Related Experiment Video
Updated: Jan 16, 2026
08:16
Collecting and Processing Drone-based Remotely Sensed Data for Use in Forest Recovery Monitoring
Published on: October 24, 2025
488
Deep learning model BiFPN-YOLOv8m for tree counting in mango orchards using satellite remote sensing data
Lalit Birla1, Anshu Bharadwaj2, Rajni Jain3
1The Graduate School, ICAR-Indian Agricultural Research Institute, New Delhi, 110012, India. lbirla64@gmail.com.
Scientific Reports
|September 30, 2025
Summary
This study introduces a novel deep learning model, Bi-directional Feature Pyramid Network (BiFPN)-YOLOv8m, for accurate mango tree counting using satellite imagery. The advanced model significantly improves upon existing methods for agricultural planning and yield forecasting.
Area of Science:
- Agricultural Science
- Remote Sensing
- Computer Vision
Background:
- Mango production is economically vital for India, necessitating accurate tree inventory for yield assessment.
- Traditional tree counting methods are labor-intensive, costly, and prone to errors.
- Satellite Remote Sensing offers a scalable solution for ecological and agricultural parameter estimation.
Purpose of the Study:
- To develop and evaluate a deep learning model for high-resolution, image-based mango tree counting.
- To compare the performance of the proposed Bi-directional Feature Pyramid Network (BiFPN)-YOLOv8m against various YOLOv8 variants and other state-of-the-art methods.
- To enhance the accuracy and efficiency of mango tree inventory for improved agricultural planning and crop yield forecasting.
Main Methods:
- Utilized a dataset of 1700 training and 300 testing high-resolution satellite images of mango orchards.
- Implemented and evaluated multiple object detection models including YOLOv8 variants (YOLOv8n, YOLOv8s, YOLOv8m, YOLOv8l, YOLOv8x), YOLOv9, YOLOv10, and the proposed BiFPN-YOLOv8m.
- Focused on assessing models based on computational efficiency, accuracy, and speed for tree counting.
Main Results:
- The proposed BiFPN-YOLOv8m model demonstrated superior performance in accurately locating and counting mango trees.
- The model achieved state-of-the-art results, outperforming other evaluated methods, even in challenging environmental conditions.
- Experimental findings highlighted the model's effectiveness in separating and counting trees within orchards.
Conclusions:
- Deep learning, specifically the BiFPN-YOLOv8m model, provides a highly effective and efficient solution for image-based mango tree counting.
- The developed approach significantly overcomes the limitations of conventional methods, offering improved accuracy for ecological and agricultural applications.
- This technology holds substantial promise for advancing precision agriculture, environmental monitoring, and crop yield prediction in mango cultivation.
