Qizhi Xu

Beijing Institute of Technology
2 previous affiliations
3Machine learning papers tracked by Lacuna
Is this you?

Research Overview

Qizhi Xu focuses on remote sensing image processing, object detection in satellite imagery, and image super-resolution. Early work addresses the challenges of identifying marine vessels in low-resolution optical and infrared data, while recent efforts explore real-world degradation models for super-resolution tasks. The research also touches upon behavioral sequence analysis and collaborative learning methodologies.

Research Trajectory

Xu’s early research primarily addresses the technical bottlenecks of object detection in challenging remote sensing environments, particularly focusing on marine vessels. In 2022, as first author, Xu developed LMO-YOLO: A Ship Detection Model for Low-Resolution Optical Satellite Imagery. This work tackled the high false detection rates common in convolutional neural network (CNN) models when applied to low-resolution optical data. The research identified that standard 8-bit rescaling schemes cause critical information loss, and that low-resolution ships present significantly fewer effective features while being highly sensitive to object-background contrast variations. By proposing the LMO-YOLO architecture, Xu provided a targeted solution for low-resolution marine object detection.

Building on this focus, Xu subsequently investigated ship detection in infrared remote sensing, a modality valued for its penetration capabilities and resistance to light interference. In 2023, Xu co-authored Progressive Task-Based Universal Network for Raw Infrared Remote Sensing Imagery Ship Detection. This research addressed the severe stripe interference inherent in raw infrared images. The work highlighted that conventional preprocessing and destriping techniques are often too time-consuming for real-world applications and can inadvertently weaken the local contrast between objects and their backgrounds, complicating the detection of dim and small targets.

Alongside these core contributions to remote sensing and computer vision, the corpus indicates participation in distinct methodological domains. This includes the research direction Modeling Sequential Dynamics with Markov Chains, which moves beyond static frequency counts to analyze the temporal dimensions of behavior, and the broader area of Modeling Cognitive Connections in Collaborative Learning.

Returning to remote sensing, Xu's more recent collaborative work shifts toward foundational image enhancement techniques that support downstream detection tasks. In 2025, Xu contributed as a middle author to Toward Real-World Remote Sensing Image Super-Resolution: A New Benchmark and an Efficient Model. This project recognized that existing super-resolution methods, typically trained on simulated paired datasets with uniform bicubic degradation, often fail to generalize to real-world scenarios. By participating in the construction of a novel real-world dataset and an efficient model, Xu helped address the limitations of simulated degradation in remote sensing image super-resolution.

Current Directions

The most recent papers in the evidence address the gap between simulated training environments and real-world deployment in remote sensing. Current work investigates how authentic, complex degradation in satellite imagery impacts the performance of super-resolution models. By moving away from simple, uniform degradation assumptions, this direction aims to create more robust image enhancement pipelines that directly improve the accuracy of downstream tasks, such as object detection and recognition in practical, noisy environments.

Profile updated

See something wrong? Report a mistake