Crowdsourcing is a distributed problem solving model that encompasses many types of tasks, and from a machine learning perspective, the development of crowdsourcing provides a new way to obtain manually labeled data with the advantages of lower annotation costs and faster annotation speed very recently, especially in the field of computer vision for image classification and segmentation. Therefore, it is necessary to investigate how to combine machine learning algorithms with crowdsourcing effectively and cost-effectively. In this paper, we propose a deep active learning (AL) framework by combining active learning strategies, CNN models and real datasets, to test the effectiveness of the active learning strategies through multiple scenario comparisons. Experiment results demonstrate the effectiveness of our framework in reducing the data annotation burden. Moreover, Our findings suggest that the strength is often observed in the case of relatively large data scale.