Currently, many tone mapping operators (TMOs) have been provided to compress high dynamic range images low (LDR) for visualizing them on the common displays. Since quality degradation is inevitably induced by compression, how evaluate obtained LDR indeed a headache problem. Until now, only few full reference (FR) image assessment metrics proposed. However, they highly depend and neglect human visual system characteristics, hindering practical applications. In this paper, we propose an...

Remote sensing image captioning aims to describe the content of images using natural language. In contrast with images, scale, distribution, and number objects generally vary in remote making it hard capture global semantic information relationships between at different scales. this paper, order improve accuracy diversity captioning, a mask-guided Transformer network topic token is proposed. Multi-head attention introduced extract features objects. On basis, added into encoder, which...

Brushless direct current motor is widely used in industrial production because of its simple structure, wide speed range and low noise. To improve the operation efficiency brushless DC reduce application costs, optimization analyzed by introducing JAYA algorithm. This method determines optimal parameters a using theory electromagnetic structure parameter selection calculation. The population diversity algorithm improved through an empirical learning strategy, adaptive strategy introduced to...

The active millimeter-wave (AMMW) scanner has been widely used for inspecting human security in public places recent years owing to its ability detect all kinds of objects under the clothes and be harmless body. However, it is really challenging concealed automatically accurately due inherent imaging noise, unknown object kind, uncertain position. Recently, many existing methods, especially deep learning-based, have achieved good performances on detection. These methods work well detecting a...

Recently, there has been significant progress in automatic pain assessment based on facial expression analysis. However, the performance of remains unsatisfactory, due to a lack analysis local pain-related action units and emotional ambiguity. In particular, ambiguous expressions complicate estimation pain. It is argued that certain regions related should receive more attention while estimating intensities. Based this, we propose multi-task hybrid Conv-Transformer method for assessment,...

ABSTRACT In ensemble learning, individual accuracy and diversities are two key factors for improving the performance, most existing methods have been designed from view of boosting diversities. Whereas stronger inevitably result in decreasing accuracies partial individuals, it brings a great challenge performance. fact, we observe that contribution individuals should be crucial factor which can effectively alleviate problem balancing diversity accuracy, but is ignored studies. Based on this,...

SAR image registration is a crucial problem in processing since the results with high precision are conducive to improving quality of other problems, such as change detection images. Recently, for most DL-based methods, has been regarded binary classification matching and non-matching categories construct training model, where fixed scale generally set capture pair blocks corresponding key points generate set, whereas it known that different scales contain information, which affects...

The concealed object detection in millimeter-wave human body images is a challenging task due to the noise and dim-small objects. Exploiting spatial dependencies mine difference between vital for discrimination of However, most approaches ignore context around object. In this paper, framework based on structural proposed suppress interference refine localizable semantic features. consists two subnetworks, region-based multi-scale weakly supervised feature refinement local context-based...

Considering that deep learning achieves the prominent performance, it has been applied to synthetic aperture radar (SAR) image registration improve accuracy. In most methods, a model is constructed classify matched points and unmatched points, in which SAR regarded as supervised two-classification problem. However, difficult annotate massive manually practice, limits performance of networks. Besides, inevitable differences among images easily cause some training testing samples are...

Abstract Facial expression recognition (FER) is still challenging due to the small interclass discrepancy in facial data. In view of significance crucial regions for FER, many existing studies utilize prior information from some annotated points improve performance FER. However, it complicated and time-consuming manually annotate points, especially vast wild images. Based on this, a local non-local joint network proposed adaptively enhance feature learning FER this paper. method, two parts...

Accurate models of 4H-SiC implanted resistors over a wide temperature range are necessary for using materials in high-temperature integrated circuits. This study presents nonuniform Gaussian resistors. The resistance values the N <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$+$</tex-math> </inline-formula> resistor and p-well exhibit nonlinear behavior with temperature. Specifically, they decrease then...

Ensemble learning performs better than a single classifier in most tasks due to the diversity among multiple classifiers. However, enhancement of is at expense reducing accuracies individual classifiers general and, thus, how balance and crucial for improving ensemble performance. In this paper, we propose new method which exploits correlation between their corresponding weights by constructing joint optimization model achieve tradeoff accuracy. Specifically, proposed framework can be...

Riemannian optimization has been widely used to deal with the fixed low-rank matrix completion problem, and metric is a crucial factor of obtaining search direction in optimization. This paper proposes new via simultaneously considering geometry structure scaling information, which smoothly varying invariant along equivalence class. The proposed can make tradeoff between information effectively. Essentially, it be viewed as generalization some existing metrics. Based on Riemanian metric, we...

An equivalent range equation with a monostatic component and bistatic synthetic aperture radar (BiSAR) compensation was proposed. There were five parameters (MEPs) in the both components. With MEPs, BiSAR can be expressed form similar to SAR (MoSAR) expression. Then, by using principle of stationary phase (POSP), we analytically derive 2-D spectrum point target, this is implemented into Doppler algorithm (RDA) that processes translational invariant (TI) data. In our RDA, spatially variant...

Vision‐based vehicle behaviour analysis has drawn increasing research efforts as an interesting and challenging issue in recent years. Although a variety of approaches have been taken to characterise on‐road behaviour, there still lacks general model for interpreting the vehicles on road. In this Letter, authors propose new method that effectively predicts based structured deep forest modelling. Inspired by learning, structure information is extracted from detected vehicle, then...

The facial expression recognition is very important for human-computer interaction. Therefore, a large number of researchers are focusing on this topic research and have acquired many valuable achievements. However, there still exist problems that need to be solved practical applications, such as the impact identity appearance differences, posture change etc. In work, dual-branch residual disentangled adversarial learning network proposed learn more accurate features by disentangling...