Abstract: The use of deep convolutional neural networks for object detection is a common method for foreign object detection in power inspection. Training a neural network requires a large number of training examples. However, it is difficult to collect pictures of the power scenes so the training examples become insufficient. In order to facilitate the training of neural networks and improve the accuracy of object detectors, this paper suggests an approach of automatically generating a large number of images that match the actual scenes by using a context modeling with existing pictures. This approach builds a context model in the convolutional neural network, synthesizes the images with Poisson blending, and combines some other methods such as scaling, rotating and filtering in the digital image processing. Example analysis in this paper verifies the data augmentation approach can augment the training examples to be used in the object detection and to generate a certain amount of images where there is a lack of data, which will improve the accuracy of object detectors.