Dr. Cheng Ju | 3D object detection | Research Excellence Award

Xi'an Innovation College of Yan'an University | China 

The research profile centers on advancing computer vision through robust, trustworthy, and high-efficiency intelligent systems, addressing both task-specific challenges and foundational methodological limitations. Core contributions focus on object detection, anomaly segmentation, feature representation learning, and adaptive model design, with strong emphasis on real-world deployment reliability. In oriented small object detection, the proposed FDA-DETR framework overcomes weak feature representation and excessive computational cost in transformer-based detectors by integrating multi-scale frequency domain enhancement, density-aware dynamic query generation, and multi-granularity attention fusion, achieving significant gains in accuracy, robustness, and efficiency for dense and complex visual scenes. In zero-shot anomaly segmentation, the explainable recursive ERSF-AS framework resolves prompt dependency, limited anomaly feature learning, and interpretability challenges through a collaborative CLIP-SAM architecture combined with semantic, spatial, and frequency priors. Its recursive reasoning paradigm strengthens cross-domain generalization and reliability in both industrial and medical environments. Collectively, these innovations contribute not only to practical solutions for complex vision tasks but also to broader theoretical advancements in feature learning, attention modeling, adaptive mechanisms, and explainable artificial intelligence. This integrated research direction establishes a scalable foundation for future exploration across diverse computer vision and intelligent perception domains, promoting trustworthy, interpretable, and deployable AI systems for real-world applications.