Article Preview
Top1. Introduction
With the change in climatic conditions, there is an increase in the frequency as well as severity of the natural calamities(Pi et al., 2020). Hurricanes are the most catastrophic disasters occurring on Earth. They are called as tropical cyclones as they take place in the tropical areas because of the presence of warm water. The sun heats up the sea waters leading to the formation of huge clouds and thus hurricanes. Hurricanes are accompanied by heavy rain, floods and high-speed winds of about 320 km/hour (Dawood et al., 2020). Hurricane Harvey, a Category 4 hurricane made landfall in the Houston region in the year 2017 with a speed of 210 km/hour killing more than 100 people. It was the most powerful hurricane that struck this region in 56 years and caused flooding of the low-lying coastal regions. It caused a huge damage of $125 billion. Timely and accurate detection of damage caused due to hurricanes becomes imperative for effectively mobilizing relief efforts to the inflicted people.
Satellite imagery is of immense importance in disaster response for identification of the impact of the disaster. Since satellite images cover a very vast area of the ground surface, they form a very useful resource for leveraging disaster damage detection. However, analysis of satellite images for detection of disaster impact is extremely challenging. Deep learning (DL), a family of machine learning, could be utilized on these satellite images for providing insight on the damages caused by the hurricane disaster (Dotel et al., 2020). Also, DL shows promising results for automation of the disaster detection tasks. DL exceeds the performance of humans in classification of complex images. It is inspired from the structure and functions of the human brain which is called Artificial Neural Networks (ANN). Human beings learn from experience, similarly DL algorithms learn by performing a task repeatedly. It is called deep learning because the DL network is composed of several layers that help the network to learn. The layers are composed of an artificial neuron which is the primary component of an ANN (Kaur et al., 2021).
Numerous architectures are available for Deep Learning. Successful understanding of images can be done through the Convolutional Neural Networks (CNN)(Pritt & Chern, 2018). A CNN is a DL network particularly useful to extract hierarchical characteristics from the images(Kaur et al., 2021). Figure 1 shows the illustration of a basic CNN. The CNN model is divided into two parts, the first being feature extraction and the second: classification (Cao & Choe, 2020). The first part is composed of the input, convolutional layer and the pooling layer while the second part is composed of the dense and the output. The input layer takes an input image of a fixed size whose resizing could be done if required. The image is then convolved with filters. The pooling layer helps in reduction of the image size. The output of the feature extraction part is known as feature map. The output of the classification part gives the classification result through the fully connected and output layer (Phung & Rhee, 2019). A CNN based architecture can be applied on the satellite images for determining the damage caused by the hurricane disaster.
Figure 1.
Schematic Diagram of a Basic CNN [7]
In this paper, a CNN based architecture has been proposed for classification of images into damaged and undamaged classes using the satellite images of Harvey Hurricane. In this article, the major contribution of the author is as follows:
- 1.
A Convolutional Neural Network based model is proposed having five convolutional layers, five pooling layers, one flattening layer and two dense layers.
- 2.
The proposed CNN network is simulated on the Adam optimizer at a learning rate of 0.0001 and 30 epochs.
- 3.
The dataset considered for simulation has 23000 satellite images of Hurricane Harvey that are classified into damaged and non-damaged classes.