SAM提取多维灰度作为输入的视觉测量误差补偿OA

To mitigate measurement errors induced by illumination variations in precision image measure-ment,an error compensation model is proposed based on multidimensional grayscale features extracted via the Segment Anything Model(SAM)and fitted using a Whale Optimization Algorithm-optimized Radial Basis Function(WOA-RBF)neural network.A mathematical model describing illumination-induced edge shift is established to characterize the nonlinear effects of light intensity and surface scattering proper-ties on measurement accuracy.Leveraging SAM's zero-shot segmentation capability,average grayscale values from heterogeneous material regions are automatically extracted as multidimensional feature inputs to represent complex image characteristics.The WOA is employed to optimize the parameters of the RBF neural network,enabling accurate compensation of edge shift errors.Comparative experiments on chromi-um-zirconium-copper fixture measurements,benchmarked against one-dimensional linear fitting,GA-LSSVM,and SVR methods,demonstrate that the proposed model achieves an RMSE of 2.07 μm,an MAE of 1.73 μm,and an R² of 0.99(with the Zernike moment sub-pixel algorithm as a representative case).Consistent accuracy and strong robustness are observed across various sub-pixel edge detection al-gorithms,indicating that the proposed approach provides an effective solution for illumination-induced er-rors in precision image measurement.