Kernelytics: Multispectral Drone Imagery and Deep Learning for Early Corn Assessment

Rising pressures from climate change and population growth have prompted farmers to explore modern technologies, such as drones and artificial intelligence to optimize conventional and manual crop monitoring practices. We present a new algorithm, Kernelytics, to efficiently assess corn fields early in the season, producing more optimal yields, allocating resources more efficiently, and identifying corn health issues earlier in the season. The proposed experimental procedure utilizes a low-cost drone to collect aerial images of the Visible Light, Red, Green, Blue, Near-Infrared (NIR), and Red Edge bands. After preprocessing the collected data, Kernelytics involves three machine learning-based assessments: object detection for corn stand count, semantic segmentation for plant and weed area, and instance segmentation for plant health. The research additionally determines the optimal model type, image size, and band combinations for each task. Keywords-multispectral imagery, remote sensing, computer vision, precision agriculture