Ran N Lati1, Sagi Filin, Radi Aly, Tal Lande, Ilan Levin, Hanan Eizenberg. 1. Department of Weed Research and Plant Pathology, Agricultural Research Organization, Newe Ya'ar Research Center, Israel; Mapping and Geo-Information Engineering, Technion-Israel Institute of Technology, Haifa, Israel.
Abstract
BACKGROUND: Weed/crop classification is considered the main problem in developing precise weed-management methodologies, because both crops and weeds share similar hues. Great effort has been invested in the development of classification models, most based on expensive sensors and complicated algorithms. However, satisfactory results are not consistently obtained due to imaging conditions in the field. RESULTS: We report on an innovative approach that combines advances in genetic engineering and robust image-processing methods to detect weeds and distinguish them from crop plants by manipulating the crop's leaf color. We demonstrate this on genetically modified tomato (germplasm AN-113) which expresses a purple leaf color. An autonomous weed/crop classification is performed using an invariant-hue transformation that is applied to images acquired by a standard consumer camera (visible wavelength) and handles variations in illumination intensities. CONCLUSION: The integration of these methodologies is simple and effective, and classification results were accurate and stable under a wide range of imaging conditions. Using this approach, we simplify the most complicated stage in image-based weed/crop classification models.
BACKGROUND: Weed/crop classification is considered the main problem in developing precise weed-management methodologies, because both crops and weeds share similar hues. Great effort has been invested in the development of classification models, most based on expensive sensors and complicated algorithms. However, satisfactory results are not consistently obtained due to imaging conditions in the field. RESULTS: We report on an innovative approach that combines advances in genetic engineering and robust image-processing methods to detect weeds and distinguish them from crop plants by manipulating the crop's leaf color. We demonstrate this on genetically modified tomato (germplasm AN-113) which expresses a purple leaf color. An autonomous weed/crop classification is performed using an invariant-hue transformation that is applied to images acquired by a standard consumer camera (visible wavelength) and handles variations in illumination intensities. CONCLUSION: The integration of these methodologies is simple and effective, and classification results were accurate and stable under a wide range of imaging conditions. Using this approach, we simplify the most complicated stage in image-based weed/crop classification models.