Automated-detection-of-hotspot-in-thermal-images

A CNN network trained for detecting hotspot in thermal imagery via transfer learning. The binary detection method determines whether thermal images contains hotspots or not. A pre trained VGG16 keras model is used to extract arbitary feature vectors that globally describe the image. Based on the extracted features, a Logistic Regression classifier is trained on the features for image classification purposes. For images that contain hotspot, the hotspot regions are localised using Otsus's adaptive thresholding techniques. See link below for demonstration
http://hotspotdl.herokuapp.com/

Install via pip

Fix opencv dependency issue when installing with pip on debian system

apt-get install -y libsm6 libxext6 libxrender-dev

Fix - add dependencies for opengl

sudo apt-get install -y libgirepository1.0-dev build-essential \
  libbz2-dev libreadline-dev libssl-dev zlib1g-dev libsqlite3-dev wget \
  curl llvm libncurses5-dev libncursesw5-dev xz-utils tk-dev libcairo2-dev

Software has been developed in Python. Install dependencies using pip

pip install -r requirements.txt

Install - Using Docker

Build docker image

docker build -t <name of image > .

Run docker container

docker run -it -e PORT=5000 -e DATABASE_URL=sqlite:///tmp/tmp.sqlite -p 5000:5000 <name of image > 

Instructions on using the VGG16 model for extracting features from training and testing data.

python feature_extraction.py [-h] -i IMAGES -o OUTPUT
optional arguments:
-h --help show this help message and exit
-i IMAGES, --images IMAGES path to input images
-o OUTPUT, --output OUTPATH path to output hdf5 file

Example - Extracting features from training images

python feature_extraction.py -i ../training_test_data/training/ -o features_training.hdf5

Example - Extracting features from testing images

python feature_extraction.py -i ../training_test_data/testing/ -o features_testing.hdf5

Once the features have been extracted, a Logistric Regression model can be trained on the extracted features

python classifier.py [-h] -train TRAIN_HDF5 -test TEST_HDF5
optional arguments:
-h --help show this help message and exit
-train TRAIN_HDF5, --train_hdf5 TRAIN_HDF5 path to trained hdf5 file
-test TEST_HDF5, --test TEST_HDF5 path to test hdf5 file

Example - Training a classifier on the extracted features

python classifier.py -train features_training.hdf5 -test features.hdf5

Output from the python script would be a model.cpickle file

Running the flask app to perform hot spot detection.

Run the following commands to create your apps database tables and perform initial migration

flask db init
flask db migrate
flask db upgrade

To run the web application use:

python app.py

A web server would be spawend and one can type in the browser localhost:5000 and you should see:

Input image- thermal image with hotspots in image

Input image - thermal image with no hotspots in image