Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Uncertainty-Aware Post-Detection Framework for Enhanced Fire and Smoke Detection in Compact Deep Learning Models

Aniruddha Srinivas Joshi, Godwyn James William, Shreyas Srinivas Joshi

TL;DR

The paper tackles the reliability-efficiency trade-off in vision-based fire and smoke detection on compact detectors. It presents an uncertainty-aware post-detection rescoring framework that combines single-pass dropout-based uncertainty with domain-specific visual features in a lightweight Confidence Refinement Network (CRN) to adjust detection confidences without altering the base detector. By evaluating on the D-Fire dataset with YOLOv5n and YOLOv8n, the method achieves significant gains in precision, recall, and mean average precision (mAP), while incurring only modest increases in end-to-end latency. The approach offers a practical pathway to more robust, real-time fire safety systems on edge devices, enabling better performance in challenging scenes without resorting to heavier models.

Abstract

Accurate fire and smoke detection is critical for safety and disaster response, yet existing vision-based methods face challenges in balancing efficiency and reliability. Compact deep learning models such as YOLOv5n and YOLOv8n are widely adopted for deployment on UAVs, CCTV systems, and IoT devices, but their reduced capacity often results in false positives and missed detections. Conventional post-detection methods such as Non-Maximum Suppression and Soft-NMS rely only on spatial overlap, which can suppress true positives or retain false alarms in cluttered or ambiguous fire scenes. To address these limitations, we propose an uncertainty aware post-detection framework that rescales detection confidences using both statistical uncertainty and domain relevant visual cues. A lightweight Confidence Refinement Network integrates uncertainty estimates with color, edge, and texture features to adjust detection scores without modifying the base model. Experiments on the D-Fire dataset demonstrate improved precision, recall, and mean average precision compared to existing baselines, with only modest computational overhead. These results highlight the effectiveness of post-detection rescoring in enhancing the robustness of compact deep learning models for real-world fire and smoke detection.

Uncertainty-Aware Post-Detection Framework for Enhanced Fire and Smoke Detection in Compact Deep Learning Models

TL;DR

The paper tackles the reliability-efficiency trade-off in vision-based fire and smoke detection on compact detectors. It presents an uncertainty-aware post-detection rescoring framework that combines single-pass dropout-based uncertainty with domain-specific visual features in a lightweight Confidence Refinement Network (CRN) to adjust detection confidences without altering the base detector. By evaluating on the D-Fire dataset with YOLOv5n and YOLOv8n, the method achieves significant gains in precision, recall, and mean average precision (mAP), while incurring only modest increases in end-to-end latency. The approach offers a practical pathway to more robust, real-time fire safety systems on edge devices, enabling better performance in challenging scenes without resorting to heavier models.

Abstract

Accurate fire and smoke detection is critical for safety and disaster response, yet existing vision-based methods face challenges in balancing efficiency and reliability. Compact deep learning models such as YOLOv5n and YOLOv8n are widely adopted for deployment on UAVs, CCTV systems, and IoT devices, but their reduced capacity often results in false positives and missed detections. Conventional post-detection methods such as Non-Maximum Suppression and Soft-NMS rely only on spatial overlap, which can suppress true positives or retain false alarms in cluttered or ambiguous fire scenes. To address these limitations, we propose an uncertainty aware post-detection framework that rescales detection confidences using both statistical uncertainty and domain relevant visual cues. A lightweight Confidence Refinement Network integrates uncertainty estimates with color, edge, and texture features to adjust detection scores without modifying the base model. Experiments on the D-Fire dataset demonstrate improved precision, recall, and mean average precision compared to existing baselines, with only modest computational overhead. These results highlight the effectiveness of post-detection rescoring in enhancing the robustness of compact deep learning models for real-world fire and smoke detection.

Paper Structure

This paper contains 24 sections, 3 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Compact Models for Fire and Smoke Detection
  4. Post-Detection Methods for Object Detection
  5. Challenges in Fire and Smoke Detection
  6. Broader Context in Fire and Smoke Detection
  7. Gap and Motivation
  8. Proposed Method
  9. Uncertainty-Aware Confidence Estimation
  10. Feature-Aware Confidence Normalization
  11. Confidence Refinement Network (CRN)
  12. Final Decision Processing
  13. Experimental Setup
  14. Dataset
  15. Model Training and Configuration
  16. ...and 9 more sections

Figures (3)

  • Figure 1: Pipeline of the proposed framework. The framework refines confidence scores using uncertainty estimation, detection region features, and a learned CRN.
  • Figure 2: Architecture of the Confidence Refinement Network (CRN). The network refines confidence scores using uncertainty estimates and visual features.
  • Figure 3: Qualitative results produced by the proposed method. The examples illustrate the effectiveness of the approach across different scenarios.