Vehicle detection in aerial images, being an interesting but challenging problem, plays important role for a wide range of applications. Traditional methods are based on sliding-window search and handcrafted or shallow-learning-based features with heavy computational costs limited representation power. Recently, deep learning algorithms, especially region-based convolutional neural networks (R-CNNs), have achieved state-of-the-art performance computer vision. However, several challenges...

Recently, deep learning-based methods have brought new ideas for ship detection in synthetic aperture radar (SAR) images. However, several challenges still exist: 1) models contain millions of parameters, whereas the available annotated samples are not sufficient number training. Therefore, most detectors to fine-tune networks pre-trained on ImageNet, which incurs learning bias due huge domain mismatch between SAR images and ImageNet Furthermore, it has a little flexibility redesign network...

Trajectory planning is one of the key and challenging tasks in autonomous driving. This paper proposes a novel method that dynamically plans trajectories, with aim to achieve quick safe reaction changing driving environment optimal balance between vehicle performance comfort. With proposed method, such complex maneuvers can be decomposed into two sub-maneuvers, i.e., lane change keeping, or their combinations, trajectory generalized simplified, mainly based on maneuvers. A fold...

With the availability of an increasing amount images from Internet and several user-friendly crowd sourcing tools such as Amazon Mechanical Turk (AMT), many large-scale ground level image datasets with semantic annotations have been collected in vision community, they fostered efficient ways to describe content. For example, visual attributes shown promising potentials recognition retrieval. However, scarcity labeled samples earth observation (EO) community (collection through...

Ship detection and recognition are crucial components of SAR ocean monitoring applications. In the literature, various features have been proposed for ship pattern analysis. However, operators often face dilemma that they little knowledge on feature selection. this paper, we first propose a novel RCS density encoding description. A two-stage selection approach is then presented. Finally, experiment conducted with high resolution imagery reveals percent correct classification as 91.54%.

Recently, deep learning has made significant progress in synthetic aperture radar automatic target recognition (SAR ATR). However, convolutional neural networks (DCNNs) are discovered to be susceptible carefully crafted adversarial perturbations. Regarding the unique scattering mechanism SAR imaging, feature such as attributed centers (ASCs) should deeply considered attack (AA) algorithms for DCNNs ATR. In this letter, an AA algorithm named ASC-STA is proposed take advantage of powerful...

The rapid advancement of deep neural networks has significantly increased demands for computational complexity and data volume. This trend is especially evident with the emergence large language models, which have rendered traditional architectures such as CPUs GPGPUs insufficient in meeting performance energy efficiency requirements. Spatial accelerators present a promising solution by optimizing on-chip compute, storage, communication resources. In exploring spatial accelerator design...

The rapid development of deep neural networks (DNNs), such as convolutional and transformer-based large language models, has significantly advanced AI applications. However, these advances have introduced substantial computational data demands, presenting challenges for the systolic array accelerators, which excel in tensor operations. Systolic accelerators are typically developed using two approaches: scale-up, increases size a single array, scale-out, involves multiple parallel arrays...

NA

Faster Region based convolutional neural networks (FRCN) has shown great success in object detection recent years. However, its performance will degrade on densely packed objects real remote sensing applications. To address this problem, an enhanced deep CNN method is developed paper. Following the common pipeline of "CNN feature extraction + region proposal classification", our primarily latest Residual Networks (ResNets) and consists two sub-networks: network network. For detecting...

This paper proposes a novel method on dynamic trajectory planning for intelligent vehicle driving under traffic environment with uncertainties. The statistical characteristics of motion are first analyzed model, in which the inputs considered to be random variables certain probability distribution. Therefore output model can calculated via unscented transformation probabilistic spread. Then overall collision candidate trajectories is assessed confidence level. Finally employed achieve...

Synthetic aperture radar (SAR) can perform observations at all times and has been widely used in the military field. Deep neural network (DNN)-based SAR target recognition models have achieved great success recent years. Yet, adversarial robustness of these received far less academic attention remote sensing community. In this article, we first present a comprehensive evaluation framework for DNN-based recognition. Both data-oriented metrics model-oriented to fully assess performance under...

Energy efficient architecture is essential to improve both the performance and power consumption of a computer system. However, modern computers suffer from severe "memory wall" problem due significant gap between processor technology memory technology. Thus, community evolving compute-centric memory-centric designs reduce data movement overhead. This paper presents comprehensive survey main challenges recent advances in energy architecture. We summarize two research directions: improving...

Deep neural networks (DNNs) have been widely utilized in automatic visual navigation and recognition on modern unmanned aerial vehicles (UAVs), achieving state-of-the-art performances. However, DNN-based systems UAVs show serious vulnerability to adversarial camouflage patterns targets well-designed imperceptible perturbations real-time images, which poses a threat safety-related applications. Considering scenario UAV is suffering from attack, this paper, we investigate construct two...

As a safety-related application, visual systems based on deep neural networks (DNNs) in modern unmanned aerial vehicles (UAVs) show adversarial vulnerability when performing real-time inference. Recently, ensembles with various defensive strategies against samples have drawn much attention due to the increased diversity and reduced variance for their members. Aimed at recognition task of remote sensing images (RSIs), this paper proposes use reactive-proactive ensemble defense framework solve...

A novel shape descriptor, named as ratio histograms (R-histogram), is proposed to represent the relative attitude relationship between two independent shapes. For a pair of shapes, shapes are treated longitudinal segments parallel line connecting centroids and R-histogram composed length ratios collinear segments. theoretically affine invariant due distance invariance transformation. In addition, computation weakens noise contribution, robust noise. Based on R-histogram, shape-matching...

NA

This paper presents a simple, novel, yet very powerful approach for real-time infrared pedestrian detection based on random projection. In our framework, firstly, feature-centric efficient sliding window scheme is proposed candidate pedestrians searching. Different from the traditional threshold or edge region of interest (ROI) generation techniques, it performs robustly under different scenes without delicate parameter tuning. Secondly, at feature extraction stage, small set features...