Project: Analysis of Mixture Composition through Peak Area and Position Calculation

Developed a 1D-CNN-based automated HPLC analysis system.

• Model Architecture: Developed a 1D-CNN using PyTorch with an optimal configuration of 256 segments and input dimension of 8192 via hyperparameter tuning.
• Signal Processing: Optimized the SNIP function for dynamic range compression and iterative peak suppression to mitigate baseline noise.
• Data Engineering: Preprocessed real HPLC data and constructed a synthetic dataset comprising millions of samples.
• Achievement: Achieved a Mean Relative Error (MRE) of under 3%, significantly outperforming traditional derivative methods and YOLO-based models.