ABSTRACT
The maintenance of road pavements is essential to ensuring the safety and efficiency of land transportation. With advancements in artificial intelligence and deep learning, new possibilities have emerged for automating pavement defect detection. However, the quality of the images used to train these models plays a crucial role in their performance, as variations in lighting or high contrast can compromise the precise identification of defects. In this study, we analyzed the impact of image preprocessing techniques on defect detection using a deep learning model. The CLAHE (Contrast Limited Adaptive Histogram Equalization), LIME (Low-Light Image Enhancement), and gamma adjustment methods were applied, along with unprocessed images. The experiments revealed that the LIME method achieved the best performance, with 78,8% mAP, 91,1% in Precision, 86% in Recall, and an F1 Score of 74%. This study highlights that preprocessing techniques are promising tools for improving the accuracy and reliability of deep learning models applied to the automated detection of defects in asphalt pavement surfaces, significantly contributing to the development of more efficient and accessible pavement management systems.