Weeds identification in corn fields based on image processing techniques and artificial neural networks

Weeds normally grow as patches and spatially distributed in field. Patch spraying to control weeds has advantages such as cost reduction, herbicide saving and reduction of environmental pollution. Machine vision system should obtain and process digital images to make control decisions. Proper identification and classification of weeds are the key steps to make control decisions and use of any spraying operation performed. In this study, a robust method based on image processing and computational intelligence was developed for segmentation from other parts of image and classification of weeds. Large crabgrass, common lamb’s quarter, velvetleaf, common barnyard grass, European black nightshade, red-rooted pigweed and European heliotrope were the weeds in the experiment. Results showed that this algorithm was precisely separated weeds from the soil. In the next step, the feature vector, which includes shape features and color features, was composed. Finally, classification of seven classes of weeds was carried out by artificial neural network (ANN). Among different ANN structures, the most optimum classifier was the 43-15-15-7 topology with accuracy 88/71 %. The results of this research indicate that the proposed system has the ability to accurately detection of weeds.