<p>Surface defect detection of daily-use ceramics is crucial for ensuring product quality, aesthetics, and usage safety. However, the task is still predominantly reliant on manual visual inspection, which suffers from low accuracy and inefficiency. This is mainly due to the complex characteristics of ceramic defects, which often exhibit multi-scale, small-size, and blurred boundaries, posing higher demands on the perception capability and inference efficiency of detection models. To address these challenges, this paper proposes an improved model based on the YOLO11 framework, named YOLO-CDNet, for efficient and accurate detection of surface defects in daily-use ceramics. Firstly, a dedicated dataset RYTC containing five types of typical defects was constructed, with samples collected and augmented via an industrial camera platform. Secondly, to tackle the insufficient small-object perception of existing models, a Semantic-Aware P2 Fusion structure (SAPF), an Auxiliary Detection Head (SAH), and a Shared Detail Enhanced Head (SDEH) are introduced, enhancing the model’s multi-scale feature modeling capability and improving its ability to discern defect edge details. Experimental results show that YOLO-CDNet achieves 89.5% mAP@0.5 on the RYTC dataset, which is 6.5% higher than the original YOLO11, while maintaining an inference time of 5.6&#xa0;ms, delivering overall performance superior to many mainstream SOTA methods. The proposed model achieves a good balance between detection accuracy and real-time performance, and holds great potential for the automatic detection of daily-use ceramic surface defects.</p>