End-to-End Deep Learning Models for Gap Identification in Maize Fields

Abstract

We propose an approach to jointly count plants and de- tect gaps in maize fields using end-to-end deep-learning

models. Unlike previous efforts that focused solely on plant detection, our methodology also integrates the task of gap identification, offering a holistic view of the state of the agricultural field. Moreover, we consider different data

sources in our experiments and explore the benefits of us- ing multispectral over RGB images, which are commonly

used in the industry. The findings suggest that multi-task learning on multispectral images significantly outperforms other model configurations, demonstrating the potential of the proposed approach for precision agriculture.