Hyperspectral Imaging and Machine Learning to Classify Herbicide-Resistant and -Susceptible Weeds in Field Crops

The increasing prevalence of herbicide-resistant weeds in western Canada continues to reduce crop yields, increase herbicide expenditures, and threaten farm profitability. Knowledge-based integrated weed management strategies are necessary to mitigate further evolution and spread of herbicide-resistant weeds. Identification and differentiation of herbicide-resistant and susceptible weeds can inform and appropriately target these strategies. However, traditional methods to identify herbicide-resistant weeds are time-consuming, labor intensive, and most require mature weed seeds. To improve the efficiency of herbicide-resistant weed identification, we utilized hyperspectral imaging integrated with machine learning to discriminate herbicide-resistant and susceptible weed biotypes. Greenhouse and field experiments were performed at Lethbridge and Lacombe Research and Development Centres (Alberta, Canada) to evaluate resistance in kochia [Bassia scoparia (L.) A.J. Scott] to glyphosate (Group 9) and fluroxypyr (Group 4), and in wild oat (Avena fatua L.) to fenoxaprop (Group 1) and imazamox (Group 2). Hyperspectral data (400–1000 nm, 204 bands) were classified using Quadratic Discriminant Analysis. In kochia, accuracies ranged from 70.9–90.4% for glyphosate and 72.9–92.7% for fluroxypyr when models were evaluated before and after herbicide treatment. In contrast, wild oat resistance was assessed after treatment only, with Quadratic Discriminant Analysis achieving 87.9% accuracy for fenoxaprop and 75.1% for imazamox. These results demonstrate that hyperspectral can detect herbicide-resistant-related spectral differences before herbicide treatment and visible herbicide symptomology, allowing for earlier and more precise intervention. This study highlights the potential of hyperspectral imaging combined with machine learning as a high-throughput, non-invasive approach for detecting herbicide-resistant weeds. Overall, the findings support the integration of advanced digital technologies into sustainable weed management practices.