CERBERUS

https://youtu.be/206LBOsw6o4

https://github.com/Nasa-Cerberus/Cerberus

Presentation of our project can be accessed by the link below:

 https://docs.google.com/presentation/d/1DcnVKv07jd0XpDVNQgLCySm4WD4RpZV4mdcVa4YzE6g/edit?usp=sharing

Our Project uses MATLAB tools to detect and classify the images as plastic debris or not. We have trained the dataset in deep convolutional neural network to get the desired output. The first goal is achieved that is detection of marine plastic debris ,classify them and effectively segregate to the respective categories.

Our network progressively makes the decisions , for example a marine bot to detect and classify marine plastic debris collected and eradicate it from the ocean body site/ garbage patch.

1. Planning : To develop a suitable network to leverage AI/ML for plastic marine debris
2. Organization: Dataset , software, Information drafting(Marine debris dataset, Marine Debris and its impacts)
3. Command: Knowledge building, accessing papers (Accessed NASA's open resources)
4. Coordination: Joining all the sections together(jotted all the information under one banner)
5. Control: Adopt to the drafted plan (Selected MATLAB)
6. Execute the drafted plan( Implemented our plan , trained our network )
7. Re-iterate -having scope for improvement and achieve better results.

Further , using Faster R-CNN , we quantify the result obtained in process one.

Thereby completing the challenge.

What does our project exactly do?

'Cerberus in Action' is a Deep Learning model that identifies, classifies and segregates plastic debris, garbage patches with high accuracy.

How does it work?

The network is built on Matlab using Deep Convolutional Neural Network where it gives appropriate accuracy of the detected image

1. Feeding and processing of the image Dataset

The images dataset is first fed into the Deep Convolutional Neural Network, which is processed into 2 segments - training and testing dataset. 

2. Testing and Training the dataset

After the processing of the dataset, we enter the learning rates, the number of the iteration the data must move in the layers to increase the training accuracy rates. Later we set the number of epochs and the frequency of iterations for the cycle of repetitions of the data for training.

3.Results:

According to the output of the model, we can deploy the robot to the particular location for any task. Hence we can pinpoint the machine to reach the place and help in the clearance of the debris.

What are the benefits of project ?

1. Easy to understand and implement.
2. All the tools in this project are widely available.
3. The dataset in this project is open source.
4. Our model gives highly accurate results
5. Results are visualized in a simple yet effective manner
6. Pin point precision of clearance of marine debris.

Future Scope:

Develop a network model for better accuracy and for the deployment for live feed of data with better hardware specifications.

Tools we used:

• Matlab 
• Deep learning and image processing tools box on Matlab
• Python editor

We used the following open data of NASA:

https://coast.noaa.gov/states/fast-facts/marine-debris.html

https://marinedebris.noaa.gov/

https://sos.noaa.gov/catalog/datasets/marine-debris-garbage-patch-experiment-drifters-and-model/

https://www.epa.gov/sites/default/files/2020-10/documents/marinelitter_booklet_10.16.20_v10epa.pdf

https://www.doi.gov/ocl/marine-debris

https://marinedebris.noaa.gov/

The representation of data in pictorial graphics, inspired our team to take reference and use these sources for our challenge .

Aspire, Inspire , Innovate and Achieve

form the blocks of Space Apps Experience.

Learning is a life process..

"One wheel alone does not turn and keep the cart in motion" -Kautilya, Arthashastra

Team Work and the spirit of critical thinking and influence over the ideology of engineering solution through persistence and consistency.

Marine Debris is a global problem. History of the problem - is because of us(humans).

Education being the key.

We are the solution, a small change towards the outlook of living, each one of us with the power of reduce, reuse and recycle can achieve the goal.

Leveraging AI/ML for plastic marine debris -the challenge , Engineering skilled solutions to hack the real world problem, a small contribution if all can contribute to the social well-being , we can achieve environment sustainable developments.

We approached the challenge through deep learning mechanism to create visualization database based on AI/ML algorithms that will aid in classifying and detecting these plastics .

Understand the potential advantages and limitations of utilizing AI/ML algorithms to classify plastic pollution.

Challenge and set backs - to acquire the proper image dataset to test and train the network.

Formation of a network in a platform and get the required output.

Linking of clusters of the code and get a structured format of output.

Better hardware specifications for training and processing data.

Patience and team work was the key to resolve all our challenges.

We would like to thank the organizers who kept the chain of continuity even during the greatest challenging time amidst the global pandemic.

https://marinedebris.noaa.gov/

https://coast.noaa.gov/states/fast-facts/marine-debris.html

https://marinedebris.noaa.gov/

https://sos.noaa.gov/catalog/datasets/marine-debris-garbage-patch-experiment-drifters-and-model/

https://www.epa.gov/sites/default/files/2020-10/documents/marinelitter_booklet_10.16.20_v10epa.pdf

https://www.doi.gov/ocl/marine-debris

Math Works

Google image data marine debris

Algorithms to organize data

Faster R-CNN

#marinedebris #ocean #cerberus

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