Classification of weed in soybean crops using unmanned aerial vehicle images

Soybeans have been Brazil's main agricultural commodity, contributing substantially to the country's trade balance. However, their production and productivity costs are affected by weeds, diseases and pests. This paper proposes a computer vision system to monitor weeds in soybean fields using images captured by a UAV. The proposed system adopts the SLIC superpixels segmentation method to detect the plants in the images and visual attributes to describe the characteristics of the physical properties of the leaf, such as color, gradient, texture and shape. Our methodology evaluated the performance of three classifiers (kNN, Randon Forest and SVM) for images captured at a height of 3 meters. The best results were obtained by the SVM classification algorithm with accuracy of 91.34%. However, the results do not yet indicate that our approach can support experts and farmers in weed monitoring in soybean crops, requiring more images and experiments.