GLD-YOLO: A Lightweight Realtime Object Detection Algorithm Based on Improved YOLOv8

With the acceleration of urbanization, the demand for inspection robots is steadily increasing. In complex environments such as urban underground utility tunnels, inspection robots are required to achieve efficient and real-time detection while meeting the demands for lightweight design and localized deployment to ensure system reliability and stability. However, existing methods often struggle to balance detection accuracy and speed under these constraints. To address this issue, this paper proposes the GLD-YOLO algorithm, aimed at improving the accuracy and speed of object detection in urban underground utility tunnels. First, to enhance the accuracy and speed of real-time object detection, we propose the GHohstHGNetV2 structure, which implements a lightweight approach to the backbone network. Second, the SPPF module within the backbone network is integrated with the LSKA (Large Separable Kernel Attention) attention mechanism, enhancing the extraction of multi-scale features. Finally, the DynamicHead is introduced to improve the accuracy and flexibility of the detector. We selected to validate the results of the experiment within the underground pipeline in city Experimental results show that the improved algorithm achieved 95% in Precision (P) and 96.5% in Mean Average Precision (mAP). Compared to the baseline model, these metrics increased by 10.4% and 6.84% respectively. Additionally, the detection speed of the algorithm reaches 144 FPS, which meets the requirement of real-time inspection robot detection. By comparing with other mainstream target detection algorithms, the effectiveness and excellence of the method is verified.