绝缘子作为输电线路中大量使用的基础元件，对于电力系统的稳定运行发挥着至关重要的作用，但因其长期暴露在自然环境中，容易发生故障，严重影响电力输送的可靠性和安全性。因此，需对绝缘子进行定期巡检，及时发现故障绝缘子。传统人工巡检方法耗费时力，而随着无人机技术的不断发展，具有图像采集功能的无人机和深度学习检测方法相结合的巡检方式变得更加方便快捷。但是，由于电力行业数据获取难度大导致数据集匮乏，这对检测模型的训练具有很大影响；同时，现有航拍图像绝缘子故障检测方法存在检测精度低的问题。因此，研究数据集增强和故障检测方法很有必要。

       针对电力行业数据集匮乏、传统数据集增强方法无法起到明显的增强效果以及基于生成对抗网络的数据集增强方法存在模式崩溃的问题，提出了双判别器加权混合生成对抗网络。首先，理论证明双判别器加权混合生成对抗网络的条件最优性，表明生成器能够学习到真实数据分布。其次，在人工合成的二维数据集、MNIST数据以及CIFAR-10这三个数据集上对双判别器加权混合生成对抗网络的性能进行验证。然后，将双判别器加权混合生成对抗网络应用于绝缘子数据集增强中，分别设计生成器、判别器和分类器的模型结构与损失函数。最后，对生成对抗网络、双判别器生成对抗网络、混合生成对抗网络以及双判别器加权混合生成对抗网络四种方法进行实验对比分析，验证了双判别器加权混合生成对抗网络对数据集的增强能力。

       针对现有航拍图像绝缘子故障检测方法存在特征信息利用不足而造成的目标检测算法检测精度低的问题，提出了基于特征深度融合的航拍图像绝缘子故障检测方法。首先，通过将K-means++聚类算法和遗传变异算法相结合获得合适的锚框参数。然后，对基于特征深度融合的航拍图像绝缘子故障检测方法的模型结构以及损失函数进行设计，并对检测模型的训练方式进行设置。最后，通过绝缘子故障检测实验验证提出方法的有效性并与其他方法进行对比分析。实验结果表明，本文提出方法对绝缘子故障检测的召回率、准确率、mAP值和F1值相对于其他检测方法都有提升，验证了提出方法的优越性。

   As a basic component widely used in transmission lines, insulators play a vital role in the stable operation of power systems. However, due to its long-term exposure to the natural environment, it is prone to failure, which seriously affects the reliability and safety of power transmission. Therefore, it is necessary to conduct regular inspection on insulators to find faulty insulators in time. The traditional manual inspection method is time-consuming and laborious. With the continuous development of unmanned aerial vehicle technology, the inspection method combining the unmanned aerial vehicle with image acquisition function and detection method in deep learning has become more convenient and faster. However, due to the difficulty of data acquisition in the power industry, the lack of data sets has a great impact on the training of detection models. At the same time, the existing aerial image insulator fault detection methods have the problem of low detection accuracy. Therefore, it is necessary to study dataset augmentation and fault detection methods.

   Aiming at the lack of data sets in the power industry, the inability of traditional data set enhancement methods to achieve significant enhancement effects, and the problem of mode collapse in the data set enhancement methods based on Generative Adversarial Network, a Dual Discriminator Weighted Mixture Generative Adversarial Network is proposed. Firstly, the conditional optimality of the Dual Discriminator Weighted Mixture Generative Adversarial Network is proved theoretically, which shows that the generator can learn the real data distribution. Secondly, the performance of the Dual Discriminator Weighted Mixture Generative Adversarial Network is verified on three datasets of artificially synthesized 2D data, MNIST datasets, and CIFAR-10 datasets. Then, the Dual Discriminator Weighted Mixture Generative Adversarial Network is applied to the enhancement of insulator dataset, and the model structure and loss function of the generator, the discriminator, and the classifier are designed respectively. Finally, the four methods of Generative Adversarial Network, Dual Discriminator Generative Adversarial Network, Mixture Generative Adversarial Network, and Dual Discriminator Weighted Mixture Generative Adversarial Network are compared and analyzed experimentally, which verifies the enhancement ability of the Dual Discriminator Weighted Mixture Generative Adversarial Network to the dataset.

    Aiming at the problem of low detection accuracy of target detection algorithms caused by insufficient use of feature information in the existing aerial image insulator fault detection methods, an insulator fault detection method in aerial image based on feature depth fusion is proposed. Firstly, the appropriate anchor frame parameters are obtained by combining K-means++ clustering algorithm and genetic mutation algorithm. Then, the model structure and loss function of the insulator fault detection method in aerial image based on feature depth fusion are designed, and the training mode of the detection model is set. Finally, the effectiveness of the proposed method is verified by insulator fault detection experiments and compared with other methods. The experimental results show that the method proposed in this paper improves the recall rate, accuracy rate, mAP value and F1 value of insulator fault detection compared with other detection methods, which verifies the superiority of the proposed method.