Application of a Mask R-CNN-Based Deep Learning Model Combined with the Retinex Image Enhancement Algorithm for Detecting Rockfall and Potholes on Hill Roads

This study presents a Mask R-CNN-based deep learning model combined with the Retinex image enhancement algorithm. The model was applied to detect the presence of rockfall and potholes in front of a vehicle traveling on hill roads. First, 611 images were sourced from online platforms to produce a training set. To increase the diversity and realism of the dataset, images captured under daytime, nighttime, foggy, and rainy conditions were included. Next, the images in the training set were categorized into three classes—low-resolution images, high-resolution images, and Retinex-enhanced images. The test results showed that the mean average precision of the proposed model was 73.54%, higher than the 64.45% achieved by the control group, which was a Mask R-CNN-based model combined with the Canny edge detection algorithm. These results attest to the superiority of the proposed model.