Project Plan

Project Plan

Project Title: Mechanatee: A Biomimetic Exploration and Environmental Monitoring Device

Team Members:

Name Email Department

Alexandra Staros astaros2021@my.fit.edu Computer Science

Cannon Bogar      cbogar2021@my.fit.edu  Ocean Eng.

Laura Mace      lmace2021@my.fit.edu  Ocean Eng.

Aiden Calenda     acalenda2022@my.fit.edu Ocean Eng.

Eden Stroman      estroman2021@my.fit.edu Ocean Eng.

Jackson Clendenin jclendenin2021@my.fit.edu Ocean Eng.

Jacob Warner  jwarner2021@my.fit.edu Ocean Eng.

James Lacey  jlacey2021@my.fit.edu Mechanical Eng.

Faculty Advisor/Client:

Dr. Stephen Wood swood@fit.edu Ocean Eng. Program Chair

Date(s) of Meeting(s) with the Client for developing the plan:

September 2nd, 2024

October 2nd, 2024

November 4th, 2024

Dates may be added to the schedule depending on the needs of the project and the discretion of the client/advisor.

Goal and Motivation:

The goal for Mechanatee is to provide a clear and concise way to monitor populations living in Floridian waters. Designed to be an underwater remotely operated vehicle, Mechanatee is designed to have an unobtrusive relationship with its environment, being careful to not to disturb wildlife while collecting research data.

Approach:

Species Recognition and Classification

The main functionality of Mechanatee will be the ability to permit Users to recognize and classify varying species of aquatic life that are native to Florida. The User can accurately identify different species in real-time as the vehicle navigates through underwater environments. Users can rely on this feature to automatically catalog the species observed during a mission or exploration. This feature would be particularly valuable for marine biologists, conservationists, and environmental agencies, providing them with reliable data in regards to biodiversity and research.

Automated Data Collection and Counting

Beyond recognizing species, the User can automatically count the number of individuals of each species it encounters using a high-definition camera. This feature is beneficial to Users who need to monitor population sizes or track changes in species distribution over time. The system will log each recognized species along with its count and timestamp, allowing Users to analyze trends and make informed decisions about the data produced. This automated counting would significantly reduce the manual effort required in traditional data collection methods, making surveying more efficient and accurate.

Image Capture and Reporting

Through Mechanatee, the User will be able to capture clear images of each recognized species. These images, along with the recognition and counting data, will provide visual evidence of each encounter. This feature allows for post-mission analysis, where Users can review and verify the species identified by the system, and how to handle the data collected.

Algorithms and Tools

There are several useful tools available. One example software is TensorFlow, which is an open-source machine learning framework developed by Google. Useful for building custom image recognition models, this provides a comprehensive array of tools for building and training machine learning models. There are many other options available like PyTorch, OpenCV, and Scikit-learn that could potentially be of aid. A thorough understanding of Convolutional Neural Networks (CNNs) will also be beneficial seeing that CNNs are the backbone of most modern image recognition. These specific algorithms are designed to adaptively learn hierarchies of features from images used. There are other supervised learning algorithms that may be beneficial, but more research into them is required.

Technical Challenges

While the idea of image recognition seems in reach, the reality of the situation is that very little is known by the team about navigating training models, or even where to start. Being the only computer science major on the team poses a challenge, but having the help of faculty and teachers who are available for consultation will make this process smoother. Very much research is needed and will be required henceforth, making this an even more time intensive process. Another challenge is selecting a processing unit that is powerful enough to run the required algorithms without slowing down.

Milestone 1 (September 30th, 2024): Hardware Selection and Initial Setup

·       Compare and select technical tools for image capture and processing

o   Evaluate different camera models suitable for the environment (waterproof, high resolution, low-light performance)

o   Assess processing units (NVIDIA Jetson, Raspberry Pi, etc.) that can handle real-time image recognition.

o   Consult with image recognition modeling expert

·       Provide demos to evaluate the tools

o   Set up a basic camera feed on the chosen processing unit.

o   Run a simple image recognition model (identifying shapes or objects) to verify the hardware set up

o   Resolve technical challenges that may arise

·       Create Design Document/Design Page on Website

o   Outline the system architecture, including the hardware setup, software plan and data flow.

·       Create a Test Plan

o   Develop a plan for testing the hardware and image recognition functionality.

Milestone 2 (October 28th, 2024): Initial Image Recognition Implementation

·       Implement, test and demo basic image recognition

o   Train a simple image recognition model using pre-labeled data relevant to Mechanatee.

o   Implement the model on the processing unit of choice.

o   Test the model with real-time camera input to ensure accurate recognition

·       Integrate and test camera feed with processing unit

o   Ensure data flow between the camera and processing unit.

o   Address latency or performance issues in image recognition.

·       Develop and test a basic user interface

o   Create a simple UI for monitoring the camera feed and recognized objects.

o   Test the UI usability and responsiveness.

·       Iterate on accuracy

o   Refine the model by adjusting parameters, adding more training data, and improving preprocessing techniques.

o   Demo the improved model with various test cases.

Milestone 3 (November 25th, 2024): Advanced Image Recognition and Preparation for Sonar Integration

·       Implement, test and demo advanced image recognition features

o   Expand the model to recognize a broader range of objects or improve recognition accuracy.

o   Implement feedback mechanisms (alerts/logging).

·       Optimize processing and memory usage

o   Analyze the performance of the system and optimize for speed and efficiency.

o   Ensure that the processing unit handles the workload within acceptable limits.

·       Prepare for sonar integration

o   Review and document the current system’s compatibility with future sonar integration.

o   Begin preliminary research and hardware selection for sonar integration.

·       Create final documentation and demo

• Document the complete image recognition system, including hardware setup, software implementation, and testing results.