A Dataset of Citrus Fruit Images
Description
This article creates a citrus image dataset, which contains citrus images of different varieties. All images were shot with Sony DSC-HX300 in Fengyuan District and Taiping District, Taichung City, Taiwan. The images are divided into the following four categories according to their varieties, namely, Murcott, Ponkan, Tankan, and Tangerines. The image format is JPG with 3648*2736 pixels, and a total of 1067 original images are collected. The image is augmented to 6042 through data enhancement methods such as image flipping and rotation. This dataset provides researchers to study different algorithms of machine learning or deep learning for image classification, object detection and other fields.
Files
Steps to reproduce
This data set is processed through three steps: image acquisition, image preprocessing, and image augmentation. The processing steps of the citrus fruit image data set will be described as follows: (1). Image acquisition The citrus image dataset was shot with Sony DSC-HX300 in Fengyuan District and Taiping District, Taichung City, Taiwan. The weather on the shooting day was sunny, and multi-angle photos were taken at a distance of 200-500 mm from the citrus. All image format is JPG, all image pixels are 3648*2736, a total of 1067 original images. (2). Image preprocessing Divide the captured images of various varieties of citrus fruits into four categories: (a) 280 images of Murcott, (b) 328 images of Ponkan, (c) 371 images of Tankan, and (d) 88 images of Tangerines, a total of 1067 images. The training set, verification set, and test set are divided according to the ratio of 70:20:10 for researchers to train their deep learning models. (3). Image augmentation In order to improve the quantity and quality of images in the training data set, this paper augments the training and verification of four citrus fruit images using six data augmentation methods including flipping the images horizontally, increasing image brightness, increasing image contrast, increasing image color, and rotating 30 degrees. All image format is JPG, all image pixels are 3648*2736, a total of 6042 images were acquired after data augmentation.