A Microbiological Image Repository of Escherichia coli and Klebsiella pneumoniae Bacterial Colonies on MacConkey Agar

This dataset consists of images of two types of bacterial strains streaked on MacConkey agar plates. The images of the bacterial colonies were taken under two different shooting conditions, “controlled” and “uncontrolled” as described in “steps to reproduce” section. These bacterial strains are: 1- Escherichia coli (E. coli): the number of images under controlled conditions is (168) and the number of images under uncontrolled conditions is (3532). 2- Klebsiella pneumoniae (K. pneumoniae): the number of images under controlled conditions is (152) and the number of images under uncontrolled conditions is (3513). IN THE REPOSITORY, YOU WILL FIND: Group 1 consists of images taken from 39 and 36 plates of K. pneumoniae and E. coli respectively, under controlled and uncontrolled conditions. Group2 consists of images taken from 25 plates of K. pneumoniae and E. coli, under controlled and uncontrolled conditions. An excel sheet detailing the numbers of images in the folders. NOTABLE FINDING: Baseline CNNs trained on this data achieved high accuracy, indicating that phone images provide sufficient discriminative signal without expert inspection. Please refer to: 1) S. A. Nagro et al., "Automatic Identification of Single Bacterial Colonies Using Deep and Transfer Learning," in IEEE Access, vol. 10, pp. 120181-120190, 2022, DOI: https://doi.org/10.1109/ACCESS.2022.3221958 2) M. Kutbi et al., "Leveraging Smartphone Imaging and Deep Transfer Learning for Bacterial Colony Classification: From Uncontrolled to Controlled Settings," in IEEE Access, doi: https://doi.org/10.1109/ACCESS.2025.3625648 HOW THIS DATASET CAN BE USED: This dataset can be utilized in any research interested in recognizing different features of bacterial colonies. The dataset can also be used to train and evaluate deep learning models for colony classification, while also supporting studies on robustness, generalization, and practical deployment, to advance computer vision and AI applications in microbiology. By combining clinically important bacteria with controlled and uncontrolled imaging, the dataset offers a realistic and accessible resource for researchers interested in developing AI methods that perform reliably in laboratory and non-laboratory environments. IMPORTANT NOTE: This dataset was expanded to include different types of bacterial strains and culture media which can be found in [DOI: 10.17632/v54x8jdx5x.1] IF YOU USE THIS DATASET, PLEASE REFERENCE THE FOLLOWING: 1. DOI: https://doi.org/10.1109/ACCESS.2022.3221958 2. DOI: https://doi.org/10.1109/ACCESS.2025.3625648 3. DOI: https://doi.org/10.17632/kx6gz3wmcf.1 4. DOI: https://doi.org/10.17632/v54x8jdx5x.1本数据集包含两种在麦康凯琼脂平板（MacConkey agar plates）上划线培养的细菌菌株的图像。该细菌菌落图像在两种不同拍摄条件下采集：受控条件（controlled）与非受控条件（uncontrolled），具体说明详见“复现步骤（steps to reproduce）”章节。这两种细菌菌株分别为： 1. 大肠埃希菌（Escherichia coli，E. coli）：受控条件下的图像数量为168张，非受控条件下的图像数量为3532张。 2. 肺炎克雷伯菌（Klebsiella pneumoniae，K. pneumoniae）：受控条件下的图像数量为152张，非受控条件下的图像数量为3513张。 ## 数据集仓库内容 本数据集仓库包含以下内容： 组1包含分别来自39块肺炎克雷伯菌平板与36块大肠埃希菌平板的图像，涵盖受控与非受控两种拍摄条件。 组2包含分别来自25块肺炎克雷伯菌平板与25块大肠埃希菌平板的图像，涵盖受控与非受控两种拍摄条件。 附带一份Excel表格，用于说明各文件夹内的图像数量。 ## 重要发现 基于本数据集训练的基准卷积神经网络（Convolutional Neural Network, CNN）取得了较高的分类精度，这表明手机拍摄的图像无需专业人员检视即可提供足够的判别信号。具体可参考以下文献： 1. S. A. Nagro等人，《基于深度学习与迁移学习的单一细菌菌落自动识别》，发表于*IEEE Access*，第10卷，第120181-120190页，2022年，DOI：https://doi.org/10.1109/ACCESS.2022.3221958 2. M. Kutbi等人，《利用智能手机成像与深度迁移学习进行细菌菌落分类：从非受控到受控拍摄场景》，发表于*IEEE Access*，DOI：https://doi.org/10.1109/ACCESS.2025.3625648 ## 数据集应用场景 本数据集可用于所有关注细菌菌落特征识别的研究工作。该数据集还可用于训练与评估用于菌落分类的深度学习模型，同时支持关于模型鲁棒性、泛化能力与实际部署的相关研究，以推动计算机视觉与人工智能在微生物学领域的应用发展。本数据集结合了临床重要菌株与受控/非受控拍摄条件，可为致力于开发可在实验室与非实验室环境中稳定运行的人工智能方法的研究人员提供贴合实际且易于获取的研究资源。 ## 重要说明 本数据集已进行扩展，新增了多种细菌菌株与培养基的相关数据，扩展数据集可通过DOI：10.17632/v54x8jdx5x.1获取。 ## 引用要求 若使用本数据集，请引用以下文献： 1. DOI: https://doi.org/10.1109/ACCESS.2022.3221958 2. DOI: https://doi.org/10.1109/ACCESS.2025.3625648 3. DOI: https://doi.org/10.17632/kx6gz3wmcf.1 4. DOI: https://doi.org/10.17632/v54x8jdx5x.1