High-Precision Noncontact Angle Measurement via Transfer Learning-Enhanced Magnetic Field Distribution Analysis

High-precision angle measurement remains challenging in extreme environments, where traditional sensors often fail due to precision limits and stability issues. Here, we introduce a non-contact angle detection method that combines transfer learning with micromagnetic simulations, enabling accurate angle prediction from magnetic field distribution images. By pre-training on 9001 simulated magnetic domain configurations, our deep neural network achieves a mean absolute error (MAE) of 0.007°. Even with limited experimental data, fine-tuning results in a testing error of 0.066°, demonstrating robust generalization across datasets. The network autonomously extracts vertical and horizontal magnetic field components from images, establishing a quantitative link between magnetic patterns and angles. The non-contact design of the method ensures robust performance in extreme environments, such as nuclear radiation zones and mining operations, while cloud integration enables real-time remote monitoring. By significantly reducing reliance on labeled experimental data, the approach lowers computational costs and enhances practicality for deployment. Additionally, tests on varied magnet geometries demonstrate the framework's adaptability, achieving sub-degree accuracy with minimal retraining. This work advances angle measurement technology by offering a scalable, non-contact solution with applications in robotics, aerospace, and industrial automation. The fusion of transfer learning and micromagnetic analysis provides a new paradigm for designing sensors that balance precision, environmental resilience, and practical deployability.