It is well known that the motion of a target induces range migration, especially for high-resolution synthetic aperture radar (SAR) systems. Ground moving imaging necessitates correction unknown migration. To finely refocus target, one must accurately obtain parameters compensating trajectory. However, in practice, these usually cannot be precisely estimated. This paper proposes new approach ground targets without priori knowledge their parameters. In devised method, azimuth compression...

This paper describes an ambiguity resolving approach for slant-range velocity estimation which utilizes the wideband characteristic of transmitted signal (multiple wavelengths). Based on wavelength dual-wavelength radar data. Then, two effective approaches are introduced to focus moving target no matter Doppler exists or not. The is estimated by number azimuth cell displacements between focused images. Both imaging methods have different properties and advantages. A performance analysis...

Pruning filters is an effective method for accelerating deep neural networks (DNNs), but most existing approaches prune on a pre-trained network directly which limits in acceleration. Although each filter has its own effect DNNs, if two are same with other, we could one safely. In this paper, add extra cluster loss term the function can force to be similar online. After training, keep and others fine-tune pruned compensate loss. Particularly, clusters every layer defined firstly pruning DNNs...

Pruning is an effective method to address the limitation of deploying deep neural networks (DNNs) on embedded systems. Most existing methods prune weights a given pre-trained DNN followed by costly fine-tuning process. In this paper, we propose new and efficient pruning algorithm which can structures filters filter shapes effectively. This achieved defining filter-wise shape-wise scaling factors indicate those be weakened. Then train network from scratch multiply with corresponding factors....

Recent years have witnessed a great advance of deep learning in variety vision tasks. Many state-of-the-art neural networks suffer from large size and high complexity, which makes them difficult to deploy resource-limited platforms such as mobile devices. To this end, low-precision are widely studied that quantize weights or activations into the low-bit format. Although efficient, usually train encounter severe accuracy degradation. In paper, we propose new training strategy based on...

Recent years have witnessed the great advance of deep learning in a variety vision tasks. Many state-of-the-art neural networks suffer from large size and high complexity, which makes it difficult to deploy resource-limited platforms such as mobile devices. To this end, low-precision are widely studied quantize weights or activations into low-bit format. Though being efficient, usually hard train encounter severe accuracy degradation. In paper, we propose new training strategy through...

Tuning-free personalized image generation methods have achieved significant success in maintaining facial consistency, i.e., identities, even with multiple characters. However, the lack of holistic consistency scenes characters hampers these methods' ability to create a cohesive narrative. In this paper, we introduce StoryMaker, personalization solution that preserves not only but also clothing, hairstyles, and body thus facilitating creation story through series images. StoryMaker...

Space-time adaptive processing (STAP) is an effective tool for moving target detection. Conventional STAP methodologies process the angular and Doppler two dimensional data vector. In practical applications, adjacent range cells are statistically dependent due to filtering, since point spreading function of a not ideal delta function. this paper, novel approach incorporating (fast time) information in presented clutter rejection, which we term space-time-range (STRAP). This method takes...

The strong target contamination has a great influence on the performance of conventional clutter suppression method. In this paper, robust approach to and ground moving detection been proposed for multi-channel SAR system. Firstly, modulus normalized vector corresponding pixel data is constructed its covariance matrix formed. Secondly, suppressed using new adaptive weight. Lastly, localization are carried out. effectiveness robustness confirmed by analysis real measured processing.

Pruning filters is an effective method for accelerating deep neural networks (DNNs), but most existing approaches prune on a pre-trained network directly which limits in acceleration. Although each filter has its own effect DNNs, if two are the same with other, we could one safely. In this paper, add extra cluster loss term function can force to be similar online. After training, keep and others fine-tune pruned compensate loss. Particularly, clusters every layer defined firstly pruning DNNs...