Application of support vector machine for identifying single corn/weed seedling in fields using shape parameters

Abstract

This study is conducted to present an application of support vector machine (SVM) method and image processing techniques for corn/weed seedlings in the fields. The original images obtained from the field are used to be preprocessed by a space transform and image processing techniques at first. Corn seedlings or small weeds are segmented by H channel using OTSU method. We found that H channel is better to reduce the effects of illumination changes. Four shape parameters extracted from the objective are used in the recognition procedure. SVM and back-propagation neural network classifiers are employed to identify single corn/weed seedling. Experimental results show that SVM classifier gives a better classification effect. SVM method with RBF kernel function achieves the highest detection accuracy of 96.5%. Using the same testing set, back-propagation neural network classifier only gives a recognition rate 83.2%.