FireNet: A Specialized Lightweight Fire & Smoke Detection Model for Real-Time IoT Applications (1905.11922v2)

Abstract: Fire disasters typically result in lot of loss to life and property. It is therefore imperative that precise, fast, and possibly portable solutions to detect fire be made readily available to the masses at reasonable prices. There have been several research attempts to design effective and appropriately priced fire detection systems with varying degrees of success. However, most of them demonstrate a trade-off between performance and model size (which decides the model's ability to be installed on portable devices). The work presented in this paper is an attempt to deal with both the performance and model size issues in one design. Toward that end, a `designed-from-scratch' neural network, named FireNet, is proposed which is worthy on both the counts: (i) it has better performance than existing counterparts, and (ii) it is lightweight enough to be deploy-able on embedded platforms like Raspberry Pi. Performance evaluations on a standard dataset, as well as our own newly introduced custom-compiled fire dataset, are extremely encouraging.

• The paper introduces FireNet, a lightweight CNN that achieves fire and smoke detection with accuracies up to 96.53% on IoT devices.
• It employs a shallow architecture with three convolution layers and four dense layers optimized for low-cost hardware like the Raspberry Pi 3B.
• The integrated system uses IoT connectivity for remote alerts, reducing false alarms compared to traditional sensor-based methods.

FireNet: A Specialized Lightweight Fire & Smoke Detection Model for Real-Time IoT Applications (1905.11922)

Introduction

"FireNet: A Specialized Lightweight Fire & Smoke Detection Model for Real-Time IoT Applications" addresses a critical need for efficient, portable, and cost-effective fire detection systems. Traditional fire alarms, which predominantly rely on physical sensors like thermal and smoke detectors, often face challenges such as false alarms and significant detection delays. These limitations necessitate alternative methods that enhance robustness and reliability.

This paper presents "FireNet," a novel neural network specifically designed for low-cost embedded platforms like the Raspberry Pi 3B. FireNet aims to achieve high performance in fire detection without compromising on model size, which is crucial for deployment on portable devices.

Related Work

Early attempts at fire detection often involved handcrafted techniques hinged on identifying motion and color properties of flames [chen2004early] [ccelik2007fire] [rafiee2011fire]. While these techniques were not computationally intensive, they suffered from high false positives and were not suited for real-world applications requiring compact and cost-effective solutions. With the advancement of deep learning (DL) technologies, several studies have explored DL-based fire detection using convolutional neural networks (CNNs) like AlexNet, VGG16, and SqueezeNet (Oussa, 2012), but these models tend to be computationally expensive, rendering them impractical for low-cost, real-time applications.

Dataset

The research identified a shortage in diverse fire datasets and sought to create a more comprehensive collection by integrating various sources, including self-shot videos from challenging environments and images with fire-like objects in the background. The authors curated a diverse training dataset consisting of 1,124 fire images and 1,301 non-fire images.

Figure 1: Few images from our training dataset.

The test dataset, designed to simulate realistic scenarios, included both fire and non-fire images, allowing for a thorough evaluation of the developed model.

Figure 2: Few images from our test dataset.

Proposed Approach

"FireNet" is a shallow CNN tailored for real-time fire detection, achieving performance levels superior to existing systems with minimal computational load. The network architecture includes three convolution layers and four dense layers, providing an optimal balance between detection accuracy and computational efficiency. This design ensures that "FireNet" maintains a high frame rate while running on low-cost hardware such as a Raspberry Pi 3B.

Figure 3: Architecture of the proposed neural network.

The above architecture serves the need for a more efficient and reliable fire detection method when compared with traditional handcrafted, image-processing techniques and even existing CNN-based models.

Implementation and IoT Integration

"FireNet" is implemented on a hardware platform integrating a Raspberry Pi 3B, a smoke sensor, distinct smoke, and fire alarms, along with a provision for remote data transmission via IoT using services like Amazon S3 and Twilio.

Figure 4: Overview of the complete unit: (a) Camera (b) Raspberry Pi 3B (c) Microcontroller (d) Cloud storage and SMS/MMS service (Amazon S3 and Twilio) (e) End-user device for receiving fire alert (visual and textual) (f) Smoke alarm (g) Fire alarm with a different sound than smoke alarm (h) Smoke sensor for sensing smoke and thus, aiding in fire-smoke differentiation.

This integrated solution facilitates a comprehensive system for early detection and differentiation between fire and smoke, thereby reducing false and delayed alarms traditionally associated with hardware-only systems.

Results

The performance of "FireNet" was evaluated on two datasets: the newly created, diverse custom-compiled dataset, and a standard fire dataset provided by Foggia et al. The results demonstrated that "FireNet" achieved high accuracy rates: 93.91% on the custom-compiled dataset and 96.53% on Foggia's dataset.

Figure 5: Training and validation curves for model accuracy.

Figure 6: Training and validation curves for model loss.

The training process of "FireNet" utilized a 70-30 split between training and validation datasets. The performance was measured in terms of accuracy, precision, recall, and F-measure. Notably, while the performance was slightly better on Foggia's dataset, the results from the diverse custom-compiled dataset highlighted the model's robust generalization capabilities.

Discussion on Similar Approaches

Numerous approaches utilizing both handcrafted and deep learning-based feature extraction have been explored (1465.87321, 1523.02456). Traditional handcrafted techniques have been criticized for their reliance on manual feature extraction, which can be cumbersome and inefficient for datasets with high volumes of image data. While deep learning approaches offer automatic feature extraction capabilities, their computational demands can be significant, posing challenges for deployment on resource-limited hardware.

FireNet distinguishes itself by proposing a solution that achieves a compelling balance between model size and performance, making it suitable for cost-effective, mobile, and real-time fire detection applications. It performs well in comparison to the more cumbersome models mentioned in related works, maintaining real-time frame rates on an economical setup without burdening computational resources.

Conclusion

The "FireNet" paper proposes a highly effective, lightweight solution for real-time fire and smoke detection in IoT applications. Its architecture, effective dataset creation, and practical implementation using a Raspberry Pi 3B reflect significant efforts to combine deep learning with IoT for fire detection. This work contributes a viable, resource-efficient model that performs reliably well, demonstrating the potential for further development and optimization in AI-driven fire detection systems. The dataset and trained model have been released for use by other researchers, encouraging continued progress and advancements in the field.