Abstract:To address the issues of severe error accumulation and low reconstruction accuracy in Fiber Bragg Grating (FBG) shape sensing technology caused by sparse measurement points and the Frenet-Serret integration method, this paper proposes an end-to-end shape reconstruction method based on cubic spline interpolation and Bidirectional Long Short-Term Memory (BiLSTM). This method employs cubic spline interpolation to convert sparse discrete strain signals into dense spatial strain sequences, in order to satisfy the physical continuity constraints of optical fiber deformation; then, a BiLSTM model is developed to construct the strain transfer characteristics along the fiber axis and the spatial geometric constraints; establishing a direct mapping from strain space to three-dimensional coordinates. Simulation and experimental results show that this method abandons the point-by-point calculation process of traditional geometric reconstruction, effectively eliminating cumulative errors. The average relative errors of the simulation results were 0.28% and 0.23%, and the endpoint errors were 0.2719 mm and 0.4639 mm. The average relative errors of the experimental results were 1.22% and 1.03%, and the endpoint errors were 7.10 mm and 5.47 mm, respectively.