✦ SEnet Project • Predicted the 4 states of soil dissolution in water, saving time in the long-term monitoring process forJapan's largest geological consulting company.• Optimized a CNN model through feature selection with SEnet, resulting in a 1.2x improvement in model accuracy, not only meeting but exceeding client requirements and expectations.✦ Metric Learning Project - Prototypical Netwrok• Clustered prediction for category and pose of rubber strip products, enabling the machine to perform automated processing for different product types.• Overcame data scarcity in various product categories by implementing Prototypical Network for Few Shot Learning.• Achieved a 99% clustering accuracy for both product types and poses.✦ Metric Learning Project - Triplet Selection• Clustered prediction for type and pose of rubber strip products, enabling the machine to perform automated processing for different product types.• Implemented Triplet Selection to cluster wood texture for both known and unknown categories in the model.• Achieved 100% clustering accuracy for known wood texture and 95% for unknown categories.✦ Spatial Attention UNET Project• Developed a model to solve the pain point of traditional Template Matching that fail with rotation and translation.• Utilized Spatial Attention to optimize the UNET model, enhancing the model's focus on detecting product defects.• Landed this technology for weld seam detection on steel pipes for a Tier 1 supplier of a renowned car brand in Japan, achieving an accuracy rate of 99%.

• Successfully conducted analysis of 20 national defense projects and strictly checked their structural safety.• Developed programs to efficiently process the aero load data on missle, reducing the workload of several days to 15-20 minutes.• Wrote technical documents detailing analysis methods, and visualizing analysis results.

Participated in two critical data science projects(industry-academia cooperation projects) on abnormal noise detection and machine health assessment, responsible for testing and modifying the programs provided by the university, and establishing a user guide for the software.