A drone-assisted deep learning framework for disaster assessment in real time and emergency response

Disasters affect citizen safety, municipal infrastructure, and the economy, making disaster management vital and difficult. Rapid urban expansion and climatic extremes have increased big catastrophe frequency and complexity. Situational awareness and quick decision-making are needed. In catastrophes, traditional evaluation methods may not give accurate and timely information owing to enormous volumes of data and quick changes. Due to these restrictions, this research offers an AI system for real-time catastrophe assessment and response using drones and the Adaptive Convolutional-Transformer Fusion Network. Convolutional feature extraction and attention-based temporal modeling allow ACTF-Net to grasp sequential interactions and dynamic spatial patterns in high-dimensional catastrophic data. Adaptive Relevance Filtering and Contextual Scaling Transformation increase data dependability and feature relevance, enabling noisy or missing data learning. Testing regularly shows that ACTF-Net outperforms previous systems like BERT and CapsNet. F1-score is 0.981, area under curve is 0.993, and catastrophe severity prediction rate is 0.96. The model’s consistent performance across datasets of different sizes across time supports real-time applications. The proposed method uses multimodal drone data, powerful preprocessing, and novel metrics like the Contextual Sensitivity Index and Temporal Stability Score to reliably categorize and assess catastrophe severity using multiple labels. The results indicate that ACTF-Net may improve situational awareness, resource usage, and response time in high-risk crisis situations.