<p>The issue of road cracking is particularly severe in the mountainous regions of industrial cities. To overcome the challenges of low detection accuracy and high computational costs associated with the SSD (Single Shot Multibox Detector) algorithm on the MRCD (Mountain Road Crack Dataset), this study introduces a novel, lightweight crack detection algorithm, NESSD (Nano Enhanced Single Shot Multi-box Detector). NESSD improves upon existing detection models by integrating a multi-scale feature fusion structure and using MobileNetV2 as the backbone network, which significantly reduces the model size without sacrificing accuracy. A key innovation of NESSD is its incorporation of HATSS (Hierarchical Adaptive Training Sample Selection), which enhances the matching process of positive and negative samples, and the use of CAWR (Cosine Annealing Warm Restarts) for efficient learning rate scheduling. These innovations enable NESSD to achieve a model size of only 2.24 million parameters, with a detection accuracy of 61.27%, demonstrating a 74.81% improvement over SSD and a 3.43% improvement over YOLOv10n. The proposed approach offers a substantial reduction in computational cost while maintaining competitive accuracy, making it a promising solution for real-time crack detection in mountainous road networks.</p>