A5041633925- Video Surveillance and Tracking Methods
- Advanced Neural Network Applications
- Sensor Technology and Measurement Systems
- Visual Attention and Saliency Detection
- Industrial Vision Systems and Defect Detection
- Advanced Measurement and Metrology Techniques
- Neural dynamics and brain function
- Functional Brain Connectivity Studies
- Plant and Fungal Interactions Research
- Research in Cotton Cultivation
- Gaussian Processes and Bayesian Inference
- Neural Networks and Applications
- Advanced MRI Techniques and Applications
Harbin Engineering University
2023-2025
Zhengzhou University
2021
Northeast Electric Power University
2020
North China Electric Power University
2020
Chinese Academy of Agricultural Engineering
2012
The ‘You Only Look Once’ v3 (YOLOv3) method is among the most widely used deep learning-based object detection methods. It uses k-means cluster to estimate initial width and height of predicted bounding boxes. With this method, estimated are sensitive centers, processing large-scale datasets time-consuming. In order address these problems, a new for estimating boxes has been developed. Firstly, it randomly selects couple values as one center separate from ground truth Secondly, constructs...
The "You only look once v4"(YOLOv4) is one type of object detection methods in deep learning. YOLOv4-tiny proposed based on YOLOv4 to simple the network structure and reduce parameters, which makes it be suitable for developing mobile embedded devices. To improve real-time detection, a fast method YOLOv4-tiny. It firstly uses two ResBlock-D modules ResNet-D instead CSPBlock Yolov4-tiny, reduces computation complexity. Secondly, designs an auxiliary residual block extract more feature...
In view of the fact that traditional least squares calibration method doppler velocity log (DVL) lacks ability to deal with ill-posed problem measurement equation and random errors at both ends equation, this paper extends total combines truncated singular value decomposition propose a DVL method. The proposed can obtain more stable accurate parameters by dealing suppressing during calibration. Simulation outcomes indicate installation error angles for x-axis z-axis obtained through...