Vegetation cover is a widely used parameter in weed research and several methods are used to asses it through digital images, ranging from simple methods (visual estimate) to more complex ones based on image analysis processing. By using photographs of a wheat experiment we first compared three simple subjective methods for the estimate of vegetation cover through digital image analysis (visual estimate, point quadrate and supervised classification) to understand which of them gave results which were less dependent on operator's subjective value assignment. We did not find a significant difference between operators in case of visual estimate (P=0.784). We then tested the relative performance of two completely automated image analysis techniques based on image transformations (IV 1 V 2 and Principal Components) frequently found in the scientific literature. IV 1 V 2 gave a better performance with respect to Principal Components. There was a highly significant correlation (p=0.947) between IV1V2 and visual estimate, i.e. the two methods of the respective categories which gave the best results. Although being very simple and potentially influenced by operator's skills,visual estimate always performed better in terms of repeatability. We conclude that visual estimate carried out by experienced observers could be regarded as a standard tool in weed research.
COMPARISON BETWEEN DIGITAL ANALYSIS METHODS FOR THE ESTIMATION OF VEGETATION COVER IN WEED RESEARCH
BOCCI, Gionata;BIGONGIALI, Federica;BARBERI, Paolo;MOONEN, Anna Camilla
2011-01-01
Abstract
Vegetation cover is a widely used parameter in weed research and several methods are used to asses it through digital images, ranging from simple methods (visual estimate) to more complex ones based on image analysis processing. By using photographs of a wheat experiment we first compared three simple subjective methods for the estimate of vegetation cover through digital image analysis (visual estimate, point quadrate and supervised classification) to understand which of them gave results which were less dependent on operator's subjective value assignment. We did not find a significant difference between operators in case of visual estimate (P=0.784). We then tested the relative performance of two completely automated image analysis techniques based on image transformations (IV 1 V 2 and Principal Components) frequently found in the scientific literature. IV 1 V 2 gave a better performance with respect to Principal Components. There was a highly significant correlation (p=0.947) between IV1V2 and visual estimate, i.e. the two methods of the respective categories which gave the best results. Although being very simple and potentially influenced by operator's skills,visual estimate always performed better in terms of repeatability. We conclude that visual estimate carried out by experienced observers could be regarded as a standard tool in weed research.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
