As the global electric vehicle (EV) market grows, so does the frequency of cyberattacks on charging infrastructure, which lacks robust defense mechanisms. Existing network attack detection methods for EV charging stations, often based on machine learning, are limited by their reliance on manual feature engineering and are not optimized for handling large volumes of network traffic. Additionally, basic deep learning models that focus solely on temporal features are inadequate for identifying cyber threats. In this paper, we propose MACROSS, an innovative end-to-end detection model that directly processes raw network packets to extract both temporal and frequency domain features using convolutional neural networks (CNNs) and Fast Fourier Transform (FFT). An attention mechanism is employed to integrate these features, and a fusion feature extraction module based on residual networks learns the fused representation to perform a classification task that detects cyberattacks targeting EV charging stations. We conducted extensive experiments on the CICEVSE2024 dataset. The results demonstrate that MACROSS effectively identifies network attack traffic targeting EV charging stations, achieving 100% and 98.4% accuracy in binary classification and attack classification, respectively, and significantly outperforms other approaches. Our model is publicly available for validation at https://github.com/kisaragi-attention/macross