Detecting small objects such as vehicles in satellite images is a difficult problem. Many features (such histogram of oriented gradient, local binary pattern, scale-invariant feature transform, etc.) have been used to improve the performance object detection, but mostly simple environments those on roads. Kembhavi et al. proposed that no satisfactory accuracy has achieved complex City San Francisco. Deep convolutional neural networks (DNNs) can learn rich from training data automatically and...

This paper presents a framework of discriminative least squares regression (LSR) for multiclass classification and feature selection. The core idea is to enlarge the distance between different classes under conceptual LSR. First, technique called ε-dragging introduced force targets moving along opposite directions such that distances can be enlarged. Then, ε-draggings are integrated into LSR model classification. Our learning framework, referred as LSR, has compact form, where there no need...

Hyperspectral unmixing is one of the crucial steps for many hyperspectral applications. The problem has proved to be a difficult task in unsupervised work settings where endmembers and abundances are both unknown. In addition, this becomes more challenging case that spectral bands degraded by noise. This paper presents robust model unmixing. Specifically, our developed with correntropy-based metric nonnegative constraints on imposed keep physical significance. Besides, sparsity prior...

In this paper, we design a new edge-aware structure, named segment graph, to represent the image and further develop novel double weighted average filter (SGF) based on graph. our SGF, use tree distance graph define internal weight function of filtering kernel, which enables smooth out high-contrast details textures while preserving major structures very well. While for external function, introduce user specified smoothing window balance effects from each node Moreover, also set threshold...

Open-vocabulary semantic segmentation (OVSS) is an open-world task that aims to assign each pixel within image a specific class defined by arbitrary text descriptions. Recent advancements in large-scale vision-language models have demonstrated their open-vocabulary understanding capabilities, significantly facilitating the development of OVSS. However, most existing methods suffer from either suboptimal performance or long latency. This study introduces ERR-Seg, novel framework effectively...

Pre-training techniques significantly enhance the performance of semantic segmentation tasks with limited training data. However, efficacy under a large domain gap between pre-training (e.g. RGB) and fine-tuning infrared) remains underexplored. In this study, we first benchmark infrared various methods reveal several phenomena distinct from RGB domain. Next, our layerwise analysis pre-trained attention maps uncovers that: (1) There are three typical patterns (local, hybrid, global); (2)...

This paper presents an interactive algorithm for segmentation of natural images. The task is formulated as a problem spline regression, in which the derived Sobolev space and has form combination linear Green's functions. Besides its nonlinear representation capability, one advantage this usage that, once it been constructed, no parameters need to be tuned data. We define on user specified foreground background pixels, solve (the coefficients functions) from group equations. To speed up...

It is known full-waveform inversion (FWI) generally ill-conditioned and various strategies including pre-conditioning regularizing the system have been proposed to obtain a reliable estimation of velocity model. Here, we propose new edge-guided strategy for FWI in frequency domain efficiently reliably estimate models with structures size similar seismic wavelength. The edges model at current iteration are first detected by Canny edge detection algorithm that widely used image processing....

ABSTRACT It is not hard to find that the forms of traditional education have limitations. The Virtual Laboratory provide a virtual experimental environment users via computers in controlling engineering by resorting reality technology. was famous for helping students study complex process experiments as well saving time and cost. Therefore, new century, it has become more popular means education. In this paper, we present geophysical laboratory system (VGLS) based on C#, Viustools, database...

Abrupt change detection is critical to monitor the occurrence of abnormal events from sensor data for situational awareness complex systems. However, various disturbances and noises applied observations may pose significant challenges robustness many abrupt methods. Recent researches have shown that bilateral filter can acquire outstanding performance on removing images while preserving edge information. In this letter, we propose two improved edge-preserving memory-based cumulative sum...

Cache-based approaches stand out as both effective and efficient for adapting vision-language models (VLMs). Nonetheless, the existing cache model overlooks three crucial aspects. 1) Pre-trained VLMs are mainly optimized image-text similarity, neglecting importance of image-image leading to a gap between pre-training adaptation. 2) The current is based on Nadaraya-Watson (N-W) estimator, which disregards intricate relationships among training samples while constructing weight function. 3)...

The advent of pre-trained vision-language foundation models has revolutionized the field zero/few-shot (i.e., low-shot) image recognition. key challenge to address under condition limited training data is how fine-tune in a parameter-efficient manner. Previously, numerous approaches tackling this have been proposed. Meantime, few survey papers are also published summarize these works. However, there still lacks unified computational framework integrate existing methods together, identify...

External memory-based neural networks, such as differentiable computers (DNCs), have recently gained importance and popularity to solve complex sequential learning tasks that pose challenges conventional networks. However, a trained DNC usually has low-memory utilization efficiency. This article introduces variation of architecture with convertible short-term long-term memory, named CSLM-DNC. Unlike the memory original DNC, new scheme memories offers different locations for read write, they...

Summary Seismic full waveform inversion (FWI) has been highly challenging due to the inherent nonlinearity, non-uniqueness and illposedness. Generally, various regularization methods are required obtain a stable solution. We have developed new adaptive edge-preserving scheme for FWI. The salient features of our algorithm include an efficient approach constrain sharp interfaces model discontinuities through total variation (TV) novel adaptively compute parameter. TV accomplishes these goals...

Current state-of-the-art action recognition approaches rely on optical flow to extract the local motion information and ignore importance of global description videos. In this paper, we present a novel architecture, named Multi-Glimpse Network (MGN), boost performance by combining Specifically, MGN makes predictions through two important modules, Local Glimpse Global Glimpse. extracts spatiotemporal features different periods using temporal sampling method. aggregates extracted develop These...