In this study, we propose a novel RGB-T tracking framework by jointly modeling both appearance and motion cues. First, to obtain robust model, develop late fusion method infer the weight maps of RGB thermal (T) modalities. The weights are determined using offline-trained global local multimodal networks, then adopted linearly combine response T Second, when cue is unreliable, comprehensively take cues, i.e., target camera motions, into account make tracker robust. We further switcher switch...

In this paper, we propose a novel weighted local cosine similarity (WLCS) and apply it to visual tracking. First, present the measure similarities between target template candidates, provide some theoretical insights on it. Second, develop an objective function model discriminative ability of components, use quadratic programming method solve obtain weights. Finally, design effective efficient tracker based WLCS simple update manner within particle filter framework. Experimental results...

Unauthorized unmanned aerial vehicles (UAVs) pose threats to public safety and individual privacy. Traditional object-detection approaches often fall short during their application in anti-UAV technologies. To address this issue, we propose the YOLOv7-GS model, which is designed specifically for identification of small UAVs complex low-altitude environments. This research primarily aims improve model’s detection capabilities backgrounds. Enhancements were applied YOLOv7-tiny including...

In this letter, we propose a novel hyperspectral image (HSI) classification method based on the joint collaborative representation (JCR) and support vector machine (SVM) models with decision fusion. First, motivated by model, adopt JCR model to deal HSI develop an effective learn contextual basis vectors for model. Second, mid-features are first extracted coefficients obtained then used train multiclass SVM classifier. After that, exploit multiplicative fusion rule combine models. We conduct...

Dental Panoramic radiography (DPR) image provides a potentially inexpensive source to evaluate bone density change through visual clue analysis on trabecular structure. However, dense overlapping of structures in DPR and scarcity labeled samples make learning accurate mapping from patches osteoporosis condition challenging. In this paper, we propose deep Octuplet Siamese Network (OSN) learn fuse discriminative features for prediction using multiple DRP patches. By exploring common features,...

In this work, we develop an effective person search algorithm with natural language descriptions. The contributions of work mainly include two aspects. First, design a baseline framework including three basic components: deep CNN model to extract visual features, bi-directional LSTM encode descriptions and the triplet loss conduct cross-modal feature embedding. Second, propose novel mutually connected classification fully exploit identity-level information, which not only introduces...

As a fundamental task in computer vision, visual object tracking has received much attention recent years. Most studies focus on short-term which addresses shorter videos and always-visible targets. However, long-term is closer to practical applications with more complicated challenges. There exists longer duration such as minute-level or even hour-level the task, also needs handle frequent target disappearance reappearance. In this paper, we provide thorough review of tracking, summarizing...

This paper presents a novel visual tracking method based on linear representation. First, we present probability continuous outlier model (PCOM) to depict the outliers within representation model. In proposed model, element of noisy observation sample can be either represented by principle component analysis subspace with small Guassian noise or treated as an arbitrary value uniform prior, in which simple Markov random field is adopted exploit spatial consistency information among (or...

In this paper, we develop an effective classification framework to classify a hyperspectral image (HSI), which consists of two fundamental components: weighted generalized nearest neighbor (WGNN) and label refinement. First, propose novel WGNN method that extends the traditional NN by introducing domain knowledge HSI problem. The proposed effectively models spatial consistency among neighboring pixels using point-to-set distance local weight assignment. addition, refinement enhance in...

In this letter, we present a novel online tracking method based on sparse representation. contrast to existing "sparse representation"-based algorithms, work adopts the representation construct both object and state models. The tracked can be sparsely represented by series of templates, also candidate samples in current frame. Furthermore, propose unified objective function integrate models, cast problem as an optimization that solved iteration manner. Finally, compare proposed tracker with...

A panoramic radiography image provides not only details of teeth but also rich information about trabecular bone. Recent studies have addressed the correlation between bone structure and osteoporosis. In this paper, we collect a dataset containing 40 images from different subjects, construct new methodology based on two-stage classification framework that combines multiple regions interest (ROIs) for osteoporosis prescreening. first stage, support vector machines (SVMs) are adopted to...

Vehicle logo detection (VLD) is a special and significant topic in object for vehicle identification system applications. Nevertheless, the range of research analysis VLD are seriously narrow real complex scenes, although it's critical role small sizes. In this paper, we make further work toward recognition real-world situations. To begin with, propose new multi-class dataset, called VLD-45 (Vehicle Logo Dataset), which contains 45000 images 50359 objects from 45 categories respectively. Our...

This paper presents a novel tracking algorithm based on the convex hull representation model with sparse representation. The tracked object is assumed to be within and candidate in meanwhile. consists of principle component analysis (PCA) subspace, constructed by all samples sparsity constraint. Then we propose objective function for our model, design an iterative solve it effectively. Finally, present framework proposed simple online update scheme. Both qualitative quantitative evaluations...

This letter presents a novel hyperspectral image (HSI) classification method based on robust joint nearest subspace and contextual prototype learning. First, we present to solve the HSI problem by exploiting set-to-class distance with metric consider both spectral spatial characteristics effectively. Second, develop an objective function learn prototypes robustly iteration algorithm it. Based learned prototypes, performance can be further improved. Finally, conduct numerous experiments...

Significant advancements have been witnessed in visual tracking applications leveraging ViT recent years, mainly due to the formidable modeling capabilities of Vision Transformer (ViT). However, strong performance such trackers heavily relies on models pretrained for long periods, limiting more flexible model designs tasks. To address this issue, we propose an efficient unsupervised pretraining method task based masked autoencoders, called TrackMAE. During pretraining, employ two...

A low-noise Low Dropout Regulator (LDO) is designed in a Semiconductor Manufacturing International Corporation (SMIC) <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$0.18~\mu $ </tex-math></inline-formula> m process, aiming at supply of clean and constant clamping voltage the operation Correlated Double Sampling (CDS) for elimination switching reset noise CMOS Pixel Sensors (CPS) applied to high energy...