Accurate and real-time detection of rice pests is crucial for protecting crop yield and advancing precision agriculture. However, existing models often suffer from limitations in small-object recognition, background interference, and computational efficiency. To overcome these challenges, this study proposes an improved lightweight detection framework, CRRE-YOLO, developed based on YOLOv11. The model integrates four enhanced components—the EIoU loss function, C2PSA_ELA module, RPAPAttention mechanism, and RIMSCConv module—to improve localization accuracy, feature extraction, and fine-grained pest recognition. Experimental results on the RP11-Augmented dataset show that CRRE-YOLO achieves 0.852 precision, 0.787 recall, 83.6% mAP@0.5, and 71.9% mAP@0.5:0.95, outperforming YOLOv11 by up to 7.8% and surpassing YOLOv8 and RT-DETR in accuracy while maintaining only 2.344M parameters and 6.1G FLOPs. These results demonstrate that CRRE-YOLO achieves an optimal balance between accuracy and efficiency, providing a practical and deployable solution for real-time rice pest detection and offering potential for integration into smart farming and edge computing applications.