Related Experiment Video
Updated: Jul 1, 2026
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
HYDRA-XAI dual-backbone disaster scene recognition using ResNet50-Swin transformer feature fusion, explainable
Akella S Narasimha Raju1, G Geetha2, B Subashini3
1Department of Computing Technologies, School of Computing, College of Engineering & Technology, SRM Institute of Science and Technology, Kattakulathur, Chennai, 603203, Tamil Nadu, India.
Scientific Reports
|June 29, 2026
Summary
HYDRA-XAI rapidly identifies disaster scenes in diverse images using a Transformer and convolutional model. This system achieves high accuracy for emergency response, aiding in critical decision-making.
Area of Science:
- Computer Science
- Artificial Intelligence
- Remote Sensing
Background:
- Effective disaster scene identification is crucial for emergency response.
- Operational streams often contain numerous non-disaster images, challenging rapid analysis.
- Few systems effectively handle human-inflicted damage scenarios in diverse imagery.
Purpose of the Study:
- To develop an end-to-end system (HYDRA-XAI) for classifying 6 types of disaster scenes.
- To leverage global Transformer context and local convolutional evidence for robust disaster identification.
- To generate multimodal evidence for recommending specific operations based on disaster class.
Main Methods:
- Utilized Frozen ResNet50 and Swin Transformer Tiny for feature extraction (2048-dim and 768-dim representations).
- Concatenated features into a 2816-dim representation, classified using XGBoost and SVM-RBF with averaged probabilities.
- Employed Grad-CAM++, Transformer attention rollout, and LIME for generating explainable AI (XAI) evidence.
Main Results:
- Achieved high performance metrics: Accuracy=0.9935, Macro-F1=0.9929, Weighted-F1=0.9935, ROC-AUC=0.9999.
- Demonstrated a small train/test gap (Macro-F1=0.0069) with stable generalization.
- Boosted escalation precision to 0.9972 using LIME-coverage thresholds and validation confidence.
Conclusions:
- HYDRA-XAI demonstrates stable generalization and prediction capabilities within its training domain (CDD).
- The model's decisions are supported by evidence-based routing, enhancing trust and interpretability.
- Further verification of external dataset robustness is recommended for broader applicability.
