Reducing weeds is a way of enhancing crops per unit area which has led to irregular application of herbicides. Many studies have shown that just 20% to 76% of fields have been devoted to weed. Use of site- specific strategies for weed management can reduce application of herbicides. In this research in order to reduce application of herbicide in corn fields, support vector machine (SVM) was designed based on machine vision system that used geometrical features of shrubs. In order to identify shrubs from background variations in algorithm of image segmentation called Prixelwise were performed. Then for shrub classification, algorithm of SVM classification was created using derivation of seven geometrical features from 100 laboratory images. Identification capacity of algorithm was determined to be 81% based on cross validation assay. In field assay 100 images were taken manually or automatically from corn raws in two different days. SVM could classify weeds at an accuracy of 93% in time of 1.16 s and 65% in time of 2.16 based on images taken. The reason for decreased classification accuracy of automatic way can be traced back to image quality and light undesirable conditions.
