High-Level Project Summary
The MarinaWatch mobile application utilizes deep learning, remote sensing devices and cloud computing technologies. Remote sensing device like drones will be used to film footage of the ocean, coastal and waterways and these footage will then be processed by deep learning models on cloud to produce statistics and visualization of marine debris pollution in filmed locations. This helps with visualizing and understanding how polluted every part of the ocean, coastal and waterways is. This data allows for understanding the cause of this pollution of marine debris and preventing further pollution.
Link to Project "Demo"
Link to Final Project
Detailed Project Description
Overview
MarinaWatch will be developed using python programming language and the Django framework for both frontend and backend of the application. As for storing the application data, SQL will be used for creating and managing the table and stored procedures at the backend of the system. The nature of the Django framework allows the application developed to be coded in different modules for each function or features, so each module will have its own directory to store the respective python files. Hence, this allows the modules or features of the application to be tested easily or to be reusable for any future applications. So there will be a few modules that will be included in the application which are user dashboard, drone footage monitoring, data analytic, statistical data for locations and data sharing to other platforms.
The MarinaWatch system uses drones as the main device for capturing footage of ocean, waterways and coastal areas. The captured footage will be sent to the cloud server, which will be the Amazon Web Service (AWS) platform for storage and processing after going through the IoT gateway using Long Range (LoRa) communication protocol. For hosting the MarinaWatch application, a serverless architecture will be used, specifically the AWS Lambda. Serverless computing is a cloud computing paradigm that provides its user the convenience of hosting applications on the cloud service without worrying about the underlying server infrastructure (Sadaqat et al., 2018). An example of serverless systems by Amazon is AWS Lambda, which are capable of scaling and evolving the application without users putting much effort in periodically patching the server for updates and maintenance.
There will be two deep learning models that will be developed, one for detecting marine debris using the VGG16 transfer learning model and other for quantifying the marine debris volume using a CNN binary classification model. Both models will be developed and hosted on the Amazon machine image infrastructure. The models developed on the AWS infrastructure will carry out all the analytical jobs such as detecting and quantifying marine debris from footage. The model will also utilize the captured footage stored in the cloud storage for training whenever there is new footage captured using the AWS Lambda trigger. The communication from the MarinaWatch mobile application will be secured with secure socket layer (SSL) protocol to ensure all communication between the mobile device and the cloud server are encrypted. The architecture works where the user makes a request for detection or quantification of footage, the request will be sent to the cloud server and passed to the deep learning models. The model will then perform the prediction from the provided footage and return the predicted results back to the mobile application.
Space Agency Data
-
Hackathon Journey
Brainstorming
Before we started doing the challenge, a quick brainstorming session was done online using a messaging application, WhatsApp. All the group members contribute by typing down their ideas into the chat and the ideas are then filtered through votes. In the end, the idea with the highest votes ended up as marine debris detection using artificial intelligence.
Document analysis
To find out more about how feasible the idea of detecting marine debris using artificial intelligence is, several literature readings on existing marine debris detection methods were conducted. Using online databases such as iEEE, ScienceDirect and ResearchGate, much of the work was found and read. After doing the research, we had come to conclude that deep learning and unmanned aerial vehicles are the most commonly used technology for identifying and quantifying marine debris as of the current trend.
Observation
Furthermore, observation methods were also used to find out what requirements can be added to the proposed system. By searching for videos and articles that are related to how people survey, clean, monitor and preserve the marine ecosystem, some requirements can be deduced so that it can solve the problem faced by these related people.
References
Sadaqat, M., Colomo-Palacios, R., & Knudsen, L. (2018, January 1). Serverless computing: A Multivocal literature REVIEW: Semantic Scholar. undefined. Retrieved September 24, 2021, from https://www.semanticscholar.org/paper/Serverless-computing%3A-a-multivocal-literature-Sadaqat-Colomo-Palacios/57258323329b29a86b1a83a77ff6ba8bd1bf3d96.
Tags
#deeplearning #drones #mobile application # cloud computing
Global Judging
This project has been submitted for consideration during the Judging process.