Dataset for the research paper 'Machine learning techniques for plant disease detection: an evaluation with a customized dataset'
Description
This is a customized dataset used in the research paper "Machine learning techniques for plant disease detection: an evaluation with a customized dataset," which was accepted for publication in the peer-reviewed journal International Journal of Informatics and Communication Technology (IJ-ICT). We constructed this dataset by cleaning and combining multiple open datasets (as cited in our article). There are 87,570 records and 97 classes in the dataset. It includes 32 different plant species and 74 distinct disease forms. The images in the dataset were programmatically chosen at random and divided into train and test sets in a 70:30 ratio. The train and test sets have 61,259 and 26,311 images respectively. The dataset contains leaf images from both laboratory setups and cultivation fields, making it more representative than most existing datasets, which only include images from lab-controlled settings and frequently contain a small number of records and classes that are unsuitable for real-world applications. To the best of our knowledge, no such datasets have been used for deep learning models. The paper is available at http://doi.org/10.11591/ijict.v12i2.pp127-139.