In this article, we propose a weakly supervised low-rank representation (WSLRR) method for hyperspectral anomaly detection (HAD), which formulates deep learning-based HAD into low-lank optimization problem not only characterizing the complex and diverse background in real HSIs but also obtaining relatively strong supervision information. Different from existing unsupervised methods, first model manner, achieves better performance without prior information is restrained by richly correct...

Hyperspectral target detection is critical in both military and civilian applications. However, it a challenging task due to the complexity of background limited samples hyperspectral images (HSIs). In this article, we propose novel learning model, called based on suppression constraint characterize high-dimensional spectral vectors. Considering insufficient samples, model trained only accurately learn distribution. Then discrepancy between reconstructed original HSIs are examined spot...

Recently, the convolutional neural network (CNN)-based approach for on-satellite ship detection in synthetic aperture radar (SAR) images has received increasing attention since it does not rely on predefined imagery features and distributions that are required conventional methods. To achieve high accuracy, most of existing CNN-based methods leverage complex off-the-shelf CNN models optical imagery. Unfortunately, this usually leads to expensive computational cost, which is hard process real...

In this article, we propose a novel filter pruning method for deep learning networks by calculating the learned representation median (RM) in frequency domain (LRMF). contrast to existing methods that remove relatively unimportant filters spatial domain, our newly proposed approach emphasizes removal of absolutely domain. Through extensive experiments, observed criterion "relative unimportance" cannot be generalized well and discrete cosine transform (DCT) can eliminate redundancy emphasize...

Convolutional neural networks (CNNs) have been successfully employed in remote sensing image classification because of their robust feature representation for different visual tasks and powerful graphics processing units (GPUs). The attendant problem is that high computational cost memory footprint hindering the application CNNs applications resource- time-sensitive situations. Based on practical deployment requirements, we pioneer a pruning-quantization joint learning model compression...

Hyperspectral anomaly detection (HAD) aims to identify samples with unknown atypical spectra from the background. Deep learning (DL)-based methods, particularly autoencoders (AEs), have proven effective in uncovering underlying profiles for HAD. However, real-world applications of hyperspectral images (HSIs), complex background land-covers and corruptions are common, leading two issues: 1) A low-dimensional manifold characterized by DL-based HAD methods can only reveal a few variation...

Neurologically, filter pruning is a procedure of forgetting and remembering recovering. Prevailing methods directly forget less important information from an unrobust baseline at first expect to minimize the performance sacrifice. However, unsaturated base imposes ceiling on slimmed model leading suboptimal performance. And significantly would cause unrecoverable loss. Here, we design novel paradigm termed Remembering Enhancement Entropy-based Asymptotic Forgetting (REAF). Inspired by...

Support vector regression (SVR) is a powerful kernel-based prediction algorithm that performs excellently in various application scenarios. However, for real-world data, the general SVR often fails to achieve good predictive performance due its inability assess feature contribution accurately. Feature weighting suitable solution address this issue, applying correlation measurement methods obtain reasonable weights features based on their contributions output. In paper, idea of...

Traditional clustering-based band selection (BS) methods treat each as individuals, and is conducted by enlarging the difference between clusters, which leads to loss of interaction information saliency evaluation. In this article, we propose a BS method named rank-aware generative adversarial network (R-GAN) address these problems. First, centralized reference feature extraction (FE) with GAN aids R-GAN combine interpretability interband relevance. Then, refined estimation provided...

The hyperfine structure absorption lines of neutral hydrogen in spectra high-redshift radio sources, known collectively as the 21-cm forest, have been demonstrated a sensitive probe to small-scale structures governed by dark matter (DM) properties, well thermal history intergalactic medium regulated first galaxies during epoch reionization. By statistically analyzing these spectral features, one-dimensional (1D) power spectrum forest can effectively break parameter degeneracies and constrain...

Despite the great potential of convolutional neural networks (CNNs) in various tasks, resource-hungry nature greatly hinders their wide deployment cost-sensitive and low-powered scenarios, especially applications remote sensing. Existing model pruning approaches, implemented by a "subtraction" operation, impose performance ceiling on slimmed model. Self-knowledge distillation (Self-KD) resorts to auxiliary that are only active training phase for improvement. However, knowledge is holistic...