Ensuring the safety and economic efficiency of nuclear facilities requires regular maintenance, which generates significant amounts of radioactive waste. Proper classification and differentiated treatment of this waste are essential for resource recovery and environmental protection. However, many regions still rely on manual waste sorting, which poses health risks to workers and is both costly and inefficient. To address these challenges, this study proposes a waste detection model named DWW-YOLO, based on YOLOv11, to enhance the accuracy and efficiency of nuclear power plant maintenance waste detection, particularly for small targets. Existing methods often struggle with small objects and low-contrast targets in complex backgrounds, leading to poor detection performance. To overcome these limitations, we introduce a dual-domain wavelet attention network and a WIOU loss function, significantly improving the model's ability to capture fine-grained features and optimize multi-scale feature fusion. Experimental results and ablation studies demonstrate the effectiveness and superiority of the proposed model, achieving higher recall and accuracy rates. This work contributes to the field by (1) proposing a wavelet feature fusion module to enhance high-frequency information, (2) integrating spatial domain information to construct a dual-domain wavelet attention network, and (3) introducing WIOU loss to dynamically adjust the loss contribution of different-scale targets, particularly improving small target detection in complex backgrounds. The proposed model provides a robust solution for intelligent waste classification and detection, aligning with the IAEA's requirements for health protection and risk minimization.