Employing canopy hyperspectral narrowband data and random forest algorithm to differentiate palmer amaranth from colored cotton

Authors: Fletcher, Reginald; Turley, Rickie

Submitted to: American Journal of Plant Sciences
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 11/25/2017
Publication Date: 11/29/2017
Citation: Fletcher, R.S., Turley, R.B. 2017. Employing canopy hyperspectral narrowband data and random forest algorithm to differentiate palmer amaranth from colored cotton. American Journal of Plant Sciences. 8:3258-3271.

Interpretive Summary: To better implement control strategies for Palmer amaranth invasions in cotton production systems, consultants and producers need tools that can help them differentiate it from cotton. ARS Scientists at Stoneville, MS trained a computer algorithm to differentiate Palmer amaranth plants from cotton plants based on light reflectance measurements of their canopies. The study focused on differentiating Palmer amaranth from cotton near-isogenic lines with bronze, green, and yellow colored leaves. Overall classification accuracies ranged from 77.8% to 88.9%. The highest accuracies were achieved for Palmer amaranth versus cotton yellow classification. Findings support further application of machine learning algorithms and light reflectance properties of plant canopies as tools for Palmer amaranth and cotton discrimination with potential application of this technology in site-specific weed management programs.

Technical Abstract: Palmer amaranth (Amaranthus palmeri S. Wats.) invasion negatively impacts cotton (Gossypium hirsutum L.) production systems throughout the United States. The objective of this study was to evaluate canopy hyperspectral narrowband data as input into the random forest machine learning algorithm to distinguish Palmer amaranth from cotton. The study focused on differentiating the Palmer amaranth from cotton near-isogenic lines with bronze, green, and yellow leaves. A spectroradiomter was used to acquire hyperspectral reflectance measurements of Palmer amaranth and cotton canopies for two separate dates, December 12, 2016, and May 14, 2017. Data were collected from plants that were grown in a greenhouse. The spectral data were aggregated to twenty-four hyperspectral narrowbands proposed for study of vegetation and agriculture crops. Those bands were tested by the conditional inference version of random forest (cforest) to differentiate the Palmer amaranth from cotton. Classifications were binary: Palmer amaranth and cotton bronze, Palmer amaranth and cotton green, and Palmer amaranth and cotton yellow. Classification accuracies were verified with overall, user’s, and producer’s accuracy. For the two dates combined, overall accuracy ranged from 77.8% to 88.9%. The highest accuracies were observed for the Palmer amaranth versus the cotton yellow classification. Producer’s and user’s accuracies range was 66.7% to 94.4%. Errors were predominately attributed to cotton being misclassified as Palmer amaranth. The overall results indicated that cforest has moderate to strong potential for differentiating Palmer amaranth from cotton when it used hyperspectral narrowbands known to be useful for vegetation and agricultural surveys as input variables. This research further supports using hyperspectral narrowband data and cforest as decision support tools in cotton production systems.