Increased efficiency of selection for leaf area in kale using artificial neural networks

ABSTRACT

The estimation of leaf area in kale is important because direct measurements are difficult and inaccurate, due to the leaf size, the irregularity of the leaf surface of some genotypes, the need for expensive equipment and intensive labor. The objective was to verify the efficiency of artificial neural networks to estimate the leaf area and verify the efficiency of the use of the estimated area in the selection process compared with the observed area. The experiment was conducted in a randomized block design with three replications, 22 accesses and four plants per plot. Multilayer perceptrons were developed using 50 leaves per access, 70% designed for training, 15% for cross-validation (early-stop) and 15% for testing. 39 perceptron multilayer network settings were tested. The RNAs were efficient to estimate leaf area from the length and width of the leaf blade. The leaf area estimated by the RNA is indicated for the selection of plants due to its easily access and due to be a non-destructive method, having high phenotypic and genetic correlation with leaf area observed and higher heritability.

Keywords:
Brassica oleracea var. acephala; multilayer perceptron; indirect selection; genetic parameters.