The absorption and scattering coefficients of biological tissue are related to the physiological state of the tissue, which are important parameters for the detection of human health. The prediction accuracy of absorption and scattering coefficients in the current optical property inversion method of the double-layer biological tissue model is greatly affected by parameters such as the thickness of the upper layer tissue, which limits the scope of application. In this study, an inversion method of absorption and scattering coefficients is proposed, which is insensitive to parameters such as the thickness of upper layer tissue, in which the spatial and temporal distribution information of diffuse reflectance is collected; then, convolution neural network algorithm is applied to predict the absorption and scattering coefficients of double-layer biological tissue. The inversion accuracy of absorption and scattering coefficients is high under the random parameters of thickness and refractive index of upper layer tissue. In the simulation experiment, using a modified Monte-Carlo simulation, the diffuse reflectance of the double-layer skin model at different space detection positions and different time is obtained, and the convolution neural network is trained and tested using the simulation data to predict the absorption and scattering coefficients of two-layer skin tissue. Results show that the mean relative errors of absorption and scattering coefficients are less than 4% when the upper layer tissue thickness and refractive index are constant, whereas when the upper layer tissue thickness and refractive index change randomly, the mean relative errors of absorption and scattering coefficients are still less than 8%. Compared with other methods, the measurement scheme and inversion algorithm proposed in this study improve the prediction accuracy and expand the practical application prospect and provide a new method for the noninvasive measurement of biological tissue optical properties.