Quantitative and Qualitative Analysis of Agricultural Fields Based on Aerial Multispectral Images Using Neural...

Area of Science:

• Agricultural Science
• Computer Vision
• Remote Sensing

Background:

• Efficient agricultural management requires comprehensive crop analysis.
• Unmanned Aerial Vehicles (UAVs) offer advanced capabilities for aerial data acquisition.
• Integrating qualitative and quantitative analysis methods enhances agricultural decision-making.

Purpose of the Study:

• To present an integrated system for comprehensive crop analysis using UAVs.
• To combine qualitative (vegetation indices) and quantitative (object detection) analyses for efficient agricultural management.
• To develop a system leveraging Detectron2 for object detection and segmentation in multispectral aerial imagery.

Main Methods:

• Utilized a convolutional neural network (Detectron2) trained on COCO-formatted data for object detection and segmentation.
• Developed a system with frontend (user interaction, annotation) and backend (image processing, project management) components.
• Implemented Normalized Difference Vegetation Index (NDVI) and Optimized Soil Adjusted Vegetation Index (OSAVI) for qualitative analysis and object detection for quantitative analysis.

Main Results:

• The system successfully performs both qualitative (NDVI, OSAVI) and quantitative (object detection) crop analyses.
• The Detectron2 model demonstrated robust performance in detecting objects, including small objects in aerial images.
• The prediction quality was assessed using the Average Precision (AP) metric, indicating reliable model performance.

Conclusions:

• The integrated UAV system provides a powerful tool for detailed crop analysis in agriculture.
• The combination of advanced AI models and vegetation indices enables efficient and accurate agricultural management.
• The system's ability to detect and quantify specific crop elements, like young lettuce, supports precision farming practices.