Regresar

Crop/weed discrimination in potato fields using computer vision techniques

Abstract:

This study presents an adapted method for crop/weed discrimination in images captured in potato fields during initial stages of growth, applying a criterion of similarity based on Mahalanobis distance. Weed detection is useful in precision agriculture to quantify and apply site-specific treatments. The images were obtained in perspective projection with a camera installed on board a tractor. The quality of the image is affected by uncontrolled lighting conditions and different plant sizes. The proposed method consists of three phases: Segmentation, training and testing. The main contribution is the ability to discriminate crop and weeds located between the crop lines and within the same furrow. The performance of the method was compared quantitatively with two existing strategies, achieving an accuracy of 89,65% with processing times less than 330 ms.