AUTHOR=Wang Pei , Tang Yin , Luo Fan , Wang Lihong , Li Chengsong , Niu Qi , Li Hui TITLE=Weed25: A deep learning dataset for weed identification JOURNAL=Frontiers in Plant Science VOLUME=Volume 13 - 2022 YEAR=2022 URL=https://www.frontiersin.org/journals/plant-science/articles/10.3389/fpls.2022.1053329 DOI=10.3389/fpls.2022.1053329 ISSN=1664-462X ABSTRACT=Weed suppression is an important factor affecting crop yields. The accurate identification of weeds will contribute to the improvement of automated weeding precision, and thus avoiding certain economic losses due to crop injury. However, the lack of data sets involved in agricultural production has limited the application of deep learning techniques in crop management. This paper collects a relatively large dataset of weeds in cereal crops, Weed25, which mainly contains 14,035 images of 25 different species of weeds. It includes images of grasses and dicot weeds. Meanwhile, weed images at different growth stages were also recorded. To verify the application value of Weed25 in the subsequent identification the high-performance deep learning detection models YOLOv3, YOLOv5, and Faster R-CNN were applied for identification model training using this dataset. The results showed that the average accuracy of detection under the same conditions is 91.8%, 92.4%, and 81.9% respectively. It presented that Weed25 had certain applicability for further development of actual field weed identification and the application of intelligent weed control technology in the future.