Detecting object hue bin in the CIECAM02 uniform color space using a multi-layer perceptron for optimal spectrum generation

In this paper, we present an innovative enhancement to an existing multi-channel light-emitting diode luminaire by integrating artificial intelligence and image analysis capabilities. The system captures an image of the object to be illuminated using a custom-developed Android application. The dominant color in the image is first identified, and then a multi-layer perceptron neural network classifies it into one of 16 hue bins within a uniform color appearance model. This classification process enables the luminaire to select a predefined spectrum designed to enhance the visual appeal of the illuminated object. To ensure robust hue bin classification, we developed a rich dataset with 7920 samples, collected using eleven smartphones under ten distinct light sources. Using this diverse dataset, thirteen multi-layer perceptron neural networks with varying input features were trained and compared, achieving hue classification accuracies, precision, and recall all ranging from 95 % to 99 % on the test set. One of the trained multi-layer perceptron networks was integrated into the Android application, with an inference time of 90 μs on a mid-range smartphone, demonstrating the practical application of artificial intelligence for on-demand hue analysis and lighting optimization. This advancement not only enhances the luminaire's functionality but also enriches the user experience by providing intelligent, on-demand lighting solutions tailored to the object's dominant color.