Automatic Crop Rows Segmentation for Multispectral Aerial Imagery

The technological advancement in remote sensing and data analysis has made significant improvements in literature due to the increasing availability of datasets and methodologies. Recently, many solutions have been proposed to automatically process and analyze data acquired with Unmanned Aerial Vehicles (UAVs) by implementing algorithms based on Machine Learning (ML) techniques. The most popular are those based on Convolutional Neural Networks (CNNs). In this work, two main methodologies have been developed and compared to perform automatic segmentation of crop rows from multispectral images. The first approach is implemented based on Support Vector Machine (SVM) and the other by means of the U-Net architecture, a state-of-the-art neural network. The two solutions are here tested to assess their classification accuracy and computational load. Despite the heavier computational load, the results showed that U-Net is more precise than SVM solutions.