LineFit. Artificial Intelligence for sugarcane planting failures.

Project description

Planting failures produce economic losses of 10% to 30% of sugarcane production on more than 26 million hectares worldwide. This project uses Artificial Intelligence to safely detect planting failures in sugarcane crops.

Data processing

A UNET-type network has been trained with two million plants for the recognition of sugarcane plants in images taken with drones. Additionally, the divide and conquer strategy has been used for a safe reconstruction of the crop lines, reducing the crop line detection problem to squares of 512 pixels on a side and then integrating to obtain the crop lines in the entire farm, as is shown in the figure above. The application was developed on Windows for farm and data management, and the Artificial Intelligence runs on a Linux machine.

Results

The figure above shows the crop lines of two fields of sugarcane of 49.2 ha and 26 ha, with a difference of 19.6% in yield expressed as plants per ha, the estimated total losses for the farm with the lowest production is 10,500 USD.

This next image shows a close-up to observe the crop lines drawn on the plants. To obtain the crop lines, each plant has been replaced by a point and the points are processed.

This third figure shows diagrams with the percentages of plants of two farms as a function of the spacing between the plants. The behavior of these graphs show patterns that are originated in different agricultural practices.