Moving object detection (MOD) in satellite videos remains challenging due to the extremely small size of interested targets and highly complex background. Both intra-frame (static) inter-frame (dynamic) information are great importance MOD. In this letter, we propose a two-stream network named dynamic static fusion (DSFNet) tackle MOD problem videos. Specifically, DSFNet is composed 2-D backbone extract context from single frame lightweight 3-D motion cues consecutive frames. Then extracted...

Infrared small target detection is a challenging task for deep learning-based methods because targets tend to disappear in the layers. To handle this problem, existing neural networks usually apply various dense and skip connections feature maintenance. Although these well-designed have achieved good performance, complex network structures reduce their efficiency. In paper, we propose simple yet efficient (RepISD-Net) infrared detection. The core of our RepISD-Net use different architectures...

Bottom acoustic parameters play an important role in sound field prediction. Acoustic deep water are not well understood. sensitive to the transmission-loss (TL) data shadow zone of water. We propose a multiple-step TL inversion method invert speed, density and attenuation Based on uniform liquid half-space bottom model, speed is inverted by using long range at low frequency obtained propagation experiment conducted South China Sea (SCS) summer 2014. Meanwhile, estimated combining with...

The deeply buried shoal dolomite reservoirs of the Lower Triassic Feixianguan Formation, giant Puguang Gas Field, NE Sichuan Basin, are exceptionally porous. influences diageneses on pore evolution studied. Through petrologic investigation, diagenetic phases divided into four stages, i.e., near-surface, pre-oil window, oil and gas window. Adobe Photoshop system is used to quantify rock texture components, porosity constitutions reservoir porosity. Porosity curves quantitatively recovered....

The properties of bottom acoustic parameters are greatly influenced on sound field prediction. Acoustic in deep water not well understood. We will discuss experience with geoacoustic inversion transmission-loss (TL) data from 100 to 600 Hz regions. Bottom sensitive the TL shadow zone water. Multiple steps method is proposed invertsound speed, density and attenuation Both a uniform liquid half-space two-layer profile assumed. Sound speeds inverted by using zones at low frequency obtained an...

The research on the vertical correlation characteristics of sound field in deep water has important implications for enhancing array gain and improving ability to detect underwater target. Based a deep-water experiment conducted South China Sea, coherence fields direct zone, shadow zone convergence are analyzed with signals received by line that covers maximal depth 1,866 m. numerical analysis based ray theory is carried out provide corresponding theoretical explanations variations at...

Multi-object tracking (MOT) in satellite videos is a new and challenging task. The difficulties stem from the extremely small objects low contrast between background. To tackle challenges of MOT videos, multi-object method proposed this paper to incorporate local motion prior into network. Specifically, we design cost volume construction module obtain offsets adjacent frames. Based on offsets, features previous frames can be propagated current frame spatio-temporal information. We conduct...

The physical ocean phenomena in the affect changes of water body, which has an important influence on sound propagation. A hyperbolic frequency modulation (HFM) signal propagation experiment towed source East Indian Ocean (EIO) was conducted summer 2019. center is 300 Hz, and hydrophone receives data from 4130 m far. This first time that we have underwater acoustic survey Ocean. phenomenon—Indian Dipole (IOD) observed. experimental IOD are processed analyzed, effects depth fluctuation...

In recent years, with the development of earth observation technology, satellite video can use optical sensors to obtain continuous images from mobile platforms, providing new data for detection and tracking large-scale moving targets. At present, target algorithms have been widely used in ground surveillance video. However, there are many challenges applying existing directly due low resolution video, non-appearance texture feature small target, signal-to-noise ratio, nonstationary camera...

A new method of echo signal detection and tracking is proposed for greatly reducing the computations miss-distance measurement. The main idea that miss- distance measurement in its nature only needs data containing signals, so raw are divided into frames pass algorithm will be further processed. Firstly, an initial scope frequency-point defined. And adaptive threshold scheme implemented to confirm if frame contains signals. Then, according result, performed renew next frame. carried out by...

The wind speed and rainfall rate above the sea surface have significant effects on ocean ambient noise, noise level can be predicted by using model combined with parameters. In this work, we study characteristics in East Indian Ocean analyze influences of noise. results show that correlation coefficient between high-band could reach 0.59 absence rainfall, 0.85 presence strong increase more than 6 dB. wind-generated source formula is also modified to make it applicable deep-sea environment...

Channel pruning, which removes unimportant channels by elaborately designing the importance criterion has become one of most successful model compression approaches. Different from static dynamic channel pruning learns and skips "unimportant" ones dynamically according to different inputs, can keep performance while reducing computational cost greatly. However, existing methods mainly focus on treats all samples equally, reduces capacity due permanent removal channels. In this paper, we...

Multi-object tracking in satellite videos has been widely used civilian and military fields. Among them, the of vehicles important applications field traffic monitoring. However, still remains challenging unsolved due to extremely small size lack appearance geometric features. In this paper, we propose an improved SORT tackle by introducing C3D CenterNet improve detection performance promote overall performance. Specifically, use as backbone extract spatio-temporal information a 3D channel...

Dynamic channel pruning is widely used in model compression to improve the efficiency of neural networks. Although dynamic can remove redundant channels dynamically, parameters still remain unchanged, which limit performance as response each neuron changes with different inputs. In this paper, we propose a novel method adaptive weight learning, adaptively adjust both and widths filters simultaneously. Specifically, design an adaptive-weight convolution module, be customized for inputs under...