A Lightweight Risk Advertising Image Detection Method Based on Mobile-ViT

To test risky advertising images efficiently on the Internet, this paper brings a recognition method based on Mobile-ViT. By integrating the advantages of CNN, training task can be finished on small data sets. Premised on multi-head attention mechanism, can further enhance the testing effect along with the increase of data sets. As the model is provided with fewer parameters, there is also less cost in training and reasoning, and real-time testing can be completed by deploying the model on equipment with limited computing power. In addition, Mobile-ViT will be integrated with other networks apart, to achieve the purpose of boosting the accuracy of the model.