Light Source Spectrum Estimation Using RGB and Multispectral Sensors

The patent segments the spectrum acquisition process into two independent parts: (1) using a multispectral sensor to capture the integral spectrum (light source × object reflectivity × response function), and (2) using a trained neural network model to separate and estimate the pure light source spectrum from this integral result. This segmentation allows the system to maintain simplicity while achieving high accuracy by decoupling the measurement function from the analysis function.

The patent introduces a neural network model as an intermediary component that processes the integral spectrum data from the multispectral sensor and transforms it into an accurate light source spectrum estimation. This intermediary model acts as a bridge between the raw sensor data and the desired pure light source spectrum, enabling accurate spectrum acquisition without requiring complex hardware modifications.

The patent replaces the conventional mechanical/mathematical interpolation processing method with an intelligent neural network-based approach. Instead of using traditional signal processing techniques to estimate the light source spectrum, the system employs a trained neural network model that has learned the complex relationships between integral spectra and pure light source spectra from training data, achieving superior accuracy by substituting mechanical processing with intelligent computation.

The patent changes the processing approach from direct mathematical interpolation to probabilistic estimation through a trained neural network. The model outputs a probability distribution over possible light source types and their corresponding spectra, allowing the system to select the most likely light source spectrum. This parameter change from deterministic interpolation to probabilistic estimation enables higher accuracy in separating the light source spectrum from the integral result.