This project is about the development of smoke detection algorithms in case of forest fires. Our main problem statement is applying classification and object detection to identify smoke / fire through images I am using Mask RCNN model for object detection in this case.
Step 1: Data gathering and collection: We have gathered data available online at below location. https://sintecsys-omdena.s3.amazonaws.com/images3.zip Dataset have almost 16.5K images (~8k masked and ~8k original images).
Step 2: Exploration of data: Masked images have been already created. The model needs to be trained on existing masked images and start predicting. Size of images is: 1920 X 1080 Step 3: Applying Image recognition and Object detection models.
Iteration 1: Simple image recognition model and tried to train with simple CNN model
Iteration 2: Alexnet model for classification
Iteration 3: Mask RCNN model
Step 4: Demonstration of Mask RCNN model
a. Overview of Mask RCNN:
b. Training and Validation Loss:
c. Precision and Recall numbers:
d. Result and comparison with Unet Masks:
Step 5: Demonstation using Flask and HTML / Javascript UI:
Step 6:
Generating Alerts: Generating alerts through emails
