Identificação de linhas de plantio em imagens aéreas de lavouras

Abstract

Finding rows on aerial images of row-planting crops is an important problem for many agricultural needs, such as cultivation and crop planning, production estimation and plant counting. This paper covers the problem of given an RGB aerial image of a crop, classify and output the location of the crop rows. Also, as the result of the proposed algorithm, rows are output as sequence of points, which can be used for further results. There are few related works to this problem, mainly because this paper uses aerial images for the field images and because it looks into the problem of planting gaps, which is very common in farming applications. The problem was tackled by finding individual plants, then finding lines from plant locations and nearby general directions, and then filtering and merging the lines to solve the problems of weed species and planting gaps. Results were evaluated by the lines lengths for true positives, false positives and false negatives for evaluation with the F-1 statistical test.