Digital image processing with Artificial Neural Network (ANN) was used to identify some soybean varieties through the form and size of the seeds. The following varieties were analyzed: EMBRAPA 133, EMBRAPA 184, COODETEC 205, COODETEC 206, EMBRAPA 48, SYNGENTA 8350, FEPAGRO 10 and MONSOY 8000 RR, 2005/2006 harvest. The image processing was constituted by the following stages: 1) Image acquisition: the samples of each variety were photographed by photographic camera Coolpix995, Nikon, with resolution of 3.34 megapixels; 2) Pre-processing: an anti-aliasing filter was applied to convert the true-color image to the grayscale intensity image; 3) Segmentation: it was made the seeds edges detection (Method of Prewitt), edge dilation and removal of needless segments; 4) Representation: each seed was represented in the form of a binary matrix 130x130, and 5) Recognition and interpretation: feedforward multiple-layer network was used with three hidden layers. The training of the network was accomplished by backpropagation. The validation of the trained ANN showed that applied processing can be used for identification of the considered soybean varieties.
soybean morphological properties; morphologic analysis of seeds; pattern recognition
