Color Regression and Sorting System of Solid Wood Floor

Abstract

Solid wood floors are very common in interior decoration, and their color is an important indicator of product quality, selected in order to achieve the overall aesthetic needed to ensure color consistency. In order to realize the sorting of solid wood floors based on color depth, so that the colors of solid wood floors could be freely graded, one image acquisition system was built to collect 108 solid wood floor images and a set of fast sorting methods for solid wood floor color depth was developed. Among these, 10 solid wood floor images were used as the test set and therefore not sorted, and 98 solid wood floor images were sorted by color depth. Among these, 80 original images were expanded 13 times to 1040, for use as a training set, and 18 were used as a validation set. The color characteristics of solid wood floors in RGB, HSV and Lab color space were extracted, and LightGBM was used to realize the color depth sorting of the solid wood floors. At the same time, two deep learning algorithms, the Vision Transformer as well as Densenet121, improved by means of an adaptive pooling layer, were used to realize the color depth sorting of solid wood floor images of different sizes. The final ranking results showed that the color ranking method using LightGBM to regress the color features exhibited the most harmonious final results.