The increasing prevalence of IoT devices introduces unique cybersecurity challenges due to their limited computational resources and exposure to dynamic threats. This paper proposes an intrusion detection framework that integrates random forest-based feature selection and deep learning techniques. Network traffic is transformed into RGB images using clustering-based feature grouping, enabling the application of convolutional neural networks (CNNs) for accurate detection. Our evaluation on two public datasets (N-BaIoT and IoT-23) shows that the proposed pipeline achieves high detection performance, with accuracies up to 99.96%, while remaining efficient for edge deployment. While our method combines established techniques, its strength lies in the integration of lightweight preprocessing and visual modeling tailored to the IoT context.