Automatic Generation of Highly Accurate Parameter-optimal Segmentation Models in Precision Agriculture

In this paper, we demonstrate how an existing AutoML tool can be enhanced to generate highly accurate yet edge-friendly segmentation models for precision agriculture. Currently, only a few AutoML tools are capable of producing TinyML models for dense computer vision tasks, such as segmentation and object detection. Furthermore, creating accurate yet parameter-efficient segmentation models is particularly challenging in precision agriculture applications, such as crop and weed segmentation. We show that such models can be achieved by invoking only two AutoML API functions and tuning just four hyperparameters. A video of the tuning process is available at: https://youtu.be/L3MyXkVTzCc.