Banana Leaf Disease Images

Huge data is one of the required resources in deep learning research to train, validate and test CNN model and get better accuracy, prediction and detection. In our study we have prepared a banana plant leaf image dataset that is used for the ‘banana disease detection’ research. It is collected from Southern Nations, Nationalities, and Peoples' Regional State Arbaminch Zuria Woreda Lante kebele and Chano kebele and Gamugofa Zone Mierab Abaya Woreda Omolante kebele where banana is widely producing area and the infection of Xanthomonas wilt and Segatoka leaf spot disease is highly observed. The data is collected from four farmers a size of one hectare farm each in three kebeles. During the data collection, the daily collected data were identified “health” or “infected” by both type of diseases. The labeled data by the first plant pathologist is verified and confirmed by the second one to make sure the quality of the collected data. Finally, the collected image was correctly labeled with three classes. Collecting images of banana plant leaf in thousands is too difficult. The researcher collects 1,288 pictures of banana leaf under three categories as “Health” banana leaf, “Xanthomonas” infected leafs and “Sigatoka” infected leafs.海量数据是深度学习研究中训练、验证与测试卷积神经网络（Convolutional Neural Network, CNN）模型以获得更优准确率、预测性能与检测效果的必要资源之一。本研究构建了一款用于"香蕉病害检测"研究的香蕉植株叶片图像数据集。该数据集采集自埃塞俄比亚南方各族州（Southern Nations, Nationalities, and Peoples' Regional State）的Arbaminch Zuria Woreda、Lante kebele、Chano kebele，以及Gamugofa专区Mierab Abaya Woreda的Omolante kebele；上述区域为香蕉主产区，且普遍发生细菌性枯萎病（Xanthomonas wilt）与Sigatoka叶斑病的侵染现象。数据采集自3个kebele中的4位农户，每位农户的种植地块面积为1公顷。采集过程中，每日采集的样本均被标注为"健康"或"受侵染"状态，由首位植物病理学家完成标注后，经第二位植物病理学家复核确认，以保障采集数据的质量。最终，采集得到的图像被准确划分为三个类别。人工采集数千张香蕉植株叶片图像难度极大，本研究共收集1288张香蕉叶片图像，分为三类："健康"香蕉叶片、受Xanthomonas侵染的叶片，以及受Sigatoka侵染的叶片。