GUIDANCE PARAMETER DETERMINATION USING ARTIFICIAL NEURAL NETWORK CLASSIFIER

Summary: The trends to change from traditional agricultural practices to modern site-specific crop management have been demanded more automated machinery. Thus, an algorithm has been developed for a vision-based guidance system to steer a tractor in the field of row crops. The algorithm treated the extraction of guidance parameters from the images as a pose recognition problem. The first algorithm task was to reduce the dimension of the data from the image size to the projection dimension of the image onto the eigenspaces built by principal component analysis. Then, the algorithm used a pose classification process using the projection points to output the tractor guidance parameters. The use of an artificial neural network (ANN) was evaluated as the classification task tool. The ANN was trained with the projection points of a set of images with known guidance parameters. The ANN performance was compared with classification using Euclidean distance and presented similar results of 0.5 degree and 4.7 cm of the absolute errors between the actual and the estimated heading angle and offset, respectively.