Towards Conservation of Sea Turtle Population using Effective and Scalable Object Detection



Sea Turtles


Sea Turtles play a quintessential role in the operation and maintenance of marine ecosystems. Their presence not only aids in the nourishment of the coral reefs but also in the survival of other marine creatures. Lately, the population of sea turtles has been adversely affected due to human interference in both direct and indirect forms. Every month, several hundreds of turtles are found helplessly succumbing to either injuries or poisoning caused by commercial activities. This has led to various marine conservation organizations making sea turtle conservation a priority among their operations. Although environmentalists and volunteers across the world are undauntedly collaborating for rescuing sea turtles, their assistance is limited by the delay in the detection of sea turtles in traumatic situations. In this work, we propose an automated sea turtle detection system for locating sea turtles in distress and alerting the nearest concerned environmentalists. At the core of our system, is a highly scalable object detection algorithm based on YOLO-v5. Our experiments demonstrate the effectiveness of single-stage object detection algorithms in the identification of turtles in diverse environments with a mAP@.5 of .99 and mAP@.5:.95 of 0.92.

Who is this talk for?

  • Anyone with a passion to learn new skills and contribute to environmental conservation.

  • Anyone who is wondering how to create a social/environmental impact leveraging Machine Learning.

  • Anyone who is keen to learn about the real-world applications of deep learning

  • Anyone who is starting or planning to start their Data Science careers and wishes to explore the possibilities.

Talk Outline:

  • Introduction [1 min]

  • Objectives [1 min]

  • Speaker Intro [1 min]

  • Motivation [2 mins]

  • The need for taking responsibility for environmental conservation.

  • Understanding how practitioners of technology can make an impact.

  • Introduction to our work [2 mins]

    • Depleting number of sea turtles and its global impact.
    • Relevant advancements in Deep Learning.
  • Some Background Information [2 mins]

    • What is object detection?
    • Why use neural networks?
    • Popular tools and algorithms.
  • Diving into our work [10 mins]

    • Goals
    • Curation of data points.
    • Pre-Processing
    • Our algorithm
    • Experiments
    • Results
    • Inference
  • Concluding Remarks and Future Avenues [1 min]

  • QnA [5 mins]


  • Basic Understanding of Machine Learning and Neural Networks
  • Some idea of Convolutional Neural Networks and Object Detection Foundations
  • Urge to learn and apply new skills to improve the world around us

Speaker Info:

Aditya Jyoti Paul

Aditya Jyoti Paul has been active in the research space, especially in Computer Vision and Image Encryption for the past three years. He has 10 years of experience working with Python and a love for simple code to do things efficiently. Here are some of his experience and career highlights, in various spheres:

Research Achievements

  • Published Researcher with 5+ international Journal and Conference papers, check out his papers on the author's page or on Google Scholar and ArXiv.
  • ML Research Intern with Samsung Research Institute Bangalore on Bengali Handwriting Recognition and was awarded the Certificate of Excellence for the same.
  • Research Assistant on a SPARC project with UC Davis and IIT Kharagpur, funded by the Ministry of Human Resource Development, Govt. of India on early diagnosis of Non-Proliferative Diabetic Retinopathy.
  • Two-time University Gold medalist for Best Research Paper in 2019 and 2020.
  • Founded Cognitive Applications Research Lab, and in just two years
    • Spearheaded research with 25+ international journal and conference publications from the team since then
    • Guided the team to 3 gold medals besides various other research awards
    • Conducted 10+ workshops on research methodology, and ML/CV and Human-Machine Interaction
  • Headed Research and Development at IEEE SRM Student Branch in SRM University.
  • Reviewer for Heliyon, an Elsevier journal, part of the Cell Press Family.

Teaching Experience and Community Efforts

Smaranjit Ghose

Smaranjit is an incoming Masters in Data Science graduate student at Columbia University. He has a diverse medical computer vision research experience spanning for over 2 years. Till date, he has impacted over 7k students across 15 open source programs and numerous national and international hackathons. His research interests include XAI and applications of deep learning to conversational agents and healthcare. He is among the top 10 most active GitHub users from India. Previously he was worked on multiple freelance machine learning projects as well as being associated with UCL, IITM, etc as a research intern.

Speaker Links:

Know more about Aditya Jyoti Paul (online alias: phreakyphoenix) at:

Check out Smaranjit Ghose's work online at

Section: Data Science, Machine Learning and AI
Type: Talks
Target Audience: Advanced
Last Updated: