The rapid advancement of information technology and the diffusion of IoT devices have increased the possibility and variety of cyberattacks targeting the IoT devices. Due to limited security measures, IoT devices are highly vulnerable to such threats. Deep learning models like ResNet and EfficientNet have achieved notable accuracy in detecting specific types of IoT attacks, but struggle with cata-strophic forgetting when adapting to new attack types. In this study, we propose a new attack detection approach integrating TreeCNN-based continual learning models with cloud and edge computing architectures, aiming at enhancing responsiveness of model construction and attack detection, and adapting to emerging IoT attacks. Experiment results demonstrated the effectiveness of the proposed approach.