Preventing attackers from interrupting or totallystopping critical services in cloud systems is a vitaland challenging task. Today, machine learning-basedalgorithms and models are widely used, especially forthe intelligent detection of zero-day attacks. Recently,deep learning methods that provide automatic featureextraction are designed to detect attacks automatically.In this study, we constructed a new deep learning modelbased on transfer learning for detecting and protectingcloud systems from malicious attacks. The developeddeep transfer learning-based IDS converts network trafficinto 2D preprocessed feature maps.Then the featuremaps are processed with the transferred and fine-tunedconvolutional layers of the deep learning model beforethe dense layer for detection and classification of trafficdata. The results computed using the NSL-KDDtest dataset reveal that the developed models achieve89.74% multiclass and 92.58% binary classification accuracy.We performed another evaluation using only20% of the training dataset as test data, and 80% fortraining. In this case, the model achieved 99.83% and99.85% multiclass and binary classification accuracy, respectively.