Co-occurrence matrix with neural network classifier for weed species classification: a comparison between direct application of co-occurrence matrix (GLCM) and haralick features as inputs

Gray level Co occurrence matrix (GLCM) texture analysis has been aggressively researched for decade for multiple applications. Co occurrence matrix retains the spatial and frequency information of the image while compresses the image into a fraction of size enabling the application of classifier engines for analysis. Haralick features are secondary features derived from GLCM. There have been countless research work done on weed classification using Haralick features outweighing the application of direct feeding of co occurrence matrix for training classifiers. Images are aquired with slight varying distances and angles to test the robustness of classifier and pre-processed using excessive Green Index method before fed into ANN (Artificial Neural Network) for training and evaluation. In this paper, we found that direct application of GLCM a column out performs the haralick feature method due to the unregulated lighting