Quality issues in battery production drive manufacturers to enhance the efficiency and accuracy of quality inspection. Building a battery surface defect dataset faces challenges, as dataset via manual construction is inefficient. Existing similar defect datasets do not match the characteristics of battery surface defects and lack in-depth descriptions of defect categories. To delve deeper into research and provide finer defect representations, we created the Battery Surface RGB Dataset (BSR), comprising 2499 battery images covering three defect types and over 6800 defect samples. In this paper, we discuss the annotation techniques we used and the challenges we faced, and conduct a statistical analysis of the dataset, comparing it in detail with publicly available defect datasets. Finally, we extensively experiment with the BSR dataset on the most representative methods in recent years, evaluating algorithm performance in defect semantic segmentation tasks, providing a benchmark and reference for subsequent research.