D expansion of Robust UNet (RUNet). The network is developed, trained, and tested over two datasets, and compared against the original 2D-RUNet and other state-of-the-art approaches. Quantitative and qualitative evaluation show the superiority of 3D-RUNet and its ability to preserve the spectral fidelity of the enhanced HSI.

This paper explores the use of unsupervised band subset selection (BSS) methods as a dimensionality reduction preprocessing stage in SLIC superpixel segmentation (BSSSLIC). Several methods for column subset selection (CSS) are used for unsupervised band subset selection and the performance of the corresponding BSS-SLIC combination is studied. CSS is the problem of selecting the most independent columns of a matrix. BSS-SLIC superpixel segmentation results are evaluated in terms of the homogeneity of the resulting superpixels. Numerical experiments with HYDICE Urban and ROSIS Pavia data sets are used to study the performance of different BSS-SLIC algorithms. The quality of the resulting segmentation is evaluated by looking at the fraction of the total number of superpixels that are homogeneous. BSS-SLIC results in the higher percentage of homogeneous superpixels when compared with SLIC using all bands.

In this paper, a superpixel-based hyperspectral image unmixing is proposed. First, superpixel segmentation is applied to the image. A low dimensional representation of the image is created by representing each superpixel by its mean spectra. Second, the superpixel image is segmented into regions. Endmember extraction is applied to each region to extract local endmembers. Abundances are computed over the full image using the extracted endmembers. The approach is compared with the global unmixing and the local unmixing using the original image. Experimental results are presented using the HYDICE Urban hyperspectral image.

This paper presents a modification to the cNMF for unmixing where the image is first segmented and the cNMF is applied to individual segments for endmember extraction. Extracted spectral endmembers from individual segments are clustered in endmember classes to describe the entire image. The approach is compared with the global cNMF. The segmentation-based cNMF better captures subtle differences between spectrally similar classes. Experimental results are presented using an image chip from an AVIRIS image collected over AP Hill.