The ability to detect man-made objects in vegetation would aid many fields of industry including – but not limited to – search and rescue as well as agriculture. The ability to locate man-made objects such as damaged modes of transportation, camping or hiking equipment, parachutes, etc. in said vegetation aids the chances of locating missing individuals involved in these cases. While much imagery of these objects exists, imagery of them in varying degrees of repair within vegetation is limited at best making standard machine learning detection methods difficult. In such cases, focusing on the vegetation may enable the detection of such objects through machine learning methods where imagery is limited. In agriculture, farming equipment may require repair while in the field where it is easily possible to misplace equipment in vegetation. This lost equipment could damage other farming equipment while work is being done. Locating lost equipment will help limit further damage of farming equipment saving not only money, but time and production as well.
For this project, the team will be given access to aerial multispectral imagery containing a variety of man-made objects located in vegetation. The goal of this project is to develop a method which can detect these random man-made objects using machine learning and computer vision techniques while investigating the benefits of multispectral data to solving this problem. The machine learning field of focus for this problem is that of anomaly detection.
At Awesense we’ve been building a platform for digital energy, with the goal of allowing easy access to and use of electrical grid data in order to build a myriad of applications and use cases for the decarbonized grid of the future, which will need to include more and more distributed energy resources (DERs) such as rooftop solar, batteries as well as electric vehicles (EVs).
Awesense has built a sandbox environment populated with synthetic but realistic data and exposing APIs on top of which such applications can be built. As such, what we are looking for is to create a collection of prototype applications demonstrating the power of the platform. Given the synthetic nature of the dataset we can make available, this would be more of a “deliver a method (and implementation of it)” type project than a “deliver insights” type project.
This involves coding some analyses and visualizations on top of said data and APIs. It would require good data wrangling + statistics + data visualization skills to design and then implement the best way to transform, aggregate and visualize the data for the use case at hand (see below). The data access APIs are in SQL form, so SQL querying skills would also be required. Beyond that, the tools and programming languages used to create the analyses and visualizations would be up to the students. Typical ones we have used include BI tools like Power BI or Tableau and notebooking applications like Jupyter or Zeppelin combined with programming languages like python or R.
If the participants don’t have any electrical background, we can teach enough of it to allow handling the given use case. For this year’s project, we have chosen a use case entitled “EV charger capacity study”. At a high level, this entails determining how many new EV chargers could be installed in a particular portion of the electrical distribution grid without overloading the capacity of the grid infrastructure at that location. This would allow distribution grid planners to determine whether or not to approve requests for “interconnection” of EV chargers; it would also allow them to plan for needed infrastructure upgrades to support more EV chargers in the future.
Local labour market information (LMI), including employment, unemployment, and participation rates, are important for planning activities conducted by municipal and First Nation governments, service providers, and many more organizations. Census of Population (Census) data, published by Statistics Canada (StatCan), provides LMI at the Census Division (CD) and Census Subdivision (CSD) levels. However, the Census is only conducted every five years. This project asks participants to build a model which estimates intercensal LMI at the CD and CSD levels.
LMI are not publicly available at the CD and CSD levels of geography for years when the Census is not conducted. StatCan’s Labour Force Survey provides data on a monthly basis. However, StatCan does not publish Labour Force Survey data at the CD or CSD levels. As such, LMI at the Economic Region level are often used in the place of intercensal data on CDs or CSDs. Figure 1 shows the CSD of the District of Kitimat, as well as the larger CD and Economic Region it is within, as an example illustrating how Economic Regions may not serve as good proxies for CSDs.
Residential radon progeny exposure is the second leading cause of lung cancer, after smoking. The two main radon isotopes are Rn-222, which is part of the uranium-238 decay chain, and Rn-220, also called thoron, which is part of the thorium-232 decay chain. There is currently much interest in the Rn-220 contribution to radon progeny exposure, which has so far been largely ignored. (Rn-220 has a relatively short half life and usually decays before it reaches the living areas in a house and it usually doesn’t show up in radon measurements. But, Rn-220 has a longer lived decay product which does reach living areas and contributes to radon progeny exposure. It can even exceed the Rn-222 contribution.)
Environmental Instruments Canada (EIC) produces a Radon Sniffer (see https://radonsniffer.com/ ), which is used by radon mitigators and building scientists to find radon entry points. These sniffers currently assume all radon is Rn-222. See the appendix for a more detailed description of how the sniffer works. We want to extend the functionality to Rn-220.
In a 2020 M2PI project, we came up with a dedicated sampling and counting sequence and developed the math to determine how much Rn-222 vs Rn-220 was in the air. This report is available to the team.
In this project, we wish to develop a method by which we can determine the presence of Rn-220 in the air, while the Radon Sniffer is continually sampling air and without having to run a dedicated thoron measurement sequence.
The stormwater infrastructure in many cities is facing challenges as the climate changes, rainfall patterns change and sea levels rise. To meet these challenges, cities are installing green infrastructure (GI) systems to absorb and retain rainfall where it lands and reduce sewer overflow. Across North America alone, cities are investing over $56.2 billion dollars in green infrastructure. By doing so, communities are becoming more resilient against climate change and are achieving environmental, social and economic benefits. Novion’s Climate Intelligence Platform helps cities monitor their green infrastructure for performance optimization, regulatory compliance, and maintenance.
Once green infrastructure has been constructed, monitoring the changes in water level after a rainfall event is required to verify its performance and net impact. This project will investigate aspects of this monitoring.
Long-distance dispersal of insects in fast moving air currents is increasingly recognized as an important driver of their dynamics at a landscape scale. Moreover, this type of dispersal has important implications for forest and agricultural crops impacted by insects. Because the detection and tracking of populations of flying insects remains challenging, it is rarely possible to determine where insects originated after they have dispersed long distances. Data from weather radars designed to detect precipitation may be useful tools for gaining insight into insect long distance dispersal because insect bodies and rain drops are often similar in size. Thus, within radar scans there is potential to quantify the density of insects departing to start long-distance dispersal as well as the movement trajectories of swarms–at least until they pass beyond the range of the radar. However, because Doppler weather radars are extremely sensitive and capable of detecting water vapor in clouds, it can be difficult to distinguish between potential insect signals and weather signals even when it is not ostensibly raining.
This project has two distinct objectives. First, we will investigate classification of radar images, based on motion, to identify insect swarms. Second, we will develop a mathematical model and numerical simulations to more easily distinguish distinguish meteorological and biotic signals.
Perfit is currently working on a virtual fitting room app that allows online shoppers to get a virtual preview of the fit of garments in their cart. The virtual fit is accomplished via simulations of cloth interacting with a customer avatar through particle collisions. Specifically, each piece of cloth is modeled using a mesh of interconnected particles. The company has made progress on collisions between cloth and avatar. However, there is a need for collisions between cloth and cloth that has remained a significant challenge for the company. For example, the capability for a virtual garment to interact with itself via cloth-on-cloth collision would allow improved wrinkle quality, and garments with pleats, such as a pleated skirt. The technical challenge facing the company is to enable the handling of cloth-on-cloth collisions in near real time. In a mathematical sense, an algorithm for cloth-on-cloth collisions would need to be developed that minimizes the number of operations required (e.g., avoids brute force particle searches) while maintaining sufficient accuracy.
Topics in geometry, physics, computational methods, and computer graphics are expected to arise while working on this problem. The preferred implementation of the solution is the Fortran programming language in order to facilitate integration with the existing physics engine. During the workshop, meshes for both customer avatar and cloth will be made available through GitHub for testing.