Development of Deep Learning Color Recognition Model for Color Measurement Processes

We present a deep learning color recognition model for the color measurement process in the paint industry. Currently, spectrophotometers are primarily used for color measurements owing to their accuracy. The measurement method involves manually injecting the sample into a spectrophotometer. Our proposed method uses a webcam with a deep learning model on the stand of a spectrophotometer. Deep learning models are widely used for image and color detection. In this study, the “you only look once (YOLO)” algorithm is applied for real-time detection of color samples. Upon training various sample images using YOLO, the model could detect the sample area in real time using a webcam. An open source computer vision (OpenCV) library was used for the color recognition model, and the detected RGB color value was converted to the international commission on illumination color space (CIELAB) value, which is primarily used in the color measuring process. However, because of the mirror-like reflection of light from a surface with specular reflection, it is difficult to implement the color value using a camera. To address this problem, we compare several specular removal methods and propose the most suitable model for the color recognition model of color samples. The accuracy of the proposed model was verified by comparing the colors of various samples. Our proposed approach can easily detect samples and color values, which can contribute significantly to automatically calculating the exact amount of coloring required for the target color.