A5015157242- Video Surveillance and Tracking Methods
- Advanced Neural Network Applications
- Infrared Target Detection Methodologies
- Human Pose and Action Recognition
- Advanced Image and Video Retrieval Techniques
- Robotics and Sensor-Based Localization
- Remote Sensing and Land Use
- CCD and CMOS Imaging Sensors
- Speech and dialogue systems
- Advanced SAR Imaging Techniques
- Industrial Vision Systems and Defect Detection
- Advanced Optical Sensing Technologies
- Speech Recognition and Synthesis
- Remote-Sensing Image Classification
- Automated Road and Building Extraction
- Fire Detection and Safety Systems
Xidian University
2020-2025
China Mobile (China)
2024
In satellite videos, moving vehicles are extremely small-sized and densely clustered in vast scenes. Anchor-free detectors offer great potential by predicting the keypoints boundaries of objects directly. However, for dense vehicles, most anchor-free miss without considering density distribution. Furthermore, weak appearance features massive interference videos limit application detectors. To address these problems, a novel semantic-embedded adaptive network (SDANet) is proposed. SDANet,...
Deep learning methods have gradually developed into the mainstream of moving vehicle detection in satellite videos. However, these require labor-intensive and time-consuming box-level annotations to predict accurate locations sizes, which is challenging for large-scale video datasets with hundreds vehicles. To address this problem, a novel coarse-to-fine dynamic refinement framework (CFDRM) proposed videos only under supervision point-level annotations. CFDRM generates initial proposal boxes...
Ship detection in aerial images remains an active yet challenging task due to its arbitrary object orientation and various aspect ratios from the bird's-eye perspective. Most existing oriented objection methods rely on angular prediction or predefined anchor boxes, making these highly sensitive unstable regression excessive hyper-parameter setting. To address issues, we replace angular-based encoding with anchor-and-angle-free paradigm, propose a novel detector deploying center four...
Deep learning (DL) based object tracking methods have achieved encouraging results on natural videos. However, directly applying these DL-based to the vehicle of optical remote sensing videos (ORSV) still faces many challenges. Different with vehicles in nature videos, most ORSV are blurry, small size, and highly similar other vehicles. Furthermore, blurred easily blend into background difficult distinguish. To solve problems, an improved Siamrpn++ clustering-based frame differencing...
Deep learning method shows its powerful classification performance with sufficient available data. However, the labeled data is limited in hyperspectral images (HSIs). Semi-supervised algorithms have unique advantages on dealing this problem. Therefore, a semi-supervised convolutional neural network proposed paper. It consists of two branches, which use samples and large number unlabeled samples, respectively. The first branch includes an encoder-decoder model to extract contextual...
Deep learning (DL) based object detection methods have been making great achievements for natural images, which guides the vehicle of optical remote sensing videos (ORSV). Compared with objects in ORSV are smaller and blurrier, most vehicles crowded. Thus, it is difficult DL to detect these small only using single-frame image. To address this problem, a motion guided R-CNN (MG-RCNN) proposed. In MG-RCNN, information from consecutive frames extracted by mean differencing method merged into...
Ship detection in aerial images remains an active yet challenging task due to arbitrary object orientation and complex background from a bird's-eye perspective. Most of the existing methods rely on angular prediction or predefined anchor boxes, making these highly sensitive unstable regression excessive hyper-parameter setting. To address issues, we replace angular-based encoding with anchor-and-angle-free paradigm, propose novel detector deploying center four midpoints for each oriented...