Classification of corn plants and weed based on characteristics of color and texture using methods of segmentation Otsu and PCA

Abstract

The corn crop is very important in Mexico. Corn is fertilized manually or with machinery. When fertilization is manual, it consists of depositing fertilizer to each corn plant. Whereas machine fertilization, involve of dropping fertilizer along the furrow continuously. Manual fertilization is effective, but it is expensive and time-consuming. Machine fertilization can be inefficient, because fertilizer is deposited in the weeds or where there is no corn plant. When the fertilizer is not absorbed by the plant, it can damage the aquifers. This project presents algorithms to classify corn plants and weeds, hoping to contribute to automated fertilization or identified weeds to apply herbicide or eliminate. We took hundreds of pictures of corn plants and weeds in corn crops. The images were segmented using the Otsu method. As well as, the images were processed with the PCA algorithm. We apply classification algorithms such as Naive Bayes, Random Forest, SVM, KNN and Backpropagation. We also apply a convolutional neural network (CNN). We finally got 99.97% as the best result with the Backpropagation classifier.