I'm Ningkun Zhou, a Machine Learning Engineer with 5 year experience, specializing in computer vision and its applications in industrial and scientific settings. I'm looking for jobs in the United Kingdom right now.

I currently work at Danieli China, where I lead the algorithm team and together we develop the Automatic Scrap Classification System. My expertise lies in integrating computer vision algorithms into production environments. Here are a few highlights of my work:

This image acquisition system I developed is fully automatic. Whenever a new layer of scrap exposed by the electromagnet, a detailed image will be acquired. The zoom parameters are calculated in real time to ensure every detailed image is in the same scale, allowing a stable semantic segmentation result later.

• YOLO for Object Detection: Implemented YOLO to detect electron magnets in scrap yards for fully automatic PTZ camera image acquisition.
• Auto PTZ Camera Zoom Calculation: Developed an calibration algorithm to automatically adjust PTZ camera zoom factor using cross-correlation techniques.

• Fine-tuned Mask R-CNN: Fine-tuned Mask R-CNN for semantic segmentation for scrap metal.
• Large Dataset: Handled large quantity of annotation data, more than 25,000 images and 1,00,000 polygons per dataset
• Continuous Optimization: Worked closely with customers, ensured 90% to 95% accuracies by vehicle. More than five projects have been accepted under my leadership.

• Zero-shot Image Classification: Applied zero-shot image classification to reduce data annotation by 50%.
• Image Search by Feature: Developed a image search system based on contrastive loss to overcome annotation inaccuracy.
• Algorithm Deployment: To deploy our algorithms, I utilized and mastered multiple communication frameworks, including RESTful API, SQL, Redis, and RabbitMQ.

Prior to my industry work, I spent three years in research in the field of Biomedical Imaging, at Dr. Jun He's lab in GIBH Chinese Academy of Sciences, where I honed my computer vision skills and applied them to solve protein structure using Cryogenic Electron Microscopy. Here are my publications during my research years:

• Shuqi Dong, Huadong Li, Ningkun Zhou, et al. "Structural basis of nucleosome deacetylation and DNA linker tightening by Rpd3S histone deacetylase complex" Cell Research, 2023. DOI: 10.1038/s41422-023-00869-1

• Le Tang, Shuqi Dong, Ningkun Zhou, et al. "Vibrio parahaemolyticus prey targeting requires autoproteolysis-triggered dimerization of the type VI secretion system effector RhsP" Cell Reports, 2022. DOI: 10.1016/j.celrep.2022.111732

Other than sample preparation and data collection, I contributed to these publication by applying several computer vision algorithms:

• Unsupervised Classification: K-means clustering of protein particles applied in fourier space to filter out projections from different orientation.

Image adapted from supplementary figures of my publication

• Density Map Reconstruction: Central Slice Theorem applied to reconstruct a high-resolution 3D density map from 2D projection images. K-means clustering and Bayesian Polishing in 3D space was performed to further optimization.

Image adapted from supplementary figures of my publication

• Auto Sample Screening: Fine-tuned EffiecientNet integrated into data collection pipeline to determine the quality of data.

• Particle Segmentation: U-Net for protein particle segmentation to determine ROI, and further improve signal to noise ratio

I graduated from the University of Wisconsin–Madison, with a major in Genetics🧬 and a minor in Computer Science💻.

• Deep Learning: EfficientNet, U-Net, YOLO, Mask R-CNN, ViT, CLIP
• Programming Languages: Python, Bash, Node.js
• Packages: TensorFlow, PyTorch, OpenCV, Scipy, NumPy, pandas, Matplotlib
• Tools: Docker, Linux, Flask, RESTful API, RabbitMQ, Redis, SQL, Git

Feel free to explore my repositories. I'm open to work right now!

LinkedIn | [email protected] | resume_in_PDF