Abstract: This paper aims to investigate the effectiveness of data augmentation techniques in foreign object detection within the electric power scenario. The focus is on enhancing the performance of detection algorithms by integrating small sample data to augment the training dataset. The variety and complexity of foreign object types in power line corridors pose a challenge as each category has a small amount of data, making it difficult to acquire a large number of samples for training. Consequently, the improvement of model performance is constrained by the quantity of training samples. This paper proposes a method that generates a new dataset of tower-kite images by fusing multiple background data and various foreign object kite samples. Additionally, basic data augmentation techniques are employed to enhance the generalization capability and robustness of the algorithm model. Based on the YOLOv5 object detection algorithm, this paper conducts training on the augmented dataset, achieving foreign object detection and recognition in power line corridors. The research outcomes of this paper offer novel approaches and methods for foreign object detection in electric power scenarios, serving as a reference for related studies and holding significant theoretical and practical implications.