Identification of Deficiencies of Iron and Manganese in Leaves of Bean Using a Classsifier.

Given the importance of common bean (Phaseolus vulgaris L.) as a source of proteins within the basic diet of human being, as well as the social and economic importance of the crop, solutions are required to the nutritional problems associated with iron (Fe) and manganese (Mn) deficiencies. These deficiencies show similar characteristics in the early stages, so they can be confused. It was proposed the creation of classifiers using artificial neural networks to identify deficiencies of Fe and Mn from digital images of samples of common bean. Eight color and four textural features were obtained and used as inputs; these features were calculated from a relative frequencies co-ocurrence matrix of paired pixels with variations of distance of 1, 5 and 10 pixels. Sixteen classes were defined from the combinations of Fe and Mn on the range of 0 to 200% in reference to the Steiner solution. Different combinations of inputs were tested, using 16, 10 and 8 classes when some intermediate classes were taken out. As the number of classes decreased or the distance between pixels increased, the global classification percentage improved. The best classifiers were obtained using eight classes and 10 pixels to generate the relative frequencies co-ocurrence matrix (62.3% overall correct global classification on average). The classifier with the best classification percentage (64.4%) was the one which used as inputs three channels of the RGB color space and four textural features (second angular moment, inertia, entropy and local homogenity). When only color or textural features were used, low percentages of correct overall classification were obtained. Therefore, both color and textural features could constitute a tool for detection of deficiencies of micronutrients in bean crop at early growth stages.