In response to the problem of requiring a large number of training samples for deep learning-based object detection, a method utilizing artificial intelligence data generation for data augmentation and automatic annotation based on large models is proposed to improve the effectiveness of object detection. This method addresses the lack of data by using artificial intelligence data generation techniques to generate data samples with prompts, and utilizes computer vision large models to automatically annotate the targets. The annotated samples are then mixed with the original data for training. Through verification, for the types of poor object detection caused by insufficient data, the (average precision at IoU threshold 0.5) has been improved from an average of less than 0.2 to an average of over 0.75. This method has reference significance for object detection tasks in scenarios where it is difficult to acquire data samples.