Related Experiment Video
Updated: May 24, 2026
08:47
Computer Vision-Based Biomass Estimation for Invasive Plants
Published on: February 9, 2024
Attention-enhanced multi-task learning for binary segmentation and fine-grained aquatic plant classification in UAV
Ashifur Rahman1,2, M M Mahbubul Syeed3,4, Razib Hayat Khan1,2
1Department of Computer Science and Engineering, Independent University, Bangladesh, Dhaka, 1229, Bangladesh.
Scientific Reports
|May 22, 2026
Summary
This study introduces an attention-enhanced multi-task learning framework for simultaneous aquatic vegetation segmentation and species classification from drone imagery. The model achieves high accuracy and efficiency, aiding biodiversity monitoring and ecosystem management.
Area of Science:
- Environmental Monitoring
- Remote Sensing
- Computer Vision
Background:
- Accurate aquatic vegetation monitoring via unmanned aerial vehicles (UAVs) is hindered by complex backgrounds and data limitations.
- Existing methods often treat segmentation and classification separately, limiting integrated species-level analysis.
Purpose of the Study:
- To develop an attention-enhanced multi-task learning framework for simultaneous binary segmentation and 14-class species classification of aquatic vegetation.
- To enable unified structural and semantic understanding of aquatic ecosystems from UAV imagery.
Main Methods:
- A shared encoder with attention-guided skip connections and joint optimization strategy was employed.
- The framework simultaneously performs segmentation and classification, validated on a new UAV dataset from Bangladesh.
- Gradient-weighted Class Activation Mapping (Grad-CAM) was used for model interpretability.
Main Results:
- The model achieved a Dice coefficient of 0.7344 and mIoU of 0.6904 for segmentation, with 98.77% classification accuracy.
- Attention mechanisms and joint learning with Gaussian blur improved performance and feature discrimination.
- The framework demonstrated a ~50% reduction in parameters and ~48.6% faster inference speed compared to single-task models.
Conclusions:
- The proposed framework offers a robust and efficient solution for integrated aquatic vegetation analysis from UAV data.
- It is suitable for real-time applications, large-scale biodiversity monitoring, and invasive species detection.
- The study highlights the importance of unified structural and semantic understanding for effective freshwater ecosystem management.