Multi-Scale Distribution Network Equipment Detection Algorithm Based on Improved YOLOv8

This paper proposes a detection algorithm for power distribution network equipment based on an improved YOLOv8 model, to address issues such as low recognition rates and poor detection performance of power distribution network equipment of different scales in the context of power inspection. The improved model introduces an enhanced spatial pyramid module, making multi-scale semantic information fusion more efficient, thereby enhancing the model's semantic representation capability for power distribution network equipment objects of different scales. A feature pyramid network integrating multi-scale location information is proposed to improve the model's localization ability for power distribution network equipment objects of different scales. Additionally, a loss function called Focal SIOU that balances multi-scale objects is introduced, enabling the model to better learn to detect poorly-performing scales of power distribution network equipment objects. Experimental results show that the improved YOLOv8 model exhibits excellent detection performance for equipment of different scales on the power distribution network equipment object detection dataset, with an average detection precision (mAP) increase of 2.4% compared to the original model. The precision for small, medium, and large-scale object detection <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$(\text{AP}_{\mathrm{S}},\text{AP}_{\mathrm{M}}, \text{AP}_{\mathrm{L}})$</tex> respectively increased by 3.0%, 4.8% and 2.7<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>, effectively improving the detection performance for power distribution network equipment objects of all scales.