Lithium-Ion-Cell 18650 Labeled Polarity-Aware Anomaly Dataset

This repository provides a polarity-aware industrial image dataset of cylindrical battery cells captured under realistic pick-and-place assembly conditions. The images were acquired during a representative battery pack assembly process and reflect visual variability typical of small and medium-sized enterprise (SME) manufacturing environments, including variations in placement, background context, and lighting. The dataset consists of cropped single-cell images (256x256 pixel) extracted from images of populated cell holders using a model-assisted auto-labeling and verification pipeline. Each cell image is labeled with its polarity (positive or negative) and grouped by condition (normal or anomalous), where anomalies correspond to visually observable deviations from nominal cell appearance such as damage or missing insulation. In addition to the primary dataset containing one image per unique cell, an extended variant with additional images captured under different lighting conditions and spatial positions is included and kept separate within the repository. The dataset is intended as a practical resource for research on supervised inspection and anomaly detection in assembly-related industrial vision tasks, with an emphasis on feasibility, transparency, and reuse rather than standardized benchmarking.本仓库提供一套面向圆柱电池电芯的极性感知工业图像数据集，采集自真实的取放式装配场景。该数据集的图像采集自典型的电池模组装配流程，能够反映中小型制造企业（Small and Medium-sized Enterprise，SME）生产环境中常见的视觉变化，涵盖放置位置、背景环境与光照条件的各类差异。本数据集包含经裁剪的单电芯图像（分辨率为256×256像素），这些图像从已装载电芯的夹具图像中提取得到，采用模型辅助的自动标注与验证流水线完成处理。每张电芯图像均标注有其极性（正极或负极），并按工况分为正常或异常两类；其中异常工况指可通过视觉观测到的、与标准电芯外观不符的偏差，例如电芯损坏或绝缘层缺失。除了每个唯一电芯对应一张图像的基础数据集外，本仓库还包含扩展变体数据集：该变体采集了不同光照条件与空间位置下的额外图像，并与基础数据集分开存储。本数据集旨在为装配相关工业视觉任务中的监督式检测与异常检测研究提供实用资源，其核心侧重在于实用性、透明度与可复用性，而非标准化基准测试。