The joint use of hyperspectral image (HSI) and light detection ranging (LiDAR) data has gained significant performance on land-cover classification. Although spatial-spectral feature learning methods based convolutional neural networks (CNNs) transformer have achieved prominent advances, contextual information described by fixed kernels all self-attention heads selected limited ability to characterize the detailed non-redundant features land-covers multimodal data. In this paper, a...

Abstract Conventional optical flow estimation methods typically recover two‐dimensional motion from RGB image sequences. Recently, due to the rise and widespread use of spike cameras, learning spiking cameras has become a hot topic in field estimation. Although existing have been designed learn by designing feature processing for streams, there is still insufficient consideration post‐processing, resulting limited accuracy To address this problem, an method based on directional disassembly...

Small target detection is a critical step in remotely infrared searching and guiding applications. However, previously proposed algorithms would exhibit performance deterioration the presence of complex background. It attributed to two main reasons. First, some common background interferences are difficult eliminate effectively by using conventional sparse measure. Second, most methods only exploit spatial information typically, but ignore structural priors across feature space. To address...

Against complex background containing the tiny target, high-performance infrared small target detection is always treated as a difficult task. Many low-rank recovery-based methods have shown great potential, but they may suffer from high false or missing alarm when encountering with intricate interferences. In this paper, novel graph-regularized Laplace approximation detecting model (GRLA) developed for dim scenes. Initially, non-convex regularizer instead of nuclear norm employed to boost...

The feature learning strategy of convolutional neural networks (CNNs) learns the deep spatial features from high-resolution (HR) synthetic aperture radar (SAR) images while ignoring speckle noise based on SAR imaging mechanism. In module, reduction by feature-adaptive projection guided a powerful embedded wavelet reconstruction mechanism can effectively learn statistics. this paper, we present Wavelet Driven Subspace Basis Learning Network (WDSBLN), following an encoder-decoder architecture,...

Traditional remote sensing scene classification methods based on low-level local or global features easily lead to information loss, additionally, the influence of spatial correlation images and redundancy feature representation are neglected. For overcoming these drawbacks, learnable multilayer energized locality constrained affine subspace coding (MELASC) – Convolutional Neural Network (CNN) framework (MELASC-CNN) which could generate orderless is proposed, it considers both diversity deep...

A method of pattern recognition tool wear based on Discrete Hidden Markov Models (DHMM) is proposed to monitor and predict failure. At the first FFT features are extracted from vibration signal cutting force in process, then vectors presorted coded into code book integer numbers by SOM, these books introduced DHMM for machine learning build up 3-HMMs different stage. And then, HMM recognised using maximum probability. Finally results failure prediction experiments were presented shown that effective.

UAV (unmanned aerial vehicle) captured images have small pedestrian targets and loss of key information after multiple down sampling, which are difficult to overcome by existing methods. We propose an improved YOLOv4 model for detection counting in images, named YOLO-CC. used the lightweight detection, replaces backbone with CSPDarknet-34, two feature layers fused FPN (Feature Pyramid Networks). expanded perception field using multiscale convolution based on high-level map generated...

When a spectral library is known, hyperspectral sparse unmixing could obtain the abundance images, which estimate fractional proportions in each pixel image scene. Dictionary pruning (DP) methods furtherly improve performance of by reducing dictionary to smaller subset. However, current algorithms usually only use single DP method subset, affects results. In this study, we propose multiple (MDP) algorithms, making them more accurate. MDP consists three methods, namely, angle mapping (SAM),...

Scene flow estimation is critical for real-world vision problems such as autonomous driving and augmented reality. Due to the popularity of 3D LiDAR sensors, scene from point clouds arouses increasing attention. Existing methods usually use a embedding-based layer find correspondences between pairs. However, only using not enough model global mutual relationship two features due local matching. In this paper, we introduce cross-transformer capture more reliable dependencies Moreover,...

Spiking cameras have shown promising advantages for optical flow estimation in high-speed scenarios. The recent work SCFlow [1] attempts to train an model using spike frames based on a multi-scale reconstruction loss. However, only the loss is unable effectively deal with details of motion, which may lead noise and blur estimated fields. To address this issue, we introduce contrastive into spike-based estimation, exploits both information positive samples negative samples. Moreover, propose...

Fixed coding style in bag of visual words (BOVW) model and strong spatial information convolutional neural network (CNN) feature representation make the vector less adaptable for scene classification. With purpose extracting learnable orderless SAR classification, high-order generalized pooling trained by backpropagation is proposed learning locally aggregated descriptors (VLADs) locality constrained affine subspace (LASC), compared with first-order style, could learn features outer product...

Feature learning of convolutional neural networks (CNNs) has gained considerable attention and achieved good performance on synthetic aperture radar (SAR) image scene classification. However, the existing feature methods is limited for generating distinguishable representations because such techniques inherently suffer from shortcomings, i.e., they do not consider local distribution deep orderless statistics multifeature style. To alleviate these drawbacks, we propose a compact global-local...

Active contour model driven by novel fitting term is proposed for image segmentation in this letter. The contains two parts: one L 1 term, which describes the change of energy between inside and outside curve; another reciprocal cross‐entropy can compute local every point on curve. more robust than when regional grey scale fluctuates greatly, detail direction curve be controlled better it. Experiments several synthetic real images have shown that achieves results efficiency other models;...

L0 gradient minimisation model, one of edge‐aware image smoothing method, also suffers from the stair‐casing effect and images with strong textures cannot be smoothed effectively weak edges or structures will overly. The authors propose a method to overcome these drawbacks above. To begin with, is subjected non‐subsampled shearlet transform obtain high‐frequency component, combine all component by maximum local energy rules decomposition image, afterwards, introducing data term associated...

The second-order pooling manner, exploring higher feature statistics than the first-order pooling, has achieved impressive performance in scene classification. However, object not only presents similarity but also exhibits diversified singularity on synthetic aperture radar (SAR) image. These make approaches to explore single-view less adaptable for SAR To solve this issue, an end-to-end training framework based multiview cross correlation attention network (MCAN) is proposed. spatial and...

The description of context information affected by speckle and class imbalance under labeled data makes the pixelwise classification for high-resolution (HR) synthetic aperture radar (SAR) image a challenging task. To address these issues, we propose global-context pyramidal class-balanced network (GPCNet) HR SAR classification. proposed structure follows an encoder–decoder architecture. In encoder module, multiscale convolutional global-local cross-channel attention (GCA) blocks are...

The transformer-based methods have demonstrated remarkable advancements in synthetic aperture radar (SAR) classification. Nevertheless, many of these ignore global statistical information and semantic feature interaction for effectively characterizing different SAR land covers under complex structure. Leveraging second-order statistics presents an efficacious approach to characterize features patches well. Motivated by this, we integrate pyramid pooling covariance techniques into each the...

Benefiting from the inherent capability of graph convolutional neural networks (GCNs) to flexibly convolve over regions with arbitrary shapes, they could model spatial topology SAR land covers. However, existing GCNs-based classification methods often work on a static and perform average pooling which have fixed weights as node style, thus lacking dynamic representation patches under complex structure. Moreover, distinguishable feature learning different GCN blocks is generally ignored,...

The vision transformer has been widely applied in remote sensing image scene classification due to its excellent ability capture global features. However, images involve challenges such as complexity and small inter-class differences. Directly utilizing the tokens of for feature learning may increase computational complexity. Therefore, constructing a distinguishable network which adaptively selects can effectively improve performance while considering Based on this, second-order...