An enhanced YOLOv8 and its application for risk prevention in underground excavation operations

Small target detection is used in underground underground excavation operations. At present, there are still challenges and problems in small target detection. Among them, small targets have low resolution in target detection, have few features that can be extracted, and are susceptible to environmental noise interference. In addition, it is difficult to accurately extract features for targets that span multiple local areas and have odd shapes. In response to these problems, this paper proposes an enhanced YOLOv8 object detection algorithm for safety precautions in underground excavation operations. Among them, a new multi-scale feature fusion module is proposed, and the original P2 layer of the benchmark model is added to the feature pyramid of the neck network for feature fusion to improve the detection accuracy of small targets. The Global Attention Mechanism is added to the backbone network of the benchmark model so that the model can focus on more important key features. In addition, it is proposed to introduce the deformable convolution DCNv3 into the benchmark model structure to replace the original fixed convolution kernel, making the convolution operation more flexible and adaptable, and increasing the receptive field required for the detection task. The improved algorithm improves the detection effect of small targets on the COCO dataset, and the algorithm is applied to the dark mining scenario, and it is also improved on the dark mining dataset. The experimental results prove the effectiveness of this method.