A5057651379- Advanced Neural Network Applications
- Vehicle License Plate Recognition
- Handwritten Text Recognition Techniques
- Industrial Vision Systems and Defect Detection
- Autonomous Vehicle Technology and Safety
- Advanced Image and Video Retrieval Techniques
- Visual Attention and Saliency Detection
- Physical Unclonable Functions (PUFs) and Hardware Security
- CCD and CMOS Imaging Sensors
- Integrated Circuits and Semiconductor Failure Analysis
- Image Processing Techniques and Applications
- Image Processing and 3D Reconstruction
- Rock Mechanics and Modeling
- Geotechnical Engineering and Underground Structures
- Video Surveillance and Tracking Methods
- Drilling and Well Engineering
University of Science and Technology Beijing
2019-2024
PLA Army Engineering University
2019
Vehicle and license plate detection plays an important role in intelligent transportation systems is still a challenging task real applications, such as on-road scenarios. Recently, Convolutional Neural Network (CNN)-based detectors achieve the state-of-the-art performance. However, it difficult to efficiently detect vehicle simultaneously most cases. With single network, can affect of due inclusion relation. In this paper, we propose end-to-end deep neural network for detecting given image,...
Benefiting from attention mechanisms, query-based detectors have a strong model capacity. They predict classification and regression by utilizing their shared queries features in the decoder. Inter-task biases cause multi-directional gradients that disturb each other to limit optimization. In this work, we introduce an decoupling (AD) for explicitly align multi-task features. Specifically, AD consists of Dense-to-Sparse Query Generator (DSQG) Split Cross-Attention (SCA), enabling query...
Multi-scale object detection in natural scenes is still challenging. To enhance the multi-scale perception capability, some algorithms combine lower-level and higher-level information via feature fusion strategies. However, inherent spatial properties among instances relations between foreground background are ignored. In addition, human-defined "center-based" regression quality evaluation strategy, predicting a high-to-low score based on linear relationship with distance to center of...
Label assignment (LA) is one of the essential phases in object detection paradigm and aims to classify samples as foreground or background. Current LA strategies generally discriminate by explicit thresholds then calculate weighted losses based on their significances. However, existing methods mostly neglect consider importance comprehensively due uneven distribution objects limitations detector structures. In this paper, we propose a hierarchical equalization loss (HEL) reconsidering...
As the license plate is multiscale and multidirectional in natural scene image, its detection challenging many applications. In this work, a novel network that combines indirect direct branches proposed for wild. The branch performs small-sized vehicle with high precision coarse-to-fine scheme using vehicle–plate relationships. detects directly input reducing false negatives due to miss of vehicles’ detection. We propose universal refinement method by localizing four corners plate. Finally,...
License plate detection is the first and essential step of license recognition system still challenging in real applications, such as on-road scenarios. In particular, small-sized oblique plates, mainly caused by distant mobile camera, are difficult to detect. this work, we propose a novel applicable method for degraded via vehicle-plate relation mining, which localizes coarse-to-fine scheme. First, estimate local region around using relationships between vehicle plate, can greatly reduce...
In view of the non-stationary characteristics electromagnetic (EM) side channel signal, this paper uses EM probe and oscilloscope to collect signal chip, analyzes by wavelet transform, extracts coefficient as feature, support vector machine complete classification four chip types. Experiments show that method can identify model quickly, stably low-cost. The experiment achieved expected results. average correct rate reaches 91.625%.