Data for: Deep learning-based early weed segmentation using motion blurred UAV images of sorghum fields
Description
Weeds are undesired plants in agricultural fields that affect crop yield and quality by competing for nutrients, water, sunlight and space. Site-specific weed management (SSWM) through variable rate herbicide application and mechanical weed control have long been recommended in order to reduce the amount of herbicide and impact caused by uniform spraying. Accurate detection and classification of weeds in crop fields is a crucial first step for implementing such precise strategies. Drones are commonly used for image capturing but high wind pressure and different drone settings have a severe effect on the image quality, which potentially results in degraded images, e.g. due to motion blur. We publish a manually annotated and expert curated drone image dataset for weed detection in sorghum fields under challenging conditions. Our results show that our trained models generalize well regarding the detection of weeds, even for degraded captures due to motion blur. An UNet-like architecture with ResNet-34 as feature extractor achieved an F1-score of over 89 % on a hold-out test-set. Further analysis indicate that the trained model performed well in predicting the general plant shape, while most mis-classifications appeared at borders of the plants. Beyond that, our approach can detect intra-row weeds without additional information as well as partly occluded plants in contrast to existing research. Github link: https://github.com/grimmlab/UAVWeedSegmentation Please cite our original publication if you have used the data in your project or in any follow-up analysis (https://doi.org/10.1016/j.compag.2022.107388): @article{GENZE2022107388, title = {Deep learning-based early weed segmentation using motion blurred UAV images of sorghum fields}, journal = {Computers and Electronics in Agriculture}, volume = {202}, pages = {107388}, year = {2022}, issn = {0168-1699}, doi = {https://doi.org/10.1016/j.compag.2022.107388}, url = {https://www.sciencedirect.com/science/article/pii/S0168169922006962}, author = {Nikita Genze and Raymond Ajekwe and Zeynep Güreli and Florian Haselbeck and Michael Grieb and Dominik G. Grimm}, keywords = {Deep learning, Weed detection, Weed segmentation, UAV, Precision agriculture}, }
Files
Steps to reproduce
Images collected by an UAV on an experimental sorghum field (BBCH 17) in Southern Germany provided data for this study. The sorghum variety “Farmsughro 180” was sown at 37.5 cm row spacing with a seeding density of 25 seeds per m². During image capture, several weed species were present on the field. These weeds comprised mostly of dicotyledons namely, Goosefoot (Chenopodium album L.), Field pennycress (Thlaspi arvense), Wild chamomile (Matricaria chamomilla), Common gypsyweed (Veronica officinalis) and Cotton thistle (Onopordum acanthium). Beyond that, a consumer-grade drone “DJI Mavic 2 Pro” fitted with a 20 MP Hasselblad camera (L1D-20c) that captures images with a resolution of 5472x3648 pixels² was used. Automated drone flight with camera pointing nadir and a capture overlap of ten percent was carried out at a flight altitude of five meters above ground level. At this altitude, the corresponding GSD was one millimeter, and therefore precise enough to recognize sorghum and weeds in early growth stages.