VIETNAM NATIONAL UNIVERSITY HO CHI MINH CITY UNIVERSITY OF INFORMATION TECHNOLOGY THESIS GRADUATION 19521770 — NGO DAC LOI THESIS ADVISOR PhD. DO DUY THANH HO CHI MINH CITY, 2023 TABLE OE CONTENTS TABLE OF CONTENTS IV.- ng HH Hư 7 LU Background 0n.2 Objective and SCOPE 200.-- Sàn HT HH TH TH TH TH HT TH Hàn HH 14 Chapter 2. THEORETICAL BACKGROUND AND RELATED WORKS.1 Wildfire Classification Using Deep Learning.2 ML-based flame defecfion.- --- àL nh HH TH HT TH TT nh HH TH 25 Chapter 3. EXPERIMENTS AND RESULTS.1 Data Collection and Data Set Generafing.-- cc Sàn HH HH HH HH HT HH Hư, 28 KSPA) ri.- óc HH HT HH HH nghi 32 3.---- HH HH HT HH TH ng nhiệt 53 K0).
56 Chapter 4: CONCLUSIƠNS.2 Limitations and Challenges. HH ng ng ky 59 4. ng HH HT TH TT HH Hàn HH 61 THANKS During the process of studying and practicing at the University of Information Technology to improve my knowledge and skills, I received a lot of attention and help from teachers, family and friends. With the deepest gratitude, I would like to thank the school and the teachers of the Information Systems Department for imparting the knowledge that serves as a foundation for me.
Ngo Dac Loi COMMENTS ABSTRACT Wildland fire detection and classification is a challenging task due to the complex nature of fire dynamics and the need for rapid and accurate identification. In recent years, deep learning models and pixel segmentation techniques have shown great potential in improving the performance of wildland fire detection and classification systems. Deep learning models, such as convolutional neural networks (CNNs), have demonstrated their ability to learn and extract meaningful features from wildfire-related images, enabling accurate fire detection. Additionally, pixel segmentation techniques, such as Maximally Stable Extremal Regions (MSER) and nonmaximum suppression (NMS), have proven effective in identifying fire boundaries and distinguishing them from other objects in the scene.
The integration of deep learning models and pixel segmentation techniques offers significant improvements in automating wildland fire detection and classification. These techniques allow for real-time analysis of imagery data, enabling early detection and timely response to fire incidents. Moreover, their ability to accurately discriminate between fire and non-fire regions reduces false alarms and enhances the efficiency of fire management operations. This paper presents an overview of deep learning models and their application in wildland fire detection and classification.
Additionally, the paper explores the role of pixel segmentation techniques, particularly MSER-NMS, in enhancing the precision and recall of fire detection algorithms. The results of the experiments are illustrated through a website to obtain a demonstration of the used algorithms’ performance.1 Background Wildfires are increasing around the globe in frequency, severity and duration, heightening the need to understand the health effects of wildfire exposure. The risk of wildfires grows in extremely dry conditions, such as drought, heat waves and during high winds. Wildfire smoke is a mixture of hazardous air pollutants, such PM2.5, NO2, ozone, aromatic hydrocarbons, or lead.
In addition to contaminating the air with toxic pollutants, wildfires also simultaneously impact the climate by releasing large quantities of carbon dioxide and other greenhouse gases into the atmosphere. With climate change leading to warmer temperatures and drier conditions and the increasing urbanization of tural areas, the fire season is starting earlier and ending later. Wildfire events are getting more extreme in terms of acres burned, duration and intensity, and they can disrupt transportation, communications, water supply, and power and gas services. Wildfires that burn near populated areas can have significant impact on the environment, property, livestock and human mortality and morbidity depending on the size, speed and proximity to the fire, and whether the population has advanced warning to evacuate.
Wildfire smoke is a mixture of air pollutants of which particulate matter (PM) is the principal public health threat. Between 2001 and 2022, Vietnam experienced a depletion of 32.7 thousand hectares of tree cover due to fires and an additional 3.39 million hectares from various other causes. The highest recorded tree cover loss attributed to fires within this timeframe occurred in 2016, accounting for 3.27 thousand hectares, which constituted 0.93% of the total tree cover loss for that particular year [1]. Human activities, both traditional and modern, contribute significantly to the incidence of wildfires in Vietnam.
Agricultural practices, deforestation, and land-use changes, often driven by the expanding need for cultivation and urbanization, increase the vulnerability of landscapes to fire outbreaks. Vietnam has recognized the importance of proactive wildfire management and has implemented strategies to prevent, control, and respond to fire incidents. These efforts involve a combination of technological interventions, community engagement, and legislative measures. The urgency of early wildfire detection cannot be overstated.
Swift identification of ignition points is paramount in deploying timely response measures, thereby curbing the uncontrolled spread of fires. Early detection also plays a crucial role in facilitating efficient evacuations, enhancing community safety, and minimizing the detrimental impact on ecosystems. Wildfire detection presents a multifaceted challenge owing to the vast and often remote terrains, diverse environmental conditions, and the imperative for real-time monitoring. Traditional methods struggle to cope with these complexities, prompting the exploration of ML and DL solutions.
These technologies offer the potential for automated, continuous monitoring, and the extraction of nuanced patterns from intricate datasets. ML algorithms, trained on diverse datasets encompassing historical wildfire data, satellite imagery, and meteorological parameters, stand out as versatile tools for discerning potential wildfire-prone areas. Classification models distinguish between normal environmental conditions and those indicative of a wildfire, while regression models predict the likelihood and intensity of fire outbreaks. The integration of multiple data sources augments detection accuracy.
