2. Why do we need to detect forest fires ?
• Forest fires pose a serious threat to natural environment and public
safety.
• Early detection of forest fires are essential, as after a certain time it is
hardly uncontrollable.
• Traditional detection of forest fires using satellite imaging takes a
relatively longer time compared to that of video processing.
• Forest fires are identified quickly by drones, thus informing the local fire
brigades to arrive at a sort period of time
3. Training DATA CNN
Forest Fire
Detection Model
Forest
Drone Camera
Storage Server
Result and Alarm
Overview of the Entire Process
4. Convolution Layer:
3 4 4
1 0 2
-1 0 3
2 3 7 4 6 2 9
6 6 9 8 7 4 3
3 4 8 3 8 9 7
7 8 3 6 6 3 4
4 2 1 8 3 4 6
3 2 4 1 9 8 3
0 1 3 9 2 1 4
=
91 100 83
69 91 127
44 72 74
• The left side is a pixel representation of an image.(in orange)
• is the convolution operator.
• The next matrix is called the filter or the kernel .(in light blue)
• The resultant matrix is obtained by some simple calculation . (in golden)
5. Convolution Layer Contd..
• The value of 91 as output is simply obtained by matrix multiplication by placing the
filter on the top left hand corner of the image matrix and then each individual pixels
are multiplied and added.(91=2*3+3*4+7*4+6*1+6*0+9*2+3*(-1)+4*0+8*3).
• For the above example the stride (Denoted by ‘s’ ) shown is 2. The stride mainly
determines the no of pixels to be jumped.
• Padding is sometime provided to image to utilize the information present at the
corners of the images.
• For a (n*n) dimension image when applied with a padding of ‘p’ and convoluted
with a filter of (f*f) with a stride ‘s’ , the output image dimension is given by:
(
𝑛+2𝑝−𝑓
𝑠
+ 1,
𝑛+2𝑝−𝑓
𝑠
+ 1)
• For a 3D image, we need to have a 3 channel filter, thus the no. of channel of the
filter must be same as the actual image.
6. Convolution Layer Contd..
• A bias (b) ∈ a real no (R) is added
to the output matrix (shown with
green)
• Some activation function such as
sigmoid or relu is taken of the
resultant output (with the bias)
• This completes a single step of
Convolution,
• The total number of parameters
depend on the size of the filter as
well as the bias added to it.
n=5, p=1, f=3, s=1, so output is (5*5) matrix. https://setosa.io/ev/image-kernels/
7. Pooling Layer:
• The main purpose of introducing the pooling layer is to speed the computation, as well as makes
some features that detects a bit more robust.
• There are different type of pooling layer. However, the most common pooling layer is the maxpool
and the average pool.
• In the maxpool, it computes the maximum value of the certain plot of matrix and in average pool,
it computes the average.
8. Flatten Layer:
• This layer converts the matrix
into a 1D array.
• This is mostly done at the last
stage of the CNN structure.
• From here the NN are formed to
learn and determine the
parameters.
• The model is prepared after the
stage.
9. Dense Layer:
• The purpose of using the dense layer is to add the fully connected layer to the neural network.
10. A drone is a small flying robot controlled remotely by a person or a computer, it can used for
various purpose
The uses are:
• Forest Patrol : Forest protection group can patrol and observe the forest area with the help
of camera.
• Detection : This drone has sensors which actually detect the fire as well as speed of the fire.
• Plant, Animal Protection and Environmental Monitoring : it is also used for tracking
migration and observe the plant and animal with the help of this device.
• Forest Administrator Rescue : It is also used for rescue the forest,
• Meteorological Detection : It is used to detect climate and weather.