In the field of aerial photography, UAVs can be used as an ideal carrier sensors. It is meaningful to plan its coverage path combined with sensor model. Previous works mainly decentralized realize UAV formation cooperative operation, but for photography scene a single task, algorithm not effective. this paper, centralized proposed, which divided into two sub-problems: planning and control. planning, due high repetition rate excessive number turns traditional PRM algorithm, constraint...

The space environment has become highly congested due to the increasing debris, seriously threatening safety of orbiting spacecraft. Space-based situational awareness, as a comprehensive capability threat knowledge, analysis, and decision-making, is significant importance ensure security maintain normal order. Various awareness systems have been designed launched. Data acquisition, target recognition, monitoring constituting key technologies make major contributions, various advanced...

The components detection of a failed satellite is an important work in space on-orbit service. However, the current methods for do not consider effects low illumination and small targets on accuracy at same time, most datasets used are manually designed. This article proposes method based image enhancement improved faster region-based convolutional neural network (R-CNN) illumination. First, dataset containing low-illumination scenarios established simulated real environment. Second,...

Object detection is one of the key tasks in an automatic driving system. Aiming to solve problem object detection, which cannot meet speed and accuracy at same time, a real-time algorithm (MobileYOLO) proposed based on YOLOv4. Firstly, feature extraction network replaced by introducing MobileNetv2 reduce number model parameters; then, part standard convolution depthwise separable PAnet head further parameters. Finally, improved lightweight channel attention modul-Efficient Channel Attention...

This paper develops an end-to-end circular-feature-based pose estimation using the time-of-flight (TOF) camera for capturing serviced satellite. First, docking ring is chosen as recognized object measurement. Second, a new ellipse detection method proposed to detect ring. Third, circular feature used estimate six-degrees-of-freedom with aid of reflective block. Combined range at each pixel location from TOF, duality can be eliminated, and roll angle recovered through geometric constraint....

Object detection is one of the key algorithms in automatic driving systems. Aiming at addressing problem false and missed both small occluded objects scenarios, an improved Faster-RCNN object algorithm proposed. First, deformable convolution a spatial attention mechanism are used to improve ResNet-50 backbone network enhance feature extraction objects; then, pyramid structure introduced reduce loss features fusion process. Three cascade detectors solve IOU (Intersection-Over-Union) threshold...

Contextual information and the dependencies between dimensions is vital in image semantic segmentation. In this paper, we propose a multiple-attention mechanism network (MANet) for segmentation very effective efficient way. Concretely, contributions are as follows: (1) novel dual-attention capturing feature spatial channel dimensions, where adjacent position attention captures pixels well; (2) new cross-dimensional interactive fusion module, which strengthens of fine location structure...

The pose determination between nanosatellites and the cooperative spacecraft is essential for swarm in-orbit services. Time-of–flight (ToF) sensors are one of most promising to achieve tasks. This paper presented an end-to-end assessment how these were used estimation. First, embedded system was designed based on ToF camera with lasers as a driven light source. Gray depth images collected detect match in real time, obtaining information. A threshold-based segmentation proposed find small set...

The detection and segmentation of key spacecraft components is a crucial prerequisite for the successful execution on-orbit capture tasks; however, existing datasets are plagued by several problems. These include lack unified dataset component segmentation, consideration motion states spacecraft, extreme illuminations. problems hinder development related research. In response to above problems, this article aims bridge gap releasing detecting segmenting components. contrast synthetic images,...

In the existing GANs, realness of an input sample is estimated by discriminator using a single scalar. Such concise measurement may convey insufficient information to guide generator, potentially leading mode collapse and gradient vanishing. Unlike we represent concept as distribution rather than So that corresponding estimates from multiple angles, providing more informative guidance generator. particular, introduce measure objectives mutation operations which generate multifarious...

Abstract Efficient and accurate semantic segmentation is crucial for autonomous driving scene parsing. Capturing detailed information efficiently through two‐branch networks has been widely utilised in real‐time segmentation. This study proposes a network named MRFNet based on strategy to solve the problem of accuracy speed urban scenes. Many do not comprehensively consider contextual from sub‐regions different directions at scales. To handle this problem, Multi‐directional Feature...

This article presents a new perspective from control theory to interpret and solve the instability mode collapse problems of generative adversarial networks (GANs). The dynamics GANs are parameterized in function space directed methods applied investigate GANs. First, linear is utilized analyze understand It proved that stability depends only on parameters. Second, proportional–integral–derivative (PID) controller designed improve its stability. can be controlled adaptively generate images...

Abstract Generative adversarial networks (GANs) are able to produce realistic images. However, GANs may suffer mode collapse in their output data distribution. Here, we theoretically and empirically justify generalizing the GAN framework multiple discriminators with one generator for improving generative performance. First, a comprehensive perspective is adopted understand why occurs. Second, an array of cooperative realness introduced into combat explore discriminator roles ranging from...

For most autonomous landing based on ground cooperation markers, the design of markers is too simple. These can provide position-related information when UAV in a high position, but low it easily cause to lose its target and result failure due small size marker narrowness camera's field view. To solve above problems, this article focuses precise UAV. An improved which be detected different heights designed, relative position coordinate system established for pose estimation, height-adaptive...

Abstract Enhancing network feature representation capabilities and reducing the loss of image details have become focus semantic segmentation task. This work proposes bilateral attention for segmentation. The authors embed two modules in encoder decoder structures . Specifically, high‐level features structure integrate all channel maps through dense relationships learned by correlation coefficient module. positively correlated channels promote each other, negatively suppress other. In...

Restricted by the cost of generating labels for training, semi-supervised methods have been applied to semantic segmentation tasks and achieved varying degrees success. Recently, learning method has taken pseudo supervision as core idea, especially self-training that generate labels. However, are noisy. In learning, training progresses, model needs focus on more classes bias towards newly learned classes. Moreover, due limitation amount labeled data, it is difficult "stabilize" knowledge....

Abstract Detecting small objects are difficult because of their poor‐quality appearance and size, such issues especially pronounced for aerial images great importance. To address the object detection (SOD) problem, a united architecture that tries to upsample into super‐resolved versions, achieving characteristics similar those large thus resulting in more discriminative is used. For this purpose, new end‐to‐end multi‐task generative adversarial network (GAN) proposed. In architecture,...

With the rapid development of aerospace technology, future on-orbit servicing and control tasks will face particularly challenging problem more unknown changeable operation targets diversified refined tasks. Accurate identification pinpointing non-cooperative satellite key components such as inspection tracking, flexible capture maintenance are important prerequisites safeguards for on- orbit Aiming at that recognition segmentation space target faced with various components, large structural...

This paper establishes a 2D grid map based on the semi-direct method, which can be used for navigation, sweeping robot and other applications. In order to implement method with help of ROS platform, monocular LCSD-SLAM after obtaining key frames, is constructed. Most current 3D conversion research uses octrees conversion, but we have proposed threshold-based method. After obtains frame, through calculation, converted into in real time. The A* path planning algorithm navigate map....

We propose a method of 3D reconstruction small-sized object based on Kinect V2 RGB-D camera and turntable, which eliminates the need costly feature extraction robust matching techniques for motion estimation. Identification detection QR code are used to calibrate system, this basis, point cloud coordinate conversion background removal realized. Our coarse registration algorithm uses fixed rotation angle turntable construct matrix between frames. Combined with ICP (Iterative Closest Point)...