Deep learning-based synthetic aperture radar (SAR) image classification is an open problem when training samples are scarce. Transfer few-shot methods effective to deal with this by transferring knowledge from the electro–optical (EO) SAR domain. The performance of such relies on extra samples, as unlabeled novel class's or labeled similar classes samples. However, it unrealistic collect sufficient in some application scenarios, namely extreme case. In case, degrades seriously. Therefore,...

Introduction Cancer-related fatigue (CRF) is a common symptom induced by chemotherapy. The main objective of the present study was to investigate whether quercetin regulates hypothalamic-pituitary-adrenal (HPA) axis and chemoattractant protein-1 (MCP-1) signaling, two factors contributing CRF in mice exposed cisplatin. Methods Male BALB/c were randomly assigned following five groups for 15 weeks: Control, CDDP, CDDP+TAK779 (an antagonist MCP-1 receptor, human CC chemokine receptor R2...

Recently, few-shot learning (FSL) has received increasing attention because of difficulties in sample collection some application scenarios, such as maritime surveillance using synthetic aperture radar (SAR) or infrared images. In real situations it is a common requirement that the model can recognize novel classes incrementally, namely class-incremental (CIL). Considering above requirement, problem recognizes incrementally when both base and class samples are scarce proposed this article....

Both quercetin and leucine have been shown to exert moderately beneficial effects in preventing muscle atrophy induced by cancers or chemotherapy. However, the combined of leucine, as well possible underlying mechanisms against cisplatin (CDDP)-induced cancer-related fatigue (CRF) remain unclear. To investigate issues, male BALB/c mice were randomly assigned following groups for 9 weeks: Control, CDDP (3 mg/kg/week), CDDP+Q (quercetin 200 mg/kg/day administrated gavage), CDDP+LL (a diet...

Estimating keypoint scale and orientation is crucial to extracting invariant features under significant geometric changes. Recently, the estimators based on self-supervised learning have been designed adapt complex imaging conditions. Such learning-based generally predict a single scalar for or orientation, called hard estimators. However, are difficult handle local patches containing structures of different objects multiple edges. In this paper, Soft Self-Supervised Estimator (S3Esti)...

Fine-grained target recognition in Synthetic Aperture Radar(SAR) or infrared imaging modal is an open problem some application scenarios where training samples are scarce.Transferring common features from Visible Optical(VO) effective for the case that SAR (infrared) scarce. However, fine-grained recognition, transferring face two issues: 1) can be divided into and coarse-grained features. For task few-shot case, how to transfer needed considered; 2) modal, parts of carry much noise because...

Driving with fatigue is one of the main factors traffic accident, so that timely driver drowsiness detection very crucial. This paper proposes a novel method by constructing Hierarchical Hidden Markov Models(HHMMs) on eyes sequences. The includes two key parts: (1) take Faster-RCNN as backbone to train robust and efficient extractor imposing additional constraints. (2) build two-layer HHMMs fully exploit temporal context information different levels. We can calculate likelihood drowsy...

Vehicle detection is one of the hot topics in lane and vehicle counting. Many works have been done on it some data sets with satellite images aerial are proposed. However, ratio targets to background small results unsatisfied. In this paper, a bottom enhanced RetinaNet model named En-RetinaNet proposed get better performance detection. The EnRetinaNet includes an feature pyramid network(FPN) bottom-top fusion before region proposal network. Enhanced network adds layer exploit more local...

The seven basic facial expression classifications are a way to express complex human emotions and an important part of artificial intelligence research. Based on the traditional Bayesian neural network framework, ResNet18_BNN constructed in this paper has been improved following three aspects: (1) A new objective function is proposed, which composed KL loss uncertain parameters intersection specific parameters. Entropy composition. (2) Aiming at special function, training scheme for...

An immersed-boundary lattice Boltzmann method (IB-LBM) is developed on a graphical processing unit (GPU) to simulate fluid-particle interaction problems. This uses the solve incompressible flow field and immersed boundary handle interaction. A direct forcing introduced in IB capture particle motion. To demonstrate of efficiency capabilities numerical method, square pipe sedimentations one many spherical particles an enclosure are simulated. The results indicate that speedup over 60 times...

This paper presents an entirely unsupervised interest point training framework by jointly learning detector and descriptor, which takes image as input outputs a probability description for every point. The objective of the is formulated joint distribution properties extracted points. essential are selected sparsity, repeatability discriminability probabilities. To maximize efficiently, latent variable introduced to represent that satisfies required properties. Therefore, original...

With the development of technology, application Internet in daily life is increasing, making our connection with closer. However, improvement convenience, information security has become more and important. How to ensure a convenient living environment question worth discussing. For instance, widespread deployment IP-cameras made great progress terms convenience. On contrary, it increases risk privacy exposure. Poorly designed surveillance devices may be implanted suspicious software, which...