An improved YOLO-Tiny-attention lightweight deep learning method is developed due to the difficulties of deploying deep learning algorithms for wind turbine blade aerial image identification and damage classification to edge computing platforms. The algorithm incorporates the Spatial-Channel Attention Module, which improves the spatial attention of the network to target localization and the channel attention to classification. The extra Yolo head is added to make the network detect multi-scale targets with higher precision. Our method demonstrates a higher detection accuracy compared to YOLOv7-Tiny, particularly with regards to small-sized objects. The whole detection process speed on a RTX3090 GPU reaches 92 FPS while ensuring detection precision (>90%), and the speed on an NVIDIA Jetson Xavier NX edge computing platform reaches 23 FPS, meeting the requirements in wind turbine blade real-time inspection.