Zeshan Chen|To rest in the highest excellence

About me

🏢 Position:
Software Engineer at China Satellite Network Group (Hainan Branch)

🎓 Education:
Master's in Software Engineering from Xi'an Jiaotong University [985]
Bachelor's from Southeast University [985]

💡 Expertise:
Artificial Intelligence, Remote Sensing Image Processing, Satellite Communication Applications

🎯 Interests:
Music, Hiking, Photography, Traveling

View Resume

Projects

Agent for YatchManager

Agent for YachtManager is an intelligent assistant system designed for a long-distance yacht supervision platform. Equipped with natural language understanding and task execution capabilities, it enables regulatory personnel to perform operations—such as vessel trajectory queries, real-time positioning, alarm response, and intelligent early warnings—through natural language commands. By integrating large language models (e.g., DeepSeek) and advanced semantic parsing techniques, the system significantly lowers the operational barrier and improves overall efficiency.

See Live Source Code

Remote Sensing Object Segmentation

Satellite images are packed with tiny, low-contrast objects—boats, cars, aircraft—that the original Segment Anything Model (SAM) often misses. I solved this with SAM-LoRA: keeping SAM’s general power while specializing it for remote-sensing scenes with minimal extra parameters.

See Live Source Code

Multi-path Coding Transmission Based on Space–Air–Ground Integrated Network

Robust multi-path data transmission framework for integrated space-air-ground networks. Uses linear coding for order-independent packet recovery and MP-QUIC–based protocol with adaptive scheduling and congestion control to improve throughput and reliability.

See PDF

TS-SSINet:Hyperspectral Image Denoising via Spatial–Spectral Interactive Learning

TS-SSINet is a two-stage deep learning model designed for hyperspectral image (HSI) denoising, which effectively addresses the challenges of spatial redundancy modeling and spectral correlation utilization, and demonstrates strong performance in handling various types of HSI noise (e.g., Gaussian, stripe, dead lines) while preserving rich spatial and spectral details.

See Live Source Code