Road Surface Condition Recognition Using RGB Images: A Dataset and Approach

The present study utilized a self-built dataset comprising RGB images of two different types of road surfaces, with the aim of identifying whether the roads contain construction waste. The dataset currently consists of 6141 images, with 3438 images depicting clean roads and 2703 images depicting dirty roads. These images were captured using a camera mounted on a vehicle and were augmented through image rotation, translation, cropping, scaling, and other transformations using natural and/or artificial lighting. All images were resized to 1028 x 926 pixels. Figure 6 illustrates representative images from the two categories of the self-built dataset. The dataset was split into training and validation sets in an 8:2 ratio, and the images were cropped to 224 x 224 pixels and normalized between 0 and 1 before being fed into the network.