Detecting malicious activities in networks has become increasingly challenging as internet usage grows, making network security vital for ensuring secure communication between devices. While machine learning (ML) and deep learning (DL) have been integrated into intrusion detection systems (IDSs), there is limited research exploring their full potential. This paper introduces a hybrid sequential-visual deep feature extractor network (HSV-DFEN) for intrusion detection, using spectrogram transformations to combine visual and sequential data processing techniques. The model leverages convolutional neural networks (CNN) and bidirectional long short-term memory (BiLSTM) networks to extract time-frequency features from network traffic, and employs ensemble learning along with principal component analysis (PCA) for dimensionality reduction. Experimental results on the CICIDS2017 dataset demonstrate that HSV-DFEN achieves an average accuracy of 99.98%, significantly outperforming existing models in detecting various types of attacks, making it an effective solution for anomaly detection in network security
