skylab-wildfiredetection

Reference: https://github.com/aiformankind/lets-stop-wildfires-hackathon

Quick Start

Checkout the following resources to get you started

1. HPWREN Camera Datasets

2. Tensorflow based Wildfire Smoke Classifier

3. Tensorflow based Wildfire Smoke Detector

4. FUEGO firecam repo

Wildfire Resources

1. FUEGO Wildfire Detection Slides by Kinshuk Govil
2. Wildland Fire Assessment System
3. The United States Fourth National Climate Assessment Volume II
4. How Wildfire Works
5. Fighting Wildfires

Tensorflow Resources

1. Tensorflow Quickstart
2. Tensorflow Tutorials
3. Install Tensorflow in PyCharm

Detailed Documentation

Understand how does Image Analysis work ?

https://www.tensorflow.org/lite/models/object_detection/overview

RefCard: http://openaccess.thecvf.com/content_cvpr_2017/poster/3562_poster.pdf

Understand how does Docker container work ?

Setup Environment

https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/installation.md

Build the Model

Goal: early detect the wildfire

We provide the following HPWREN camera image sequences as training dataset.

Train Dataset https://tinyurl.com/yy3evlfx

HPWREN Cameras: Smoke/No Smoke datasets We separated the images into smoke and no smoke categories for anyone who is interested in building a smoke classifier. HPWREN Cameras: Bounding Box Annotated Smoke Dataset (Pascal VOC annotation format) We annotated these images using bounding box for smoke detection.

root@c963a4e2e324:/tensorflow# python models/research/object_detection/legacy/train.py --logtostderr --train_dir=/tensorflow/train/ --pipeline_config_path=/tensorflow/train/faster_rcnn_resnet101.config

— Any model exported using the export_inference_graph.py tool can be loaded here simply by changing PATH_TO_FROZEN_GRAPH to point to a new .pb file. By default we use an "SSD with Mobilenet" model here. See the detection model zoo for a list of other models that can be run out-of-the-box with varying speeds and accuracies.

Ref: https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md —-

Subfolder 20170625-BBM-bm-n-mobo >>> contains the image sequences captured by a HPWREN camera when a fire broke out >>> images ~40 mins before and after the ignition of fire 20170625-BBM-bm-n-mobo >> 1498416633_+00240.jpg (first appearance of smoke)

1498413993_-02400.jpg (2400 seconds before detecting the first fire) 1498418793_+02400.jpg (2400 seconds after detecting the first fire) python models/research/object_detection/export_inference_graph.py --input_type image_tensor --pipeline_config_path /tensorflow/train/faster_rcnn_resnet101.config --trained_checkpoint_prefix /tensorflow/train/model.ckpt-1229 --output_directory /tensorflow/inference_graph

label_meta: https://tinyurl.com/y2g2o2jk

You need to build a model that uses previous last seen image sequences (moving window) to predict the presence of smoke in the current image.

root@c963a4e2e324:/tensorflow# jupyter notebook --ip 0.0.0.0 --no-browser --allow-root

MODEL_NAME = '/tensorflow/inference_graph' MODEL_FILE = MODEL_NAME + '.tar.gz' DOWNLOAD_BASE = 'http://download.tensorflow.org/models/object_detection/'

Path to frozen detection graph. This is the actual model that is used for the object detection.

PATH_TO_FROZEN_GRAPH = MODEL_NAME + '/frozen_inference_graph.pb'

List of the strings that is used to add correct label for each box.

PATH_TO_LABELS = '/tensorflow/input/annotations/label_map.pbtxt'

Loading label map Label maps map indices to category names, so that when our convolution network predicts 5, we know that this corresponds to airplane. Here we use internal utility functions, but anything that returns a dictionary mapping integers to appropriate string labels would be fine

category_index = label_map_util.create_category_index_from_labelmap(PATH_TO_LABELS, use_display_name=True)

Detection

For the sake of simplicity we will use only 2 images:

image1.jpg

image2.jpg

If you want to test the code with your images, just add path to the images to the TEST_IMAGE_PATHS.

PATH_TO_TEST_IMAGES_DIR = '/tensorflow/test/shapes' TEST_IMAGE_PATHS = [ os.path.join(PATH_TO_TEST_IMAGES_DIR, 'image{}.jpg'.format(i)) for i in range(1, 4) ]

Size, in inches, of the output images.

IMAGE_SIZE = (12, 8)

============================================================== we will run your program against a new test dataset. The test dataset will have the same directory structure as the train dataset. In the end, your program will output a CSV with the following format. Output File folder_name, smoke_detected(yes/no), image_filename_where_smoke_detected, index of image in the sequence (zero indexed, -1 if no smoke detected) Example: 20170625-BBM-bm-n-mobo,yes,1498416633_+00240.jpg,14 20170625-BBM-wk-w-abbc,no,-,-1

================================================

References Live Collab https://colab.research.google.com/github/aiformankind/lets-stop-wildfires-hackathon/blob/master/Challenge_1B_WildfireSmokeImageClassifierForDemo.ipynb#scrollTo=Ps5obHk8RI20

Local Jupyter

Either

docker container start c963a4e2e3243dd2664b870c5bf1efa21c83bffcb21d97ed479c22dc18056cec Or docker run -e TZ=US/PACIFIC -it -p 8888:8888 -p 6006:6006 --name=wildfiredetection aiformankind/wildfiredetection:0.0.1

docker ps docker exec -it wildfiredetection /bin/bash

Once inside docker: root@c963a4e2e324:/tensorflow# root@c963a4e2e324:/tensorflow# pip show tensorflow

root@c963a4e2e324:/tensorflow#jupyter notebook --ip 0.0.0.0 --no-browser --allow-root http://(c963a4e2e324 or 127.0.0.1):8888/?token=f14d680afa89c6230ecdd7b8a5186d0a3678d78dd90e13ac

http://127.0.0.1:8888/notebooks/object_detection_wildfire.ipynb

V2 Latest Code: https://github.com/aiformankind/lets-stop-wildfires-hackathon/blob/master/wildfire_smoke_challenge_2.md

https://colab.research.google.com/github/aiformankind/lets-stop-wildfires-hackathon/blob/master/Challenge_1B_WildfireSmokeImageClassifierForDemo.ipynb

https://aiformankind.github.io/lets-stop-wildfires-hackathon/wildfire_smoke_challenge_1B.html

=== Analysis of Code

Labelling and Bounding Box https://github.com/heye0507/lets-stop-wildfire-hackathon

Unbalanced Data https://towardsdatascience.com/deep-learning-unbalanced-training-data-solve-it-like-this-6c528e9efea6

Transfer Learning

https://adventuresinmachinelearning.com/transfer-learning-tensorflow-2/

https://builtin.com/data-science/transfer-learning

https://www.tensorflow.org/tutorials/images/transfer_learning

====

https://www.meetup.com/AI-for-Mankind/events/265983290/

Image Augmentation https://github.com/albumentations-team/albumentations/blob/master/docs/hall_of_fame.rst https://www.slideshare.net/VladimirIglovikov/albumentations-fast-and-flexible-image-augmentations

Smoke Detection and Satellite Image Analysis https://fuego.ssl.berkeley.edu/smoke-detection/ https://drive.google.com/file/d/1SECoJiJakNzkDrmczi_iRZIEBAH9yUuG/view http://www.wfas.net/