In order to improve the accuracy of spectral confocal displacement measurement system, the influence of surface scattering characteristics of samples was studied. Firstly, the principle of spectral confocal displacement measurement system is introduced. Based on scalar scattering theory, the axial response of spectral confocal under the influence of surface scattering characteristics is deduced, and the relationship model of scattering influence on displacement measurement is established. Then, the displacement measurement error caused by the peak wavelength curve shift caused by scattering is studied theoretically and simulated. The results show that when the surface roughness of the sample is large, it will produce a large scattering effect, which will lead to a significant decline in measurement accuracy. At the same time, the spectral confocal shift measurement is affected by the incident characteristics of each wavelength. In order to correct the influence of scattering, a multivariate scattering correction method combined with generalized regression neural network (GRNN) is proposed to process spectral data, and a scattering correction algorithm model is established. Finally, an experimental platform was built, and samples were selected for displacement measurement experiments. The experimental results show that the measurement performance of the system decreases with the increase of roughness. For the sample with roughness of 20 nm, the maximum displacement measurement error is reduced from 12.6 μm to 1.9 μm, and the average displacement measurement error is reduced from 8.1 μm to 0.86 μm, which improves the displacement measurement accuracy and verifies the correctness of theoretical analysis and the effectiveness of the proposed scattering correction method.This research result has certain reference significance for improving the accuracy of spectral confocal displacement measurement system.Displacement measurement technology can be used to detect surface morphology, film thickness, three-dimensional structure of living cells in biomedicine, so it has a wide range of applications. Spectral confocal displacement measurement technology uses a broad spectrum light source, establishes the coding correspondence between each wavelength and its respective axial focusing position by using the principle of optical dispersion, decodes the information according to the obtained reflection spectral characteristics of the sample surface, and obtains accurate axial position or tiny displacement data, thus realizing precise measurement; This method has ultra-high distance measurement resolution of nanometer level, and has universal adaptability to environment and measured materials, and has obvious application advantages and development prospects in the field of precision manufacturing of nondestructive measurement. When the spectral confocal method is used for precise displacement measurement, the scattering characteristics of the sample surface will cause some scattered beams to enter the system, which will make the received effectively reflected spectral response signal have invalid scattering noise, and the measurement data will drift, causing measurement errors. Based on the scalar scattering theory, in the spectral confocal displacement measurement system, the shift of peak wavelength curve caused by scattering is studied by constructing the functional relationship between surface scattering and light intensity, and the displacement measurement error caused by scattering is analyzed and studied. In order to correct the influence of scattering, a multiple scattering correction method combined with General Regression Neural Network (GRNN) is proposed to establish a scattering correction model to process the spectral data. Finally, the effectiveness of the correction method is verified by experiments.The influence of surface scattering characteristics of samples on the measurement of spectral confocal shift is studied. When the surface of the sample is not smooth, it will lead to the scattering of reflected light on the surface (Fig.2). Analyze the relationship among root mean square roughness, incident angle and scattered light intensity (Fig.3, Fig.4). Compare the spectral data shifts under different root mean square roughness (Fig.5, Fig.6). The method of multivariate scattering correction combined with GRNN is used to process the spectral data and establish the scattering correction model. Experiments verify the effectiveness of the scattering compensation algorithm in spectral confocal.In the experiment, all kinds of measured samples can't be absolutely smooth. The accuracy of the spectral confocal displacement measurement system mainly depends on the reflection spectrum received by the system, and surface scattering is the main source that affects the reflection spectrum error. By analyzing the influence of surface scattering on the error of reflection spectrum, a scattering compensation algorithm is established to reduce the measurement error. The spectrum received by the system is mixed with scattered light. When the scattering situation is serious, the reflected light intensity will decrease Fig.3, Fig.4) and the peak wavelength will shift (Fig.5), which will lead to the decrease of measurement accuracy. The measurement error increases with the increase of root mean square roughness (Fig.6). The scattering correction algorithm established by multivariate scattering correction and GRNN generalized regression neural network can reduce the error caused by surface scattering and improve the accuracy of spectral confocal displacement measurement system.In order to improve the accuracy of spectral confocal displacement measurement system, the influence of surface scattering characteristics of rough samples is studied and analyzed. Firstly, the working principle of spectral confocal displacement measurement system is introduced. Based on scalar scattering theory, the functional relationship between surface scattering characteristics and spectral response of samples is constructed, and the peak wavelength drift and displacement measurement error caused by scattering are analyzed. The scattering characteristics of the sample surface will shift the reflection spectrum curve and affect the displacement measurement accuracy, especially when the root mean square roughness δ is large, the scattering influence is great and the system resolution is obviously reduced. Then, in order to correct the influence of scattering characteristics, a scattering error correction model is established by using multivariate scattering correction method combined with GRNN generalized regression neural network, and it is verified by experiments. The experimental results show that the maximum displacement measurement error is reduced from 12.6 μm to 1.9 μm, and the average displacement measurement error is reduced from 8.1 μm to 0.86 μm when measuring the roughness sample with δ of 20 nm. In addition, the measured data of sample blocks with roughness of 12 nm, 50 nm and 100 nm are compared and analyzed. The results show that the surface scattering increases with the increase of roughness, which reduces the performance of the system. After the scattering error correction, the measurement accuracy of the system is improved. The research results have guiding significance for further improving the performance of the spectral confocal displacement measurement system and promoting the engineering application of the system.