CoFly-WeedDB: A UAV image dataset for weed detection and species identification

Marios Krestenitis; Emmanuel K Raptis; Athanasios Ch Kapoutsis; Konstantinos Ioannidis; Elias B Kosmatopoulos; Stefanos Vrochidis; Ioannis Kompatsiaris

doi:10.1016/j.dib.2022.108575

CoFly-WeedDB: A UAV image dataset for weed detection and species identification

Data Brief. 2022 Sep 5:45:108575. doi: 10.1016/j.dib.2022.108575. eCollection 2022 Dec.

Authors

Marios Krestenitis¹, Emmanuel K Raptis^{1

2}, Athanasios Ch Kapoutsis¹, Konstantinos Ioannidis¹, Elias B Kosmatopoulos^{1

2}, Stefanos Vrochidis¹, Ioannis Kompatsiaris¹

Affiliations

¹ Information Technologies Institute, The Centre for Research and Technology, Hellas, Thessaloniki 57001, Greece.
² Department of Electrical and Computer Engineering, Democritus University of Thrace, Xanthi 67100, Greece.

Abstract

The CoFly-WeedDB contains 201 RGB images (∼436 MB) from the attached camera of DJI Phantom Pro 4 from a cotton field in Larissa, Greece during the first stages of plant growth. The 1280 × 720 RGB images were collected while the Unmanned Aerial Vehicle (UAV) was performing a coverage mission over the field's area. During the designed mission, the camera angle was adjusted to -87°, vertically with the field. The flight altitude and speed of the UAV were equal to 5 m and 3 m/s, respectively, aiming to provide a close and clear view of the weed instances. All images have been annotated by expert agronomists using the LabelMe annotation tool, providing the exact boundaries of 3 types of common weeds in this type of crop, namely (i) Johnson grass, (ii) Field bindweed, and (iii) Purslane. The dataset can be used alone and in combination with other datasets to develop AI-based methodologies for automatic weed segmentation and classification purposes.

Keywords: Deep convolutional neural networks; Precision agriculture; Semantic segmentation; UAV dataset; Weed detection.