Publications by authors named "Langyue Zhao"
Article Synopsis
- Drone aerial imaging is becoming crucial due to advancements in optical sensor technology, but efficient multi-object tracking remains a challenge, especially with traditional methods that separate identification from tracking.
- A new Transformer-based framework is proposed, which integrates object detection and tracking into one process using self-attention mechanisms, simplifying the tracking pipeline and improving performance significantly.
- The system employs innovative techniques like trajectory detection label matching and cross-frame self-attention to enhance tracking accuracy and stability, with experimental results validating its effectiveness on datasets like VisDrone and UAVDT.
Article Synopsis
- ASDD-Net is a deep learning algorithm designed for detecting defects in polycrystalline silicon solar cells using electroluminescence images, featuring advanced techniques like downsampling and feature fusion to improve detection accuracy.
- The model includes modules like Enhanced Cross-Stage Partial Network Fusion and Hybrid Attention CSP Net, which enhance the extraction of various defect features and help the model analyze defects of different sizes and shapes effectively.
- With a mean Average Precision (mAP) of 88.81% on the PVEL-AD dataset and a processing speed of 69 frames per second, ASDD-Net demonstrates superior detection performance compared to current state-of-the-art models and shows good generalization on other datasets.
