A5103164915- Advanced SAR Imaging Techniques
- Traffic Prediction and Management Techniques
- Underwater Acoustics Research
- Network Security and Intrusion Detection
- Geophysical Methods and Applications
- Smart Agriculture and AI
- Data Mining and Machine Learning Applications
- Machine Learning and ELM
- Seed Germination and Physiology
- Technology and Security Systems
- Image and Signal Denoising Methods
- Wireless Sensor Networks and IoT
- Allelopathy and phytotoxic interactions
- Plant tissue culture and regeneration
- Advanced Image Fusion Techniques
- Infrastructure Maintenance and Monitoring
- Transportation Systems and Logistics
- Advanced Malware Detection Techniques
- Vehicle License Plate Recognition
- Fire Detection and Safety Systems
- Advanced Neural Network Applications
- Anomaly Detection Techniques and Applications
China Railway Group (China)
2023
Air Force Engineering University
2019-2021
Fruit image classification is the key technology for robotic picking which can tremendously save costs and effectively improve fruit producer's competitiveness in international market. In field, deep learning technologies especially DCNNs are state-of-the-art have achieved remarkable success. But requirements of high computation storage resources prohibit usages on resource-limited environments such as automatic harvesting robots. Therefore, we need to choose a lightweight neural network...
In the field of intrusion detection, there is often a problem data imbalance, and more unknown types attacks make detection difficult. To resolve above issues, this article proposes network model called CWGAN-CSSAE, which combines improved conditional Wasserstein Generative Adversarial Network (CWGAN) cost-sensitive stacked autoencoders (CSSAE). First all, CWGAN that introduces gradient penalty L2 regularization used to generate specified minority attack samples reduce class imbalance...
In the process of subway construction, there is a complex nonlinear relationship among engineering quantity, material cost, labor cost and total which cannot be accurately calculated by manpower alone. Therefore, reasonable deployment human, material, capital other resources in future requires high-precision highly reliable analysis prediction model. Aiming at this problem, paper proposes Bayesian network model construction project based on principal component K2 algorithm basis analyzing...
With the increase of environmental, human and other factors, traditional cost management method subway mechanical electrical engineering is unable to achieve accurate estimation, resulting in a large difference between estimation results actual results, affecting design implementation entire project. In order solve this problem, paper first develops data acquisition system collect real project cost. Secondly, linear discriminant based on principal component analysis proposed. Through impact...
Abstract In the radar automatic target recognition filed, extracting representative features from high resolution range profile (HRRP) is key issues, which determines accuracy and reliability of recognition. this paper, we propose a novel stacked discriminative auto-encoder(S-disAE), center loss integrated into auto-encoder, it can force learned feature with large distance close to their class representation center, so as reduce intra-class variations while keeping different classes...
Abstract Aiming at the problem of coherent speckle noise in SAR image processing, this paper improves network based on convolutional neural method, adjusts hierarchical relationship hierarchy, and introduces principles batch normalization residual to eliminate. The gradient disappears when level is too deep. loss function improved by using multiple convolution kernels different sizes which are used extract a variety levels feature information. Finally, peak signal-to-noise ratio (PSNR) edge...
Aiming at the problem of low detection accuracy and high false positive rate caused by noise doping in network data, same time improving speed, a intrusion based on stacked denoising sparse autoencoder extreme learning machine (sDSAE-ELM) is proposed. First, used to automatically extract robustness characteristics then for classification. Experiments NSL-KDD dataset show that method sDSAE-ELM has strong when processing high-dimensional noisy while shortening training time. And have been achieved.
Abstract Extracting key features from the radar high resolution range profile (HRRP) determines accuracy and reliability of target recognition. Aiming at problem feature extraction recognition in HRRP recognition, a one-dimensional stack convolutional autoencoder (1D-sCAE) method is proposed. Firstly, constructed to extract deep signals by unsupervised learning. Then, multiple autoencoders are stacked construct 1D-sCAE, classified fine-tuning network with label data. At same time, for...