- Geophysical Methods and Applications
- Advanced SAR Imaging Techniques
- Advanced Fiber Optic Sensors
- Emotion and Mood Recognition
- Face recognition and analysis
- High voltage insulation and dielectric phenomena
- Industrial Vision Systems and Defect Detection
- Robotic Path Planning Algorithms
- Machine Learning and Data Classification
- Fluid Dynamics Simulations and Interactions
- Advanced Vision and Imaging
- Polymer-Based Agricultural Enhancements
- Neural Networks and Applications
- Face and Expression Recognition
- Topological and Geometric Data Analysis
- Thermochemical Biomass Conversion Processes
- Image and Object Detection Techniques
- Biodiesel Production and Applications
- Explainable Artificial Intelligence (XAI)
- Robotics and Sensor-Based Localization
- Synthetic Aperture Radar (SAR) Applications and Techniques
- Ultrasonics and Acoustic Wave Propagation
- Control and Dynamics of Mobile Robots
- Cell Image Analysis Techniques
- Sparse and Compressive Sensing Techniques
Nanjing University of Science and Technology
2020-2023
Categorical variables often appear in datasets for classification and regression tasks, they need to be encoded into numerical values before training. Since many encoders have been developed can significantly impact performance, choosing the appropriate encoder a task becomes time-consuming yet important practical issue. This study broadly classifies machine learning models three categories: 1) ATI that implicitly perform affine transformations on inputs, such as multi-layer perceptron...
Computer vision methods have raised great interest in the manufacture of printed circuit boards (PCBs). The detection and localization circular holes on PCBs are very important for positioning or defect detection. In this paper, a new method such is presented. proposed first uses an adaptive edge detector to obtain image. Then it utilizes symmetry circle transforms PCB image into distance space where Euclidean relationship shown. rapid center search (RCCS) algorithm used locate each circle....
In order to analyze the propagation characteristics of ultrasonic signals generated by two-point partial discharge inside oil-immersed transformer, Finite Element Method is used for modeling and simulation. The temporal, spatial frequency at point transformer are obtained. Compared with single-point local case, overall sound pressure amplitude place where overlap occurs approximately doubled in case. distribution ranges observation points also expanded from 10-30kHz single 10-40 kHz...
Shadow in synthetic aperture radar (SAR) results dark areas or shadows on the SAR image, leading to challenges object classification tasks. Although Convolutional Neural Networks (CNN) have shown good performance classification, it cannot generalize well under shadow and features are corrupted due pooling. In this paper, we propose an explainable shadow-aware transformer with ASC for shaded classification. Our method integrates attention-based transformers, CutMix, a physical module enhance...
Due to obstructions in the natural environment, practical emotion recognition applications often experience reduced accuracy. In order address this technological challenge, paper proposes an approach embedding adaptive dual-channel attention, specifically designed for recognizing facial expressions obscured faces. This method employs attention mechanism focus on uncovered feature areas, enhancing algorithm's capability concentrate features. Additionally, it utilizes a dual-stream residual...
The objective of this study was to maximize the biofuel content water hyacinth obtained from microwave pyrolysis of. Box-Behnkenof response surface methodology applied for experimental design. Under optimal condition: 1110W power, 3.5m reaction time, 2% amount absorbent, maximum 20.17%.
Abstract It is difficult to obtain a large number of high‐quality labelled Synthetic Aperture Radar (SAR) image data. In order solve the classification task an SAR with limited data, this paper proposes dual‐network model pre‐training and fine‐tuning based on contrastive learning pseudo‐label training strategy. The algorithm firstly obtains initial weight network through unsupervised then takes advantage information realise fine tuning dual network. effectively reduces samples required...
This paper introduces a novel method for road target segmentation in the context of unmanned driving based on stereo disparity maps. The proposed utilizes topological persistence threshold analysis to address challenges associated with selecting appropriate values. approach involves converting 3D vehicle images into uv-disparity maps, extracting planes using v-disparity and calculating occupancy grid maps u-disparity Persistence diagrams are then constructed by generating results under...
In addressing the path planning challenges encountered by Automated Guided Vehicles (AGVs) within smart warehousing transportation and storage, as well inherent issues of traditional A* algorithm, we propose an optimized enhanced algorithm for optimization. Building upon conventional selects appropriate heuristic functions weighting coefficients, integrating Bezier curves to achieve final improved algorithm. The function is used optimize problem long paths. Combined with curve, tortuous that...