In this project, we address the problem of concept drift in image data, caused by the introduction of new classes in Multi-class Classification. Our proposed method combines a pre-trained transformer model with a dimension reduction algorithm and a distance-based algorithm to detect drifting samples. Through experiments and simulations of the drift on image datasets, we showed the effectiveness of our method in detecting drifting samples and addressing concept drift in classification tasks, as well as providing semantic insights into data distribution. We plan to experiment our method with different image datasets to include more diverse data sources such as remote sensing data, satellite images, and medical images or image data with different resolutions and qualities.