HYBRID YOLO CNN ARTIFICIAL INTELLIGENCE FRAMEWORK FOR MULTI-CLASS DISASTER IMAGE ANALYTICS IN LOCAL GOVERNMENT EMERGENCY RESPONSE AND HEALTHCARE MANAGEMENT

Abstract

Disasters such as floods, wildfires, structural collapses, and mass-casualty incidents impose severe operational burdens on local government authorities and emergency healthcare systems. Existing computer vision frameworks address these challenges in isolation, lacking integrated multi-task capacity to simultaneously detect disaster types, locate victims, classify medical severity, and assess infrastructure damage. This article proposes a Hybrid YOLO-CNN Artificial Intelligence Framework (HYCAIF) that unifies YOLOv8-based object detection with a Convolutional Neural Network (CNN) feature backbone, Squeeze-and-Excitation (SE) attention, Convolutional Block Attention Module (CBAM), and a Feature Pyramid Network (FPN) neck into a cohesive multi-head architecture. The framework is designed explicitly to serve local government disaster response functions, including GIS-integrated emergency alerting, medical triage dispatch, and administrative dashboarding, thereby bridging an underexplored gap between sub-national governance and deep learning. Experiments on a composite dataset comprising the Aerial Image Dataset for Emergency Response (AIDER) and the Detection in Optical Remote Sensing Images (DOTA) benchmark yield a mean Average Precision at IoU 0.5 (mAP@0.5) of 91.7%, Precision of 90.2%, Recall of 89.6%, and F1-Score of 89.9%, representing improvements of 9.6, 9.7, 10.3, and 10.6 percentage points over the best-performing baseline, respectively. The model achieves a sustained inference throughput of 71 frames per second on commodity GPU hardware, satisfying real-time deployment constraints. These results affirm the practical viability of the proposed framework for adoption by municipal authorities, civil protection agencies, and emergency healthcare dispatchers.