Object detection in autonomous driving scenarios has become a popular task recent years. Due to the high-speed movement of vehicles and complex changes surrounding environment, objects different scales need be detected, which places high demands on performance network model. Additionally, devices have varying capabilities, lightweight model is needed ensure stable operation with limited computing power. To address these challenges, we propose called BiGA-YOLO based YOLOv5. We design...

Object detection in aerial images is a challenging task because of the complex background and various orientations objects. Currently, many detectors have made significant progress improving mAP scores, but they do not achieve improvements efficiency model size. In this letter, we propose detector for resource-limited satellite network, namely, simple convolutional neural networks (simple-CNNs), which can be directly applied actual application scenarios using small sample data. Specifically,...

Object detection method of vessels based on deep learning technology can extract the vessel position and category information in remote sensing image. However, buildings artificial facilities port area will cause false alarms recognition results, which decrease accuracy. Therefore, containing object semantic segmentation tasks convolutional neural networks, this paper proposes a ship identification scene-mask R-CNN. Based network, location extraction, target classification scene...

Traffic sign recognition (TSR) is an indispensable component for vision-based system of self-driving car. Promising results have been achieved which especially benefit from the rapid development deep neural networks recently. However, there are few works focusing on algorithms' performances towards different complex conditions, such as weather and viewpoint variations. In this paper, we propose a new real-world TSR dataset, dataset with several fine-grained conditions fine labeled involving...

Racetrack memory, an emerging low-power magnetic promises a competitive replacement for traditional memory in the accelerators. However, random access racetrack is time and energy expenditure CNN accelerators because of its large amount invalid-shifts. In this article, we propose automatic-addressing architecture that builds novel data layout to guarantee next round can be always satisfied at in-situ or rigorously adjacent cells current round, producing fully serialized footprint drive...

Abstract Cellular image analysis system is a complex that plays critical role in disease diagnosis and pharmaceutical research. The of data one the most aspects system. However, there are differences distribution cellular images, including cell morphology, density etc. This often requires careful algorithm customization, strict parameter tuning, or even inefficient manual processing, leading to low levels automation. In this work, an efficient end‐to‐end segmentation algorithm, ECS‐Net,...

With the development of satellite technology, space information networks (SINs) have been applied to various fields. SINs can provide more and complex services receive tasks. The existing resource scheduling algorithms are difficult play an efficient role in such a environment resources We propose allocation scheme based on reinforcement learning. Firstly, according characteristics SINs, we established cloud computing architecture manage uniformly. Next, adopt variable granularity clustering...

<abstract> <p>Accurate classification and segmentation of polyps are two important tasks in the diagnosis treatment colorectal cancers. Existing models perform separately do not fully make use correlation between tasks. Furthermore, exhibit random regions varying shapes sizes, they often share similar boundaries backgrounds. However, existing fail to consider these factors thus robust because their inherent limitations. To address issues, we developed a multi-task network that...

Catering to the public nature of Ad hoc network in open channel and data communication being easily eavesdropped, this paper proposed an antieaves dropping algorithm which is based on coding. The RSA signature introducing timestamp homomorphic mechanism detect tampering replay attacks, as basis for calculating safety, used one measurement indicators router by node safety establish t pieces routing entry. It generates encoding vectors random number when source encoded will be divided into n...

The training of Deep Learning (DL) model can be accelerated significantly with Graphics Processing Unit (GPU). Accessing a server GPUs attached and directly on it is the most common way using GPUs, but this may cause unbalanced load low GPU utilization in multiple-server-multiple-application situation. This paper gives mathematical definition problem presents YARN-based solution. Key solution to build resource scheduling platform customized Application Master (AM) optimized for DL....

Aiming at the anti-eavesdrop security demandand potential safety hazard in Ad hoc, a secure randomlinear network coding algorithm was proposed. Thealgorithm has source and protecting codingvectors. The information methodand optimizing finite region is defined to guaranteethat properly protected. Different codingvectors schemes were designed for adaptingsecurity path without. We applied Hoc Networks tomeet demand on anti-eavesdrop. anti-eavesdropalgorithm based simulated analyzedin NS2...

Due to the limitation of acknowledgment, complexity software system and interference noises, traditional risk evaluation methods face challenge poor uncertain inputs. Therefore, it is crucial perform estimation on against a information database. This paper presents grey-based rough set approach by introducing grey theory into based analysis, aiming for an effective method evaluation.

The order of magnitude object detection instances in satellite images is larger than that conventional images, and many small are clustered together images. Most objects datasets perpendicular to the ground, while parallel ground their orientation varies greatly. Conventional with horizontal bounding box, a box may contain multiple dense lot background information, resulting disproportionate proportion box. oriented helps locate more accurately obtain azimuth information easily. Therefore,...

It is difficult to accurately forecast the arriving time of tidal bore on Qiantang River, and destroying huge. For above issues, this paper proposes tide line detection based image. With method, we can predict bore, then reduce casualty incidents. This method mainly includes extraction warning strategy, realizes River's accurately. According experimental results, get has good adaptability, extract line, meet requirements real-time reliability at same time, laid a foundation next step early...

The paper proposes a mathematical model about the relationship between curvature radius changes of interventricular septal configuration and pressure difference left right ventricular based on Young-Laplace model. method noninvasive detection is discussed. In this which can simulate deformation in differential designed produced. After scanning by CT, boundary extracted calculated. Software ANSYS used to septum verify experiment. Then ventricle was established Hooke's law combined with...

Summary The tasks of a space‐based information network are complex and diverse, but the resources environment minimal. existing methods challenging to match task demand resource supply accurately. Aiming at problem accurate prediction from resource, we propose adjustment strategy. First, multidimensional algorithm based on improved particle swarm optimization back propagation (IPSO‐BP) neural network. PSO is used optimize weight threshold BP make up for defects that easy fall into local...

The existing image matching methods for remote sensing scenes are usually based on local features. most common features like SIFT can be used to extract point However, this kind of may too many keypoints the background, resulting in low attention main object a single image, increasing resource consumption and limiting their performance. To address issue, we propose method that could implemented well resource-limited satellites images ship by leveraging line A keypoint extraction strategy...