A5053646874- Generative Adversarial Networks and Image Synthesis
- Advanced Image Processing Techniques
- Infrared Target Detection Methodologies
- Advanced Neural Network Applications
- Digital Media Forensic Detection
- Advanced Measurement and Detection Methods
- Video Surveillance and Tracking Methods
- Advanced Algorithms and Applications
- Industrial Vision Systems and Defect Detection
- Advanced Computing and Algorithms
- Image Processing and 3D Reconstruction
- Face recognition and analysis
Space Engineering University
2021-2023
Aiming to solve the problems of false detection, missed and insufficient detection ability infrared vehicle images, an target algorithm based on improved YOLOv5 is proposed. The article analyzes image characteristics then discusses in detail. uses DenseBlock module increase shallow feature extraction. Ghost convolution layer used replace ordinary layer, which increases redundant graph linear calculation, improves network extraction ability, amount information from original image. accuracy...
CycleGAN is widely used in various image translations, such as thermal-infrared–visible-image translation, near-infrared–visible-image and shortwave-infrared–visible-image translation. However, most translations are for infrared-to-visible the wide application of laser imaging has an increasingly strong demand laser–visible In addition, current translation mainly aimed at frontal face images, which cannot be effectively utilized to translate faces a certain angle. this paper, we construct...
The range-gated laser imaging instrument can capture face images in a dark environment, which provides new idea for long-distance recognition at night. However, the image has low contrast, SNR and no color information, affects observation recognition. Therefore, it becomes important to convert into visible then identify them. For translation, we propose laser-visible translation model combined with spectral normalization (SN-CycleGAN). We add layers discriminator solve problem of quality...
In this paper, the classic template matching sequential similarity detection algorithm(SSDA) has a large amount of calculation and slow operation time, an improved SSDA is proposed. By introducing integral image, point-by-point summation in window transformed into addition subtraction four positions which reduces for accumulation summation. The laser image lena are used as experimental images to be matched, with gray scale changes noise added matching. Compared traditional normalized...
Generative adversarial network (GAN) has become a hot research topic in the field of image processing. As an unsupervised training model, GAN been widely used computer vision, especially style transfer. The purpose is to make generator generate false image, and discriminator cannot tell whether input real or generated image. Compared with traditional models, model these advantages transfer: composed two different networks, loss function automatically learned by playing games each other....
One of the important research directions in field target detection computer vision, among which deep learning-based can extract advanced features and has higher accuracy than traditional algorithms. The inference speed convolutional neural networks embedded platforms is low, practical application value low. Therefore, for background optoelectronic device camera detection, on platform NVIDIA Jeston Nano, ResNet18 network used to identify photoelectric target. Use TensorRT accelerate process...