Proceedings

Find matching any: Reset

Add filter to result:

Ground Level Hyperspectral Imagery For Weeds Detection In Wheat Fields

¹D. J. Bonfil, ²U. Shapira, ²A. Karnieli, ²I. Herrmann, ²S. Kinast

1. ARO, Gilat Research Center
2. Ben-Gurion University of the Negev

Weeds are a severe pest in agriculture resulting in extensive yield loss. Applying precise weed control has economical as well as environmental benefits. Combining remote sensing tools and techniques with the concept of precision agriculture has the potential to automatically locate and identify weeds in order to allow precise control. The objective of the current work is to detect annual grasses and broadleaf weeds among cereal as well as broadleaf crops by implementing field spectroscopy tools. The Spectral Camera HS (Specim) with 1600 pixel per line and 849 bands in the range of 400-1000 nm was selected to obtain ground level images in wheat and potato fields. For each image, fractional coverage (FC) assessment was implemented for five classes: wheat; potato; soil; grass weed (GW); and broadleaf weed (BLW). The images were radiometrically corrected, transformed to relative reflectance values, and spectrally resampled to 91 continuous bands in the range of 400-850 nm. Spectra from 21 images, all together more than 1800 pure pixels, of the five classes were obtained. 28 classification models were calibrated and validated by Partial Least Square Discriminant Analysis (PLSDA) method. The Variable Importance in Projection (VIP) analysis has shown that the red-edge region is most or second most important spectral region in the majority of the cases. 14 models were chosen to classify each of the images. For each model and class the classification results were correlated with the FC field assessment resulting in R² values reaching 0.84 and 0.77 (in different models) for BLW and GW, respectively. It can be concluded that BLW detection is achieved with higher accuracy than GW and that the red-edge region is highly important for weed detection.

Keyword: Remote Sensing, Hyper spectral, Vegetation, Classification

D. J. Bonfil U. Shapira A. Karnieli I. Herrmann S. Kinast Remote Sensing Applications in Precision Agriculture Poster 2012

Download paper