A method and system for predicting optical properties of a sample using DRS (Diffuse Reflectance Spectroscopy) are presented. Embodiments include obtaining a plurality of diffuse reflectance values based on diffusely reflected optical energy obtained from a detector that increases source detector separations, wherein the optical energy is irradiated to the sample and diffusely reflected within the sample. can An embodiment includes providing diffuse reflectance values to a hybrid deep neural network architecture (HDNNA), wherein the HDNNA includes a Fully Connected sub-network and a Convolutional sub-network, a merged neural network and It contains an output neural network. HDNNA can predict the optical properties of a sample by using the diffuse reflectance of the sample. It involves providing diffuse reflectance values to fully connected subnetworks and convolutional subnetworks. In this case, the convolutional subnetwork may be a 1D-CNN (One Dimensional Convolutional Neural Network). A fully connected subnetwork can interpret diffuse reflectance values as feature vectors. 1D-CNN can interpret diffuse reflectance values as tensors. A fully connected subnetwork consists of a plurality of neural network layers, and nonlinear mapping may be performed on each layer to generate an output value from an input value. 1D-CNN captures shape properties of diffuse reflectance values, where the relationship between different diffuse reflectance values can be derived using shape properties. The shape properties can be mapped to the optical properties of the sample. Intermediate features generated as outputs by fully connected and convolutional subnetworks can be merged using merge neural networks, which can then be non-linearly mapped to generate intermediate features. Intermediate features generated as outputs by the merge neural network can be input to the output neural network. The output value of the output neural network may be regarded as a predicted value for the optical properties of the sample. It may include training a hybrid deep neural network architecture (HDNNA). During the training phase, a reference value of the diffuse reflectance may be determined by specifying a range of reference values of the optical property. The reference value of the diffuse reflectance may be determined according to a specified reference value of the optical characteristic. Embodiments may include predicting optical properties of a sample during a training phase of a hybrid deep neural network architecture based on a reference diffuse reflectance value. The predicted value of the optical characteristic may be compared with a reference value of the optical characteristic. An embodiment may include minimizing a difference (error) between a predicted value of the optical characteristic and a reference value of the optical characteristic by using back propagation. and providing a mean square weighted error cost function to minimize an error between a predicted value of the optical property and a reference value of the optical property. Minimizing the cost function can improve the accuracy of hybrid deep neural network architectures to predict optical property values. The cost function may include a weighting factor assigned to the optical characteristic based on a reference value range of the optical characteristic. In this case, the range can mean the difference between the maximum reference value and the minimum reference value of an optical property, and the weighting factor allows equal weights for error backpropagation of different optical properties, so the reference of the optical property through optimal update of the parameter It is possible to accurately predict the optical properties of a sample regardless of the range of values. Once the hybrid deep neural network architecture is trained, it can be used to predict the optical properties of the sample after determining the diffuse reflectance values using source detector separation. These and other aspects of the embodiments herein will be better understood and understood when considered in conjunction with the following description and accompanying drawings. It is to be understood, however, that the following description is presented by way of illustration and not limitation, while showing examples and numerous specific details thereof. Many changes and modifications can be made within the scope of the embodiments of the present specification without departing from the spirit of the present invention, and the embodiments of the present specification include all such modifications.
展开▼
