Pseudo Mode

https://docs.google.com/presentation/d/1uIO8rqFBV6ch5huO-jjh9HHBWQHem9tv/edit?usp=sharing&ouid=103163715339672625996&rtpof=true&sd=true

https://github.com/blackprince001/NASA-2021-Space-Apps-Challenge

Our project is primarily focused on identifying marine debris in waterbodies. From our research we found out that much of the marine debris research focuses on floating plastic debris, but it is important to recognize that only approximately half of all plastic is positively buoyant, i.e., it floats. The rest of them sink underwater and likewise remain undetected. Moreover after some amount of time in the ocean, floating plastic debris may become sufficiently fouled with biological growth that the density becomes greater than seawater, and it sinks. Hence our prototype would primarily focus on detecting the surface concentration of marine debris as well as the debris that's has been living underwater. 

Through our model, we would also probably calculate the age of the plastic (i.e how long it has been on the sea).

Basically our project is divided into two section : the hardware (the model that's structured in the form of a turtle and likewise would be released on to the water bodies) and the software (the server and user end part that would update the map information regarding the concentration and hotspots of marine debris)

Hardware (The turtle) :

How we built the hardware model:

Image : Hardware model

• Building the Hardware model  

Components for the eyes:  

  -Raspberry Pi Camera Module 2 - 4 pieces for the anterior part of the turtle and for the pivot areas of the turtle propellers

  -White LEDs for illumination under the sea.

Image : front view of model

• Motherboard(shell):

-Raspberry pi 4 - running the TensorFlow model and sending feedback through saildrone. Covered by a 3D printed water resistance shell-like case.

Image : Motherboard

• Wheels  

- Servo motors : mobility in the ocean using tracked distance mapping and pathfinder 

Image : Wheels

Modeling of the wheels will have small flaps inclined at the vertices of the perpendicular bars that hinge the wheel into shape. 

This will serve to displace water when the servo motors are activated, causing the wheel to rotate and move Herman the turtle.

But for the lateral character movement, Herman is going to rely on servo motors controlling the limbs. The use of both limbs to steer left and right in water will enhance faster completion of tasks in a defined automated area.

Mode of Operation

• Our hardware is developed through an AI route-mapped robot model that uses an OpenCV trained image data Tensorflow model to detect, monitor and quantify plastic waste under the ocean and report instances of found material as api queries to server. 
• Live frames from the raspberry pi camera modules will be analyzed for signs of plastic waste from trained data on how plastics in the ocean will look like. Output data will then be recorded and saved on the raspberry pi model B storage device.
• The Quantified data will include picture frames of debris in the ocean ,which will be classified by the model and update it to our server.
• Our hardware model will be Integrated with the Saildrone(NASA) platform api for the sole purpose of sending parsed data from observational routines in the ocean to a server for live map showcase and analysis

Software(Server/Database/User End):

How we built our software model?

• Initially we started with creating a web based UI for our web app. While using our webapp, our end user would initially encounter our home page, which has a navigation bar, to redirect you to pages that contain our live map which shows the hotspots of the shores where marine debris were identified, virtually reality based hardware model, documentation (giving a brief insight about the available datasets) and the creators of this project. The UI was built purely through HTML,CSS and JS. 
• We developed our live map through the available open datasets, that contained the plastic wastes collected/identified by individuals(localites/organization) in the seashore and henceforth performed visualization and analysis on them. 
• The live map was developed using ArcGis Software tool and the visulaization for the datasets were made using Jupyter Notebook. 
• For object detection we used tflite model(single shot detectors).
• Furthermore we would be integrating multiple datasets from the NASAs portal, so that it can contribute in finding the hotspots of the marine debris and also develop a roadmap for the hardware model. For eg : the sentinel data obtained from esa that shows the chlorophyll concentration in the water bodies, can help us predict which zones are likely to have wastes. Greater concentration of chlorophyll, means there is a greater chance of the debris beingidentified. Also the data from NASA can be used to measure the oceanic temperature and the green house gas rate and likewise predict the chances of marine debris being there. 

