Weed species discrimination based on SIMCA analysis of plant canopy spectral data

Adoption of a site-specific weed management system (SSWMS) can contribute to sustainable agriculture. Weed classification is a crucial step in SSWMS that could lead to saving herbicides by preventing repeated chemical applications. In this study, the feasibility of visible and near infrared spectroscopy to discriminate three problematic weeds was evaluated. A greenhouse experiment was conducted to classify three common weed species: water-hemp (Amaranthus rudis), kochia (Kochia scoparia), and lamb's-quarters (Chenopodium album). Soft independent modelling of class analogy (SIMCA) method was used to classify these weed species based on canopy spectral reflectance. Five different pre-processing methods were evaluated to remove the irrelevant information from spectral reflectance. Analysis of data indicated that the second derivative pre-processing method applied to NIR (920-2500 nm) spectra was the best to discriminate three weed species with 100% accuracy for 63 test samples. The SIMCA model on NIR wavebands exhibited the highest discrimination power ratio. The results showed the model distance value for most developed classes in NIR range was more than three, which indicated its superior ability to discriminate weed species with low risk of misclassification. Furthermore, the discrimination power of different wavelengths obtained from the best models indicated that 640, 676, and 730 nm from the red and red-edge region, and 1078, 1435, 1490, and 1615 nm from the NIR region were the best wavelengths for weed discrimination.