Mango Orchard Aerial Image Dataset for Research on Fruit Trees with UAVs in the Precision Agriculture

Background: Agriculture plays a vital role in the economic development of many countries worldwide. The integration of cost-effective artificial intelligence (AI) solutions into farming practices has the potential to significantly enhance productivity and improve the livelihood of farmers. One of the major challenges in automating agricultural tasks such as pesticide spraying on fruit trees, is the lack of sufficient labeled datasets. This limitation hinders the development of robust AI models for accurate tree detection, segmentation and spraying. A well-structured and diverse database is therefore essential for training effective AI models that can support precision agriculture. Methods: In this work, we propose the systematic collection of aerial images of mango orchards and the creation of mango fruit image database captured using a DJI Air 2S UAV (Unmanned Aerial Vehicle). We also proposed image processing algorithms to divide, classify and annotate the images. These images are collected from different Indian farms where high numbers of mango orchards are found. The database contains 3917 mango tree images, which were manually sorted after collection. The images, which are large in size, are divided into smaller frames to make further image processing tasks easy and then those frames are classified based on the tree availability using different image processing techniques such as HSV, tree contours, contour area and eccentricity threshold. The count we got is 69607 frames with the presence of trees and 8713 frames with the absence of trees. Later, image processing algorithm was designed to annotate tree positions in images in (You Only Look Once) YOLO format within a specified directory structure, with tasks such as identifying tree positions based on color thresholds and drawing bounding boxes around them. Results: A total of 58,608 frames were annotated using the YOLO format. Frames that did not contain mango trees were excluded. Annotation was performed through user-interaction-based methods to ensure high-quality labeling. As a result, all annotated images are confirmed to contain mango trees or parts of mango trees only. These images are accessible to anyone interested in the research in related domains. Through this paper, we are publishing our UAV-captured original mango orchard image dataset and its auto-annotated image dataset for the Image processing, AI, ML and DL based UAVs application developments.