Early season dominant weed mapping in maize field using unmanned aerial vehicle (UAV) imagery: Towards developing prescription map

Weed mapping plays a crucial role in precision agriculture by providing detailed information on the spatial distribution and density of weeds within a field. This study aimed to create a dominant weed map in a maize ( Zea mays ) field using aerial images captured using a UAV. A Phantom 4 Pro UAV equipped with Sequoia (multispectral) and CMOS (RGB) sensors captured images 17 m above ground level. The dominant weeds were cheeseweed ( Malva parviflora ) and bindweed ( Convolvulus arvensis ). All images were converted into one orthomosaic image using the Pix4D software and then transferred to the ENVI software for classification into four separate classes (soil, maize crop, and two dominant weeds). K-means and ISO-data were used as unsupervised classification methods, while Support Vector Machine (SVM), Maximum Likelihood (ML), Minimum Distance (MD), and Neural Network (NN) were used as supervised classification algorithms. Classification evaluation was performed using overall accuracy (OA) and kappa coefficient. The most accurate result of the algorithm was used to create a prescription map for the herbicide treatment. The K-means and ISO-data algorithms achieved 44.46 % and 40.70 % accuracy, respectively, with kappa coefficients <0.25, indicating their inefficiency in identifying weeds due to low accuracy. Among the supervised algorithms, NN and SVM had the highest accuracies (96.44 % and 95.77 %, respectively), followed by ML (94.33 %), and MD (93.06 %). The supervised classification was more precise due to the higher accuracy and kappa values. This study demonstrated that the two main components of a precision weed management system, a weed map (location and coverage of weeds) and a user-adjustable prescription map, are effective during critical weed management periods to reduce herbicide use and environmental contamination.