Recent progress on salient object detection is beneficial from Fully Convolutional Neural Network (FCN). The saliency cues contained in multi-level convolutional features are complementary for detecting objects. How to integrate becomes an open problem detection. In this paper, we propose a novel bi-directional message passing model At first, adopt Multi-scale Context-aware Feature Extraction Module (MCFEM) feature maps capture rich context information. Then structure designed pass messages...

Detecting salient objects in cluttered scenes is a big challenge. To address this problem, we argue that the model needs to learn discriminative semantic features for objects. end, propose leverage captioning as an auxiliary task boost object detection complex scenarios. Specifically, develop CapSal which consists of two sub-networks, Image Captioning Network (ICN) and Local-Global Perception (LGPN). ICN encodes embedding generated caption capture information major scene, while LGPN...

The ROI (region-of-interest) based pooling method performs operations on the cropped regions for various samples and has shown great success in object detection methods. It compresses model size while preserving localization accuracy, thus it is useful visual tracking field. Though being effective, ROI-based operation not yet considered correlation filter formula. In this paper, we propose a novel pooled (RPCF) algorithm robust tracking. Through mathematical derivations, show that can be...

Recently, anchor-free object detectors have shown promising performance in oriented detection on remote sensing images. However, the objects images always large variations arbitrary orientations, sizes, and aspect ratios, which makes existing methods hard to obtain satisfactory results. In this article, we propose a novel detector, center-boundary dual attention (CBDA) network (CBDA-Net), for fast accurate CBDA-Net, construct CBDA module, utilizes mechanism extract features center boundary...

Multi-object tracking in unmanned aerial vehicle (UAV) videos is an important vision task and can be applied a wide range of applications. However, conventional multi-object trackers do not work well on UAV due to the challenging factors irregular motion caused by moving camera view change 3D directions. In this paper, we propose UAVMOT network specially for views. The introduces ID feature update module enhance object's association. To better handle complex motions under views, develop...

Convolutional neural networks (CNNs) have been widely applied in the context of ship detection synthetic aperture radar (SAR) images, but performance is still not ideal scenarios with clutter interference. Mining frequency-domain information to suppress sea SAR has attracted wide attention. However, existing methods do process adaptively, which results degradation performance. To overcome this problem, article proposes a novel deep learning network called YOLO-FA. YOLO-FA contains proposed...

Deep learning-based methods have achieved great success in target detection tasks of computer vision, but when it comes to Synthetic Aperture Radar (SAR) image ship detection, some new challenges appear because the wide swath images, diverse appearances ships and lack detail information, which make inefficient less effective. Aiming these issues, this paper, a lightweight feature optimizing network (LFO-Net) based on popular single shot detector (SSD) model is proposed for polarization SAR...

We propose a new model for fast and accurate video object segmentation. It consists of two convolutional neural networks, Dynamic Targeting Network (DTN) Mask Refinement (MRN). DTN locates the by dynamically focusing on regions interest surrounding target object. The region is predicted via sub-streams, Box Propagation (BP) Re-identification (BR). BP stream faster but less effective at objects with large deformation or occlusion. BR performs better in difficult scenarios higher computation...

Existing defocus blur detection (DBD) methods usually explore multi-scale and multi-level features to improve performance. However, regions normally have incomplete semantic information, which will reduce DBD's performance if it can't be used properly. In this paper, we address the above problem by exploring deep ensemble networks, where boost diversity of detectors force network generate diverse results that some rely more on high-level information while ones low-level information. Then,...

Microwave I/Q down-converters are frequently used in image-reject super heterodyne receivers, zero intermediate frequency (zero-IF) and phase/frequency discriminators. However, due to the electronic bottleneck, conventional microwave mixers face a serious bandwidth limitation, imbalance, even-order distortion. In this paper, photonic fundamental sub-harmonic presented using polarization division multiplexing dual-parallel Mach-Zehnder modulator (PDM-DPMZM). Thanks all-optical manipulation,...

Existing defocus blur detection (DBD) methods generally perform well on a single type of unfocused scene (e.g., foreground focus), thereby suffering from the performance degradation for other types scenes. In this paper, we present first exploration full-scene DBD, and propose separate-and-combine framework to achieve excellent diverse We firstly structure DBD dataset (named as DeFBD+) through collecting more scenes background focus, full focus out focus) with pixel-level annotations. Then,...

Microwave I/Q up-converter is widely used in modern electronic systems to achieve single-sideband (SSB) up-conversion and vector modulation. However, due the frequency-dependent mixer quadrature coupler, conventional up-converters suffer limited operating bandwidth. In this paper, a photonic microwave proposed using polarization-division multiplexing modulator. The channels are built all-optical methods without any devices, so wide bandwidth obtained. Balanced detection also realized reduce...

Long-tailed distribution of remote sensing data generally limits the object recognition performance deep neural networks. We notice that too many samples from head class will induce network to learn features tail being biased towards head. To solve this, we propose a novel center-wise feature consistency learning (CFCL) mechanism for long-tailed recognition. Firstly, implement head-tail center generation procedure builds two teacher models extract knowledge and respectively, so as avoid...

Self-localization of sensor nodes is one the key issues in wireless networks. Based on analysis traditional range-free algorithms such as centroid and APIT (approximate perfect point triangulation test) schemes, effect random deployment all node localization researched. And then, an improved algorithm (ICLA) based quality perpendicular bisector proposed. In ICLA, are categorized into several kinds localized, respectively. Extensive simulation results indicate that ICLA obtains a better...

A wideband photonic microwave single sideband (SSB) up-converter and an in-phase/quadrature (I/Q) modulator are proposed experimentally demonstrated. The intermediate frequency (IF) local oscillator (LO) signals applied to the submodulators of a dual-parallel Mach-Zehnder (DPMZM) for up-conversion, phase generated radio signal can be arbitrarily tuned through working point DPMZM. Using polarization division multiplexing DPMZM, I/Q constructed SSB up-conversion or modulation. In experiment,...

To address multi‐sensor real‐time track‐to‐track association problem of aircraft platforms in a complex environment, where sensor biases are time‐varied, targets distributed closely and different sensors report targets, an anti‐bias algorithm based on distance detection is proposed according to the statistical characteristics Gaussian random vectors. First, vector between homologous tracks derived its feature analysed; second, rough minimum average refined χ 2 distribution illustrated...

Existing works mainly focus on crowd and ignore the confusion regions which contain extremely similar appearance to in background, while counting needs face these two sides at same time. To address this issue, we propose a novel end-to-end trainable region discriminating erasing network called CDENet. Specifically, CDENet is composed of modules mining module (CRM) guided (GEM). CRM consists basic density estimation (BDE) network, aware bridge network. The BDE first generates primary map,...

Recent state-of-the-art methods on focus region detection (FRD) rely deep convolutional networks trained with costly pixel-level annotations. In this study, we propose a FRD method that achieves competitive accuracies but only uses easily obtained bounding box Box-level tags provide important cues of regions lose the boundary delineation transition area. A recurrent constraint network (RCN) is introduced for challenge. our static training, RCN jointly fully (FCN) through box-level...