This paper provides a comprehensive survey on deep facial expression recognition (FER), addressing the challenges of overfitting and expression-unrelated variations in real-world conditions. It covers datasets, algorithms, and techniques for deep FER systems, including pre-processing, deep feature learning, and classification. The survey highlights the evolution of FER from traditional methods to deep learning, emphasizing the importance of large datasets and advanced neural network architectures. It discusses specific techniques such as face alignment, data augmentation, and normalization to improve performance. The paper also reviews state-of-the-art deep neural networks for static and dynamic image sequences, detailing their architectures and training strategies. Additionally, it explores methods for enhancing feature representation, such as auxiliary blocks, loss layers, and network ensembles, and addresses challenges like inter-identity variations and intra-class variations. Finally, it identifies future directions and opportunities in the field of robust deep FER systems.This paper provides a comprehensive survey on deep facial expression recognition (FER), addressing the challenges of overfitting and expression-unrelated variations in real-world conditions. It covers datasets, algorithms, and techniques for deep FER systems, including pre-processing, deep feature learning, and classification. The survey highlights the evolution of FER from traditional methods to deep learning, emphasizing the importance of large datasets and advanced neural network architectures. It discusses specific techniques such as face alignment, data augmentation, and normalization to improve performance. The paper also reviews state-of-the-art deep neural networks for static and dynamic image sequences, detailing their architectures and training strategies. Additionally, it explores methods for enhancing feature representation, such as auxiliary blocks, loss layers, and network ensembles, and addresses challenges like inter-identity variations and intra-class variations. Finally, it identifies future directions and opportunities in the field of robust deep FER systems.