With the rapid development of information technology, the collection and analysis of image data play an increasingly important role in various fields. Deep learning (DL) and Optical Character Recognition (OCR) algorithms, as cutting-edge technologies in artificial intelligence and machine learning, have greatly promoted the progress of image data processing. In order to further understand the performance of different DL models in data OCR recognition image data acquisition, this article iteratively trains different models on the same dataset (COCO-Text dataset) and collects OCR image data. Finally, different models can be analyzed, and the accuracy, precision rate, recall, and F1 scores of GAN (Generative Adversarial Network) are 0.94, 0.93, 0.92, and 0.93, respectively. The analysis shows that when the number of iterations is sufficient, GAN has better OCR image data acquisition performance than other deep learning models; when the number of iterations is insufficient, the OCR image data acquisition performance of GAN decreases significantly. When the number of iterations is sufficient, CNN has better OCR image data acquisition performance; when the number of iterations is insufficient, CNN can still maintain good OCR image data acquisition performance.
