System and method for the early visual detection of forest fires using a deep convolutional neural network
View Patent ↗A Wildfire Early Detection System uses Aerial Video Clips of surveilled areas—obtained through a Network of Hardware components including UAVs and Tethered Aerostats equipped with a Camera and an AI-enabled Embedded Computer—and an Aerial Training Dataset, digitally combining variations of Smoke-Plume-Clips with variations of Background-Clips in nine different positions in the first frame of said Background-Clips with a relative size to the background calculated by perspective and programmed to follow the background to stay apparently “static” in the same place relative to the background for all the remaining frames of the Clip. A Computer-Vision Algorithm trained with that Aerial Training Dataset is used to recognize early fire Plumes of Smoke in those Aerial Video Clips with use of a Multiplication-free Neural Network for Wildfire detection, an AddNet based Discriminator CNN, a GANs used as both event detectors and smoke-plume scene synthesis and Block-based detection.
1. A Wildfire Early Detection System comprising:
at least one Hardware Network Component comprising one of (a) a Drone; (b) a Tethered Aerostat; (c) a Tower, or (d) a Manned Aircraft, wherein said Hardware Network Component is equipped with:
one or more Cameras comprising: an RGB, Near Infrared (Night Vision) or thermal Camera, wherein the one or more Cameras are configured to provide Aerial Video Clips of surveilled areas; and
an AI-Enabled Embedded Computer configured to train a Computer-Vision Algorithm with an Aerial Training Dataset comprising at least 324,000 Training Clips obtained by combining and digitally placing at least 300 Smoke-Plume-Clips having at least 5 variations of Smoke-Color, Plume-Height and Plume-Width in at least 120 Background-Clips having at least 5 variations of Terrain-on-Clip, Clouds, Height-of-Camera and Visibility, wherein said Smoke-Plume-Clips are digitally placed in nine different positions in a first frame of said Background-Clips with a relative size to a background calculated by perspective and programmed to follow the background to stay apparently “static” in the same place relative to the background for all remaining frames of the Clip;
wherein the trained Computer-Vision Algorithm is used to recognize early fire Plumes of Smoke in said Aerial Video Clips.
2. The Wildfire Early Detection System of claim 1 wherein Camera functions are controlled from a web app and from the Embedded Computer, and wherein said Camera and Said Embedded Computer are powered by solar panels, and configured to operate autonomically 24/7.
3. The Wildfire Early Detection System of claim 2 wherein said Camera and Said Embedded Computer have a Power Consumption between 20W and 60W.
4. The Wildfire Early Detection System of claim 1 wherein at least one of said Hardware Network Components is a Tethered Aerostat made of PVC and filled with Helium, and having an airfoil shape that stabilizes Roll, Pitch and Yaw axes, and having a lift vs drag ratio that tends to infinity so that a Tether and ground form an angle of between 80 and 100 degrees wherein area to volume ratio is of at most 3.5 m 2 /m 3 .
5. The Wildfire Early Detection System of claim 4 wherein said Tethered Aerostat further comprises an Automatic Helium Filling Valve used to refill the Aerostat with Helium in case of low pressure while said Aerostat is on air, wherein said Automatic Helium Filling Valve comprises the Tether, an Upper-Ring acting as a male connector, which is attached to the Tether in such a way that it allows the passage of Helium from the Valve to the Aerostat through an Upper-Pipe, a Non-Return-Valve added to the Upper-Pipe, a Lower-Ring acting a female connector, that is attached to a ground unit containing a Helium Cylinder, wherein the Lower-Ring has a lower pipe which connects said Lower-Rong to said Helium Cylinder and wherein when the Upper-Ring and the Lower-Ring are connected together, they are sealed by a Rubber-Seal to make sure there are no leaks.