Identifying hair fescue in wild blueberry fields using drone images for precise application of granular herbicide

Abstract

Hair fescue is a common perennial grass weed in wild blueberry fields in Nova Scotia, Canada and reduces yield if not adequately controlled. Hair fescue is typically managed with a broadcast propyzamide application which is costly for growers. Current research on the management of hair fescue in wild blueberry fields focuses on the development of a precision granular herbicide applicator with individual nozzle control. The integration of aerial images could benefit current work by providing a large-scale perspective on hair fescue location. This research suggests a new strategy to identify infield areas with hair fescue infestation, generate prescription maps, and integrate them into a widely available navigation system of heavy farming equipment to improve the processing workflow of hair fescue detection on a larger scale. This method proposes the integration of multi-resolution segmentation on multispectral (Blue, Green, Red, Red Edge, Near-Infrared) drone images to group hair fescue plants into segments with similar spectral characteristics. The best performing segmentation was classified and tested on wild blueberry fields in Nova Scotia. Results were assessed by comparing classified boundaries to reference data collected with high resolution RGB images. Findings from hair fescue and wild blueberry spectral profiles showed that hair fescue could be identified using blue, green, and red bands and showed no advantages over multispectral cameras. Results from the classification of wild blueberry and hair fescue showed high Kappa coefficient values of 0.90 and 0.94 and overall accuracy of 95 and 97% were achieved for both fields surveyed. Results also demonstrated an average potential cost saving of $CAD1246.43 ha-1 when compared to a broadcast application of granular dichlobenil. This work represents the first attempt to integrate aerial imagery and a precision granular herbicide applicator to manage important weeds in wild blueberry fields.

Keywords

• Hair fescue
• Image segmentation
• Drone
• Multispectral imaging
• Image classification