DL, with its prowess in image and pattern recognition, plays a pivotal role in wildfire detection. Architectures like Convolutional Neural Networks (CNNs) excel in processing extensive datasets such as satellite imagery, sensor data, and social media feeds, elevating the precision of detection systems. DL models exhibit an innate capacity to comprehend intricate spatial and temporal patterns. We have encapsulated our contributions in the following summary: e We present a unique dataset of side-by-side RGB/IR imagery collected during a prescribed fire near Flagstaff, Arizona in 2021 [2].
The images are jointly labeled by two human experts with fire/no-fire and smoke/nosmoke labels.2 Drones for Firefighting Services e We deploy ML pre-trained on diverse datasets fire for classification. Additionally, we illustrate a comparative analysis of the results. 10 e Implemented and deployed advanced object detection architectures, including NMS and MSER, leveraging their capabilities for efficient and accurate identification of objects in diverse contexts. e We construct a website designed to visualize fire detection outcomes, allowing users to upload images for analysis and presentation of results.2 Objective and scope 1.1 Objective Train machine learning models: The goal is to create machine learning algorithms capable of proficiently detecting and categorizing wildland fires.
This entails acquiring and preprocessing labeled datasets comprising both fire and non- fire areas and leveraging these datasets to train models adept at precisely classifying images associated with fires. This process will involve utilizing datasets sourced [9] to enhance the algorithm's ability to recognize and distinguish fire-related patterns. Enhance accuracy and efficiency: The goal is to improve the accuracy and efficiency of wildfire detection and classification by leveraging the power of deep learning models. Deep learning algorithms, such as convolutional neural networks (CNNs), can learn complex features from images, enabling more accurate detection and classification of fires.
Adaptability and scalability: The aim is to develop models and algorithms that can adapt to different environments and are scalable to handle large datasets. Integration with Web Backend: Integrate the ML and DL models into the web backend, ensuring smooth communication between the frontend and backend components for image processing. Result Visualization on Web Interface: Present the detection results on the web interface in a visually interpretable manner, providing users with immediate feedback on the presence or absence of wildland fires in the uploaded images.2 Scope Machine learning algorithms play a pivotal role in the analysis of both infrared (IR) and RGB images, offering a sophisticated approach to detecting early signs of potential fire outbreaks. The efficacy of these algorithms is notably 12 enhanced through meticulous training on diverse datasets, allowing them to discern and recognize intricate patterns across a spectrum of environmental conditions.
This adaptability is crucial in achieving a higher level of accuracy in fire detection, transcending the limitations posed by varying atmospheric and geographical factors. In addition to their fundamental function of classifying fire images, contemporary studies are increasingly focusing on more advanced vision tasks within the realm of fire detection. A notable example is the real-time identification and precise localization of fire or smoke areas, often represented with bounding boxes. This evolution in the capabilities of machine learning algorithms transcends traditional fire detection methodologies, paving the way for a more nuanced understanding of fire dynamics and facilitating proactive intervention strategies.
Furthermore, the utilization of established object detection frameworks has proven to be instrumental in achieving real-time fire detection across diverse environments. This is particularly evident in the realm of surveillance videos, where the algorithms can dynamically identify and track potential fire occurrences, contributing to enhanced situational awareness and rapid response mechanisms. In summary, the multifaceted application of machine learning algorithms in fire detection not only involves their ability to classify fire images accurately but extends to advanced vision tasks such as real-time identification and localization of fire or smoke areas. The continuous evolution of these algorithms, coupled with their adaptability to diverse datasets, signifies a significant stride towards more effective and sophisticated fire detection methodologies, particularly in dynamic and complex environmental scenarios.3 Thesis structure This thesis is divided into five chapters, as follows: ¢ Chapter 1: Introduction * Chapter 2: Theoretical background and related works ¢ Chapter 3: Experiments and results ¢ Chapter 4: Conclusions 14 Chapter 2.
THEORETICAL BACKGROUND AND RELATED WORKS 2.1 Wildfire Classification Using Deep Learning Classify pixels within our FLAME? dataset into four categories: Flames with Smoke (YY), Flames with No Smoke (YN), Smoke with No Flames (NY), and No Flames No Smoke (NN). We employed several well-established machine learning and deep learning classification models. These include ResNet, DenseNet121, SqueezeNet on the dataset. The above models Using Flame2 Collect Data dataset Split data into Split Data train set and test set Using ResNet, Train KNN Model SqueezeNet DenseNet models FIGURE 2.1 Flow chart of training process.1 Categories of dataset Label Number of IR images | Number of RGB images | Class Fire, Smoke 25,434 25,434 YY Fire, No Smoke 14,317 14,317 YN No Fire,No Smoke | 13,700 13,700 NN No Fire, Smoke 0 0 NY } Ww Ww; M Ị b WX, + WX, + 2X; + Wx, +b a *; ef =| Might, + WX,+ WX, + WX, + — , activation Hyhay + X;+ ,x„ +L FIGURE 2.2 Simple Neural network [17] As depicted in figure 2.2, it becomes evident that within a feedforward neural network, each layer is equipped with its distinctive weight matrix and biases, symbolizing the weights associated with individual neurons within the layer.
The utilization of the nn.Linear layer facilitates the realization of the matrix multiplication operation between input data and the weight matrix, coupled with the addition of the bias term for each respective layer [17]. 16 10 ReLU -10 on iv n FIGURE 2.3 ReLU layer [17] ReLU (Rectified Linear Unit) layer is a fundamental component in neural networks, serving as an activation function applied to the output of each neuron in a specific layer. It introduces non-linearity to the model by transforming negative inputs to zero and leaving positive inputs unchanged. Mathematically, the ReLU activation function is defined as f (x) = max (0, x), where x is the input to the neuron.
The ReLU layer aids in capturing complex patterns and relationships within the data, allowing the neural network to learn and adapt during the training process. With RGB images, each pixel is represented by three color channels: Red, Green, and Blue.