Image Segmentation between Crop and Weed using Hyperspectral Imaging for Weed Detection in Soybean Field

Abstract

The goal of this study was image segmentation between crop and weed using hyperspectral imaging for weed detection. This technique is useful for selective weeding using spot spraying or a robotic system. A hyperspectral image consists of a large number of pixel spectra. Therefore, image segmentation was executed pixel by pixel. After each pixel spectrum was extracted from the image, it was identified as soil or plant. If the pixel spectrum was identified as plant, it was categorized as crop or weed. For the pixel discriminant model between soil and plant, simple NDVI thresholding was employed. The pixel discriminant model between crop and weed consisted of normalization, generation of explanatory variables and discrimination, and four types of models were developed and validated. Finally, segmented images between crop and weed were generated from the hyperspectral images by applying these models. As the validation results, pixel discrimination between soil and plant was performed with 99.9%. Also, most pixel discriminant models between crop and weed had a success rate of more than 90%. As a result of image segmentation between crop and weed, most hyperspectral images were segmented correctly. This study demonstrated the possibility of weed detection using hyperspectral imaging.