Image : Signals from hotspots

Image : Computer Vision model trained on images of waste and debris scaling off desired output

image : Data Visualizations from NASA datasets and likely data to be obtained by robot

• https://collections.eurodatacube.com/chl_water_quality_saturday/
• https://gml.noaa.gov/ccgg/carbontracker-ch4/documentation.html
• https://collections.eurodatacube.com/s5p-no2-tropno-daily-check/
• https://search.earthdata.nasa.gov/search?fst0=Oceans
• https://nsidc.org/data/nsidc-0116?_ga=2.99776728.773505565.1633276918-1029359020.1633146821
• https://www.marinedebristracker.org/data

• Initially we used the datasets from noaa that contained the data on the debris waste updated by the app users and likewise we optimized and built a precise map that shows the sea shore hotspots of marine debris along with a detailed explanation of the type and concentration of the waste found. 
• We would be integrating multiple datasets from the NASAs portal, so that it can contribute in finding the hotspots of the marine debris and also develop a roadmap for the hardware model. 
• The sentinel data obtained from esa that shows the chlorophyll concentration in the water bodies, can help us predict which zones are likely to have wastes. Greater concentration of chlorophyll, means there is a greater chance of the debris being identified. Also the data from NASA can be used to measure the oceanic temperature and the greenhouse gas rate and likewise predict the chances of marine debris being there. 
• The sentinel-5P data obtained from esa would allow us to monitor the tropospheric nitrogen level over the ocean and derive assumption about the concentration of the waste. The same applies for the methane level tracker datasets obtained from Global Monitoring library. We would combine these two datasets along with other datasets from NASA such as the NASA Ocean Biogeochemical Model assimilating satellite chlorophyll data(that measures the oceanic temperature), the polar path finder data from nsdic that monitors the motion of the ice (as Sea ice accumulates more than its fair share ) so as to monitor where the plastic waste is coming from and henceforth build a roadmap for our hardware model, the information from airsar alongtrack giving us an idea about the oceanic wave speed and direction.
• We would be integrating all these datasets and henceforth derive a predictive model that would predict the concentration and hotspots for marine debris and build a road map for our hardware model.

What inspired our team?

Out of all the problem statements we skimmed, leveraging ai/ml for marine debris caught our eye. We realized that there weren't any map that shows you the concentration of wastes dumped in the water bodies and due to this most of the people were unaware of the fact , as to how badly the marine debris is affecting the organisms within as well as we human beings. So we wanted to communicate with the mass about this buried and serious issue of marine debris by building a live map that would monitor the amount of plastics and their age; hence we chose this problem statement.

Challenges Faced:

1) it quite difficult design the model

2) I t was quite difficult to build this in a limited amount of time that I learned from this challenge

Challenges faced while building the Hardware model: 

• We faced a couple of challenges for selecting the components for hardware as we wanted to make it as much as environmental and pocket friendly. 
• Choosing the algorithm was a difficult task and we had to resort to a heuristic approach 
• During code execution we faced a couple of problems as our systems didn't provide a very high computational speed . 
• Sometimes our model gave wrong results or very less accuracy, for eg: when our model was shown the picture of a plastic bottle it detected it as a plastic bag

Challenges faced while building the software model :

• It was quite hard to find accurate datasets as there were multiple options available
• Cleaning the datasets took a long time as the datasets were huge 
• Direct datasets regarding the marine debris weren't available. So had to resort other way and perform analysis and assumptions from the existing datasets. 
• The software tool we are using to develop a map that said the concentration of marine debris on the shores only supported importing a limited amount of data

How did our team resolve these challenges ? 

• Primarily we approached the SMEs and thankfully they were able to clear most of our doubts 
• Most of our team members come from diverse tech background. So in a nutshell, we got to learn a lot from each other and also were able to help the other when they were stuck at some instance. 
• We also expanded our idea, by considering and putting forward each one of our thoughts . Through this approach we were also able to resolve the problem we were facing. For eg: when one of our team member had difficulty in visualizing the data , the other would work along with team member and together they would solve the problem , quickly as well as precisely

• https://www.epa.gov/trash-free-waters/toxicological-threats-plastic
• https://earthdata.nasa.gov/learn/sensing-our-planet/unwelcome-enrichment-in-the-arctic
• http://www.ismar.cnr.it/
• https://ourworldindata.org/plastic-pollution
• https://slides.ourworldindata.org/plastic-pollution/#/title-slide
• https://www.condorferries.co.uk/marine-ocean-pollution-statistics-facts
• https://www.kaggle.com/sansuthi/plastic-input-into-worlds-oceans/discussion/249222
• https://www.w3schools.com/, www.google.com, nasa.gov/sites/default/files/atoms/files/fy20_adc_guide.pdf

#marine #ai #ml #debris #model #algorithm #code

This project has been submitted for consideration during the Judging process.