Submitted to: Remote Sensing Applications: Society and Environment
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 2/2/2016
Publication Date: 5/17/2016
Citation: Westbrook, J.K., Eyster, R.S., Yang, C., Suh, C.P. 2016. Airborne multispectral identification of individual cotton plants using consumer-grade cameras. Remote Sensing Applications: Society and Environment. 4:37-43.
Interpretive Summary: Boll weevils can infest isolated cotton plants as well as cotton fields, but finding and removing isolated cotton plants is difficult within large cotton production areas. Although aerial imaging using consumer-grade cameras achieves sufficient spatial resolution to detect isolated (volunteer and regrowth) cotton plants in mixed vegetation, each image pixel represents direct measurement of only one of the three spectral bands (red, green, and blue) and interpolation of the remaining two spectral bands. We present an analytical technique which calculates the median intensity of each spectral band for soil, cotton, other crops, and weeds as the respective spectral identities. Results of this study indicate that consumer-grade cameras can acquire multispectral images (red, blue, green, and infrared bands) from which to detect individual volunteer cotton plants at an early growth stage (before blooming). The results will aid eradication programs in identifying cotton fields, leading to expedited completion and maintenance of boll weevil eradication in the U.S.
Technical Abstract: Although multispectral remote sensing using consumer-grade cameras has successfully identified fields of small cotton plants, improvements to detection sensitivity are needed to identify individual or small clusters of plants. The imaging sensor of consumer-grade cameras are based on a Bayer pattern, which alternates red, green, and blue filters over individual sensor pixels. However, each pixel of the imaging sensor of consumer-grade cameras represents direct measurement of only one of the three spectral bands (red, green, and blue) and interpolation of the remaining two spectral bands. We present an analytical technique in which endmember sets were derived from bimodal histograms of each spectral band for cotton, other vegetation types and soil, and linear spectral unmixing was used to identify individual cotton plants. For some field environments, we achieved significant misclassification rates as low as 0.125 and 0.146 for validation tests of remote sensing identification of volunteer okraleaf cotton plants and volunteer conventional cotton plants, respectively. Results of this study indicate that consumer-grade cameras can acquire multispectral images of sufficient quality to detect individual cotton plants at an early growth stage, which will aid eradication programs in identifying and locating volunteer plants.