Detection of Abnormal Regions in a Field by Semantic Segmentation Using Remote Sensing Data

Abstract

In recent years, “smart agriculture,” which introduces ICT into agriculture to improve the efficiency, automation, and productivity of agricultural work, is attracting attention. For advanced smart agriculture, this paper proposes a semantic segmentation model to detect and classify abnormal regions in a field using satellite or aerial images. The performance evaluation of the proposed model is conducted for the Agriculture-Vision Challenge Dataset, a dataset of aerial images of fields and abnormal regions. The results show that the proposed model can detect abnormal regions with higher performance than conventional deep learning models.