A novel framework for human object classification in flooded environments using YOLOv12n and FastGRNN-MBT

Area of Science:

• Computer Science
• Artificial Intelligence
• Image Processing

Background:

• Floods are devastating natural disasters causing significant destruction.
• Accurate human object detection in flooded environments is critical for disaster response.
• Existing methods often lack high classification accuracy and misclassify objects.

Purpose of the Study:

• To present a novel Fast Gated Recurrent Neural Network-based Mean Binary cross entropy Taylor concept (FastGRNN-MBT) for human object classification in flooded environments.
• To improve the accuracy and reliability of human detection during flood disasters.

Main Methods:

• Image pre-processing using adaptive median filter and Retinex algorithm for enhancement.
• Object detection utilizing You Only Look Once version 12 nano (YOLOv12n), fine-tuned with the CAO algorithm.
• Feature extraction and final human object classification using the proposed FastGRNN-MBT model.

Main Results:

• The FastGRNN-MBT model achieved high performance metrics: 96.979% recall, 97.763% precision, 96.226% accuracy, and 96.988% F1-score.
• Demonstrated significantly improved classification accuracy compared to existing methods.
• Achieved efficient execution and inference times (147.236s and 28.178s, respectively).

Conclusions:

• The FastGRNN-MBT model is highly effective and suitable for classifying human objects in flooded environments.
• This method offers a significant advancement in disaster response technology for human detection.
• The proposed approach outperforms current state-of-the-art classification methods.