Fivemind

Team Name: 
Tinman

Fivemind is a land management and bushfire mitigation tool. Using Satellite imagery, the tool automatically detect changes in vegetation and human built structures within specific terrestrial areas. These change observations are combined with terrain and climate data to produce a spatial risk assessment. Land Managers and Fire Authorities can then capture and evaluate higher resolution imagery of areas of concern by deploying UAVs or Drones over flight paths generated by the tool.

Less bushfire and greater crop yields can be achieved by monitoring vegetation more closely – computer vision and machine learning processes are generally more precise than the traditional masking methods. NVDI provides general trends across large areas, where as computer vision can identify specific places of change and group associated causes.

Fivemind consists of a satellite imagery pipeline and a web app interface. The pipeline called Spacelab processes imagery bands made available by Sentinel-2 multi-spectral instruments. It's main purposes is to create NVDI and EVI vegetation masks, NRG false colour images and gradient histograms. The role of the web interface is to easily identify risk level by colour code and create Drone missions across land areas of interest.

The project uses machine learning and computer vision technologies in a number of ways. A TSP variation of a Self Organising Feature Map is used to calculate near-optimal Drone flight paths from selected way points. The map algorithm is also used to group land areas with similar feature values. Histogram of Gradient descriptors and radial RANSAC detect vegetation and surface changes over time.

The generated flight path is downloadable and compatible with DJI flight quad copters and drones.

screenshot:https://postimg.org/image/k9uam8dnb/

Used Datasets: 
Dataset Name: 
Sentinel-2 Satellite MultiSpectral Instrument Data
Publishing Organisation/Agency: 
European Space Agency
Jurisdiction of Data: 
International
How did you use this data in your entry?: 
Bands 4, 3, 2 and 8 are used to make vegetation masks and false colour images. Band 8 is used in a computer vision process to detect changes over time.
Dataset Name: 
Bureau of Meteorology weather stations - All Australia
Publishing Organisation/Agency: 
Bioregional Assessment Programme
Jurisdiction of Data: 
Australian Government
How did you use this data in your entry?: 
Automated finding nearby weather stations for a machine learning process.
Dataset Name: 
Precis forecast – South Australia
Publishing Organisation/Agency: 
Bureau of Meteorology
Jurisdiction of Data: 
Australian Government
How did you use this data in your entry?: 
Fetched current weather conditions for use in a machine learning process.
Dataset Name: 
Bushfire Management Planning Data - Identified Assets at Risk
Publishing Organisation/Agency: 
SA Country Fire Service
Jurisdiction of Data: 
Government of South Australia
How did you use this data in your entry?: 
Geocoded the asset street addresses and used the risk level in a machine learning process.
Dataset Name: 
Geoscience Australia, 3 second SRTM Digital Elevation Model (DEM) v01
Publishing Organisation/Agency: 
Bioregional Assessment Programme
Jurisdiction of Data: 
Australian Government
How did you use this data in your entry?: 
Elevation data is used in a machine learning process.
Event Location: 
Adelaide