High-Level Project Summary
We created a Web Dashboard using streamlit to Track Marine Debris with Object Detection Model (YoloR). Users can upload video data and measure the number of plastic objects in the video as seen on the interface. It resolves the problem of estimating regional pollution levels using satellites by using videos of that region instead. This is crucial in timely intervention and identifying the regulations needed to keep the plastic levels in check. Region with the most/least plastic pollution can automatically be identified. This technology makes the need to feed information redundant as well and can be interpreted and used by all individuals regardless of technical expertise as is at present.
Link to Project "Demo"
Link to Final Project
Detailed Project Description
Our project is a Web Dashboard that uses the power of Object Detection to track marine debris from video data. We can upload a video from our local machine on the Web Dashboard and it will detect and show the tracked plastic objects in the video automatically by colour coding them. The certainty of the detection can also be varied manually which changes the number of detected objects. It works by using an Object Detection Model called YOLOR and making inferences from the input video and showing output on the Web Dashboard. It benefits us greatly if we want to detect the plastic pollution levels of a region and take measures accordingly which is cost effective and rapid compared to data analytics using satellites which need to be processed before being acted upon. We can collect video data of that region and use our Web Dashboard to infer the region's debris levels to localise the initiatives. We hope to use this Web App to get data in Real-Time from all around the world and replace the currently available debris tracker that depends on manual processes. Through this model, we also hope to improve the understanding of environmental systems of the global population who due to lack of knowledge in translating scientific datasets such as satellite images restrict themselves from researching about related information. We trained the YOLOR model on DeepPlastic Data on Google Colab and used Streamlit to develop the Web Dashboard. Our coding language was Python.
https://docs.google.com/document/d/1j36kriPfvBXW-snmkb5vvlS97LlmzRXAMnil-PsUmYk/edit?usp=sharing ( highlights our research and solution)
Space Agency Data
When brainstorming the tool, we were intially confused about the mechanism of AI/ML models and how they could be used to report marine debris, but, referring to the satellite images provided by NASA in addition to the current intiatives being undertaken by JAXA, ESA and CSA helped us identify how to use technology to detect plastic waste. The team heavily relied on the space agencies' sites to understand the concept of marine debris accurately before researching their programmes and technologies that are currently serving this purpose. By referring to these and a satellite footage specifically provided by NASA which displays current movement and plastic flow was key in understanding the periodic trends. However, it also led us to the conclusion that despite the data analytics, the detection was only focused on specific regions which prompted us to question " Can we develop something that can measure marine debris in any and all water bodies- from the smallest of streams to the Pacific Ocean?" which helped us code for own solution that provides results depending on the input given to it which is thus customised and personalised. The web Debris Tracker by National Geographic and Morgan Stanley served as a springboard and inspiration for our ML model, in understanding the fundamentals required for a data analytics website.
We used open-source DeepPlatics Data to train our YOLOR model.
Hackathon Journey
Though it's hard to describe the multitude of emotions that we experienced through the course of this hackathon, if we were to put it as a mathematical function, it's similar to a sine curve. We started off at point zero as strangers, showed a gradual upward trend during our research, reached a peak when we were clear about the objective and back to zero when brainstorming. But the path didn't stop and only made us more determined to explore the scope of AI/ML in the data analytics using our solution. Choosing the marine debris challenge was a much debated choice due to the various aspects involved. However, perhaps the comfort of designing a solution for the water was something that sunk in all of us as we chose this topic and discussed our solutions. Water pollution and marine debris is an area that has long been discussed but rarely acted upon, which perhaps was one of the motivating factors in designing something that appealed to the masses and implored them to act by improving their scientific literacy using a much simplified way of presenting data. The biggest challenge we faced was with presenting our model. Due to the mechanism used to develop it, the link generated is only temporary and can be shared with a single individual at a time. This very revelation set panic in all of us as we wondered what to present, however, it is then that we thought that instead of providing a link, a demonstration of the functionality can be shown instead which has proved to be a viable solution for our team. The entire team would like to thank their parents who have been a part of all our moments in this hackathon journey by providing us with suggestions when we were unable to think of those as well as being our emotional pillars of strength.
References
DeepPlastic Data, Google Colab, Deepnote, Streamlit, YoloR
https://debristracker.org/data
https://www.esa.int/Enabling_Support/Preparing_for_the_Future/Discovery_and_Preparation/The_Discovery_Campaign_on_Remote_Sensing_of_Plastic_Marine_Litter
https://www.forbes.com/sites/allenelizabeth/2020/04/27/how-satellites-and-machine-learning-are-being-used-to-detect-plastic-in-the-ocean/?sh=5c278cd24fd1
https://www.asc-csa.gc.ca/eng/satellites/everyday-lives/monitoring-and-protecting-our-ecosystems-from-space.asp (understanding the importance of satellite images in ecosystem regulation)
https://www.nasa.gov/feature/esnt2021/scientists-use-nasa-satellite-data-to-track-ocean-microplastics-from-space
https://sos.noaa.gov/catalog/datasets/marine-debris-garbage-patch-experiment-drifters-and-model/
https://www.bbc.com/news/science-environment-53521001
https://www.nytimes.com/2020/10/07/world/australia/microplastics-ocean-floor.html
https://www.nytimes.com/2008/06/22/magazine/22Plastics-t.html
Tags
#plastic,#debris,#sea,#ocean,#ai,#ml,#detector,#trash, #PODS
Global Judging
This project has been submitted for consideration during the Judging process.