改进YOLOv8的绝缘子缺陷检测研究
Research on Insulator Defect Detection of Improved YOLOv8

为了维持电力传输的可靠性、安全性和可持续性，绝缘子的缺陷检测成为电力巡检中一项重要的任务。为此，本文提出了绝缘子缺陷检测改进模型P-YOLOv8。通过集成Slim-neck模块和CBAM (Convolutional Block Attention Module)注意力机制模块，显著提升了模型对小目标物体的特征提取能力，从而更好地适应绝缘子缺陷数据集的特性。CBAM注意力机制模块的引入，使得神经网络能够更加聚焦于绝缘子缺陷信息的捕捉；Slim-neck模块是一种用于目标检测的神经网络结构，Slim-neck模块中的模块首先使用轻量级卷积，GSConv它强化了关键特征，使用深度可分离卷积降低了模型的计算成本。通过对YOLOv5和YOLOv8模型的对比分析，我们决定采用经过改进的P-YOLOv8模型作为绝缘子缺陷预测的优选模型。P-YOLOv8在精确度、召回率和mAP-kp上分别达97.2%、98.4%、99.3%，较YOLOv5显著提升。此外，P-YOLOv8在光照强度环境下表现优异，精确度、召回率和mAP-kp分别为92.17%、90.2%、97.6%，较YOLOv8有所提升，这些数据充分证明了改进P-YOLOv8模型不仅在网络模型精度上保持了高水平，同时对光线等复杂环境因素具有较强的鲁棒性。因此，它能够有效地应对复杂环境下的绝缘子缺陷预测任务，为绝缘子的安全监测提供了有力的技术保障。
In order to maintain the reliability, safety, and sustainability of power transmission, defect detection of insulators has become an important task in power inspection. Therefore, this article proposes an improved model P-YOLOv8 for insulator defect detection. By integrating the Slim-neck module and the Convolutional Block Attention Module (CBAM) attention mechanism module, the model’s feature extraction ability for small target objects has been significantly improved, thus better adapting to the characteristics of insulator defect datasets. The introduction of the CBAM attention mechanism module enables neural networks to focus more on capturing insulator defect information; The Slim-neck module is a neural network structure used for object detection. The modules in the Slim-neck module first use lightweight convolution GSConv, which enhances the recognition of key features, and uses depth wise separable convolution to reduce the computational cost of the model. Through comparative analysis of YOLOv5 and YOLOv8 models, we have decided to adopt the improved P-YOLOv8 model as the preferred model for predicting insulator defects. P-YOLOv8 achieved accuracy, recall, and mAP kp of 97.2%, 98.4%, and 99.3%, respectively, significantly improving compared to YOLOv5. In addition, P-YOLOv8 performs excellently in light intensity environments, with accuracy, recall, and mAP kp of 92.17%, 90.2%, and 97.6%, respectively, which is an improvement compared to YOLOv8. These data fully demonstrate that the improved P-YOLOv8 model not only maintains a high level of network model accuracy, but also has strong robustness to complex environmental factors such as light lines. Therefore, it can effectively cope with the task of predicting insulator defects in complex environments, providing strong technical support for the safety monitoring of