Fast Surface Topography Reconstruction Method for Profilometer Measurement based on Neural Continuous Representation

Surface profilometers are indispensable for the measurements on precise machining and optical manufacturing industries. However, to acquire accurate surface topography, existing equipment requires line-by-line sampling to cover the target surface, which extremely limits the measurement efficiency. To address this problem, this paper proposes a learning-based high-efficient measurement approach to accelerate the surface topography measurement, which adopts the sparse sampling to reduce the measurement time and learns continue surface representation to reconstruct dense-accurate surface topography. Specifically, a neural network model extracts the features from sparse sampling data and learns an implicit continuous function of the target surface; the dense accurate surface topography can be reconstructed from the continuous representation. The proposed approach greatly reduces sampling time consumption and preserves the accuracy of topography. To facilitate the training process, we synthesize a large-scale machined surface topography dataset to train the baseline model and capture a moderate real-world dataset to fine-tune the model performance in practical situations. Both the simulation and real workpiece experiments verified the performance of the proposed method.