Then, machine learning (CNN) is applied to predict the likelihood of soil-borne illnesses being present based on conditions in the soil. Before this, preprocessing is required using GIS to transform the images to account for atmospheric changes, such as light going through the atmosphere and back scatter or noise.

The data needs to be corrected to take these factors into account, and the GIS format allows for easy cleaning of data later on.

When it comes to the machine learning classification process, pixels can be changed using algorithms, which can then be applied to RGB and multispectral images.
Pictures taken by drones can be analyzed using RGB to detect changes in plant health, which can indicate the presence of soil-borne pathogens. By applying machine learning algorithms to the image data, it’s possible to classify areas of the soil based on the presence or absence of soil-borne illnesses.

A point cloud is generated by taking many pictures of a single area with overlap, resulting in a 90% overlap of multiple images stacked upon each other. This allows the conversion of a 2D image into a 3D image. Point clouds are particularly important for soil and vegetation mapping, especially for areas with smaller crops.

They are essential for identifying where crops are, as the size of the crops may be smaller than other crops in the area.