6533b823fe1ef96bd127e849

RESEARCH PRODUCT

12 - Feasability study of a wheatears counting system per vision

subject

description

Working on a feasibility study of wheatears counting, a colour component texture's analysis method was developed. The agronomic goal is yield prediction before harvest evaluating mean number of wheatears per squared meter according to the field variation knowledge. To this counting system, we evaluate six textural parameters (two statistical parameters and four Haralick features from co-occurrence matrix) on the main colour systems and vegetation indices used in agronomic applications. A new hybrid system provides a representation of wheatears' pictures taken under natural conditions with a better extraction of wheat. A method based on distances measurements (Euclidian, Mahalanobis) allows to extract wheatears with few errors corrected by mathematical morphology. Although we encounter difficulties from light intensity's variation and high entropy in the scene (ears' covering and shadows), results allow to extract disturbed wheatears and last recent images give an higher accuracy in segmentation.