Surveillance drones are remarkable devices for monitoring, as they have strong spatial and remote sensing capabilities. The prompt detection of peripheral damage to the blades of wind turbines is necessary to reduce downtime and prevent the potential failure of wind farms. Computer vision breakthroughs with deep learning have developed and been refined over time, mainly using convolution neural networks. From this perspective, we suggest a deep learning model for monitoring and diagnosing the blade health of wind turbines based on images captured by surveillance drones. The main limitations of standard monitoring devices are their poor detection accuracy and lack of real-time performance, making it complex to obtain the attributes of blades from aerial images. Based on the foregoing, this study introduces a method for increasing detection accuracy when carrying out operations in real time using You Only Look at Once version 3 (YOLOv3). We train and evaluate three deep learning models on the wind turbine image dataset. We find that many aerial images are unclear because of blurred motion. As avoiding such low-resolution images for training can affect accuracy, we use a super-resolution convolution neural network to reconstruct a blurred picture as a high-resolution one. The computational results demonstrate that YOLOv3 outperforms traditional models in terms of both accuracy and handling time.

监控无人机具有强大的空间和遥感能力，是监控的杰出设备。必须迅速检测出风力涡轮机叶片的外围损坏，以减少停机时间并防止风电场潜在故障。深度学习在计算机视觉方面的突破已经发展并不断完善，主要是使用卷积神经网络。从这个角度出发，我们提出了一种基于监视无人机捕获的图像的，用于监视和诊断风力涡轮机叶片健康的深度学习模型。标准监视设备的主要局限性在于其检测精度差和缺乏实时性能，这使得从航拍图像中获取叶片的属性变得很复杂。基于以上所述，这项研究介绍了一种使用“仅一次查看”版本3（YOLOv3）实时执行操作时提高检测精度的方法。我们在风力涡轮机图像数据集上训练和评估三个深度学习模型。我们发现由于运动模糊，许多航拍图像不清楚。由于避免使用此类低分辨率图像进行训练会影响准确性，因此我们使用超分辨率卷积神经网络将模糊图片重构为高分辨率图片。计算结果表明，YOLOv3在准确性和处理时间方面均优于传统模型。我们发现由于运动模糊，许多航拍图像不清楚。由于避免使用此类低分辨率图像进行训练会影响准确性，因此我们使用超分辨率卷积神经网络将模糊图片重构为高分辨率图片。计算结果表明，YOLOv3在准确性和处理时间方面均优于传统模型。我们发现由于运动模糊，许多航拍图像不清楚。由于避免使用此类低分辨率图像进行训练会影响准确性，因此我们使用超分辨率卷积神经网络将模糊图片重构为高分辨率图片。计算结果表明，YOLOv3在准确性和处理时间方面均优于传统模型。