Event Details

When:

Time: 8:30–9:30 am

Location: Room 2C61 AGRI

DETECTING KOCHIA AT THE FIELD SCALE USING MULTI-TEMPORAL UAV IMAGERY

Abstract

Kochia (Bassia scoparia) is one of the most problematic agronomic weeds in Canada. This annual broadleaf weed typically thrives in saline and acidic areas and in regions with moderate precipitation. Currently, the control of kochia in crops has become problematic owing to its resistance to multiple herbicides, including those in groups 2, 4, 9, and 14. Consequently, controlling kochia patches in fields is challenging, and mapping them can help manage kochiainfested areas effectively. The main aim of this study was to explore the spectral characteristics of kochia and the potential use of UAV imagery (Quantix mapper) for mapping kochia in crop fields using a machine learning algorithm. Data were collected from 12 lentil and 4 wheat fields. Data collection was initiated during the mid-growing season, when kochia was visible. Kochia classification in lentil and wheat crops was performed using the random forest (RF) algorithm. The model was trained using multispectral bands and vegetation indices. To evaluate the performance of the RF, crop-based and combined models were created: lentil, wheat, and lentil– wheat models. Prior to performing the classification, understanding the spectral differences between kochia and crops (wheat and lentil) was crucial for successfully training the machine learning model. UAV temporal data gathered during the mid-growing season were used to quantify kochia from the lentil crop using vegetative indices. It was observed that kochia and crops had spectral differences over the last two dekads, with the greatest difference during the last dekad before crop harvest. Three models were trained and evaluated using the last two dekads. The trained models achieved 94.8% or higher accuracy when tested on the held-out lentil and wheat datasets. When the spatial classification of kochia patches was performed on selected dates within the last dekad using the crop-based lentil and wheat models, classification accuracies of 97.5% and 100% were achieved for lentil and wheat, respectively. The combined lentil-wheat model classified kochia patches with 97.3% accuracy in lentil fields and 99.5% in wheat fields. This finding supports accurate spatial classification of kochia patches in lentil and wheat crops, facilitating in creating effective kochia control strategies.

Via Zoom https://usask-ca.zoom.us/j/92779144503?pwd=6ygF4x51o0aemg19h9xfYdb1GLdTUF.1

Everyone is welcome to attend

Contact

Debbie Banks