Remote Sensing-Based Agricultural Damage Assessment: A Semantic Segmentation Method for Analyzing Corn Field Changes

International Conference on Communication Networks and Machine Learning (CNML), pp. 400–405

Abstract

The increasing impact of natural disasters on agricultural production highlights the vital need for accurate estimation of changes in crop cultivation areas as pertains to agricultural recovery and formulation of food security strategies. In this study, the authors apply an integrated approach of remote sensing data and semantic segmentation technology by taking the June 2023 event in Illinois, termed the derecho event, as a case study. DeepCBAMNet is applied to analyze the pre-and post-disaster remote sensing images, yielding an overall model accuracy of 93.39% and a mean Intersection over Union (mIoU) of 81.88% on the training dataset. The study reports an area affected by the disaster of 69,352.01 acres with a 91.98% accuracy rate when compared with that reported by the United States Department of Agriculture (75,401.60 acres). The methodology is a key advancement in the assessment of agricultural disasters, as it gives a new technical approach to the rapid and accurate assessment of disaster impact, thus allowing government authorities to put in place more productive post-disaster recovery strategies and agricultural policies.

Authors 3

1. Meng Wu

   Sanda University

   Affiliation as printed

   School of Information Science and Technology, Sanda University, Shanghai, China

2. Zeyu Li Aachen Computer Science and Artificial Intelligence Department

   Leiden University

   Affiliation as printed

   Computer Science and Artificial Intelligence Department, Leiden University, Leiden, Holland

3. Yi Wei

   Sanda University

   Affiliation as printed

   School of Information Science and Technology, Sanda University, Shanghai, China