Performance Evaluation of Anomaly Detection with a New Battery Surface Anomaly Dataset

With the continuous improvement of battery technology and its expanding applications, there has been a surge in demand for high-performance lithium-ion batteries. However, various surface defects such as minor scratches, dust, and stains may occur during the battery production process. Traditional detection relies on error-prone manual inspection, which is inefficient. In contrast, automatic detection based on deep learning can greatly enhance speed and accuracy. In this paper, we introduce a dataset specifically for the task of anomaly detection in industrial scenarios named "Battery Surface Anomaly Dataset", abbreviated as BSA Dataset. The BSA Dataset contains 5500 battery images, including 500 with annotated anomalies, collected through Automatic Optical Inspection (AOI). It provides high-quality data support to promote the application and development of deep learning technologies in the field of battery anomaly detection. Moreover, we have conducted extensive experiments with existing advanced anomaly detection algorithms to validate the utility and challenge of the dataset. These results serve as a benchmark for researchers comparing their models.