A deep learning based mobile application that identifies Colony Collapse Disorder in beehives.
Eligibility: Please confirm you meet all the following criteria
You are aged between 14 - 18 as of August 1, 2020
You live in the United States or its territories
You are not employed by, or directly related (grand parents, parents or siblings) to a current General Motors (GM) or Ashoka employee
You have been working on this project for at least three months
You consent to us possibly featuring your work on social media
You confirm you have the rights to use and share any content uploaded on this entry form
Eligibility: Date of Birth
What gender do you identify with?
Help us stay in touch!
3871 Appian Street : California : Pleasanton (94588)
Date You Started Your Project Started
Project Stage: Select the description below that best applies to your approach.
Established (successfully passed early phases, have a plan for the future)
1. The Problem: What problem are you helping to solve?
One of the most prominent issues currently facing our environment is bee mortality. As primary pollinators of more than one third of crops produced worldwide, these bees are critical to maintaining a balance in both our food supply and wildlife dependent on these plants. However, due to a phenomenon known as CCD (Colony Collapse Disorder) caused by the weakening of the insects’ immune system, these bees have been dying out by the millions.
2. Your Solution: How are you planning to solve this problem? Share your specific approach.
SafeHive is an innovative deep learning approach to make CCD diagnosis more accurate and easily accessible. It uses TensorFlow and a Convolutional Neural Network to take the weather, the bees’ food supply, and a picture of the hive as input from the user and output the risk percentage for CCD. In order to accomplish this, the mobile application first parses and separates the bees from the honeycomb in the hive image. It then utilizes TensorFlow to calculate the probability that they display certain signs of CCD, such as Deformed Wing Disorder (DWD) in the bees or Varroa Mite Infestation in the honeycomb. These probabilities, along with the weather and food source data, then get fed into the Convolutional Neural Network. Hidden layers in the network train their activation functions using a training dataset and, ultimately, output a percent that denotes the likelihood of the hive being affected by CCD in the present or the future. Having such predictive technology just a tap away on their smartphones will help many beekeepers prepare ahead of time for a potential outbreak and readjust their practices to prevent a similar situation from arising in the future.
4. Personal Journey: What’s the story behind why you decided to start this project?
CCD first caught my attention earlier this year in my AP Biology class, when we watched a video explaining how bees were rapidly disappearing due to the mysterious Colony Collapse Disorder. I was shocked by how much ambiguity remained around an issue of such large magnitude and how little was being done to address it. Drawn by curiosity, I continued researching into the disorder until I discovered that it was heavily influenced by several different factors, namely virus infestations, weather, and food source. It was here that I realized I could combine my newfound interest in hive protection with my long-time passion for artificial intelligence. After sifting through the pros and cons of many computer vision models and neural network structures, I finally settled on TensorFlow and a Convolutional Neural Network as the most reliable and accurate options for this project.
5. Video (Keep it simple, a video made on a hand-held phone is great): Please upload a 1-minute video to YouTube that answers the following “I am stepping up to be a Changemaker because...”
7. The X Factor: What is different about your project compared to other programs or solutions already out there?
Due to the lack of a significant public or governmental push towards addressing bee mortality, there are currently no other applications available to assist beekeepers in diagnosing or preventing CCD. Since research in the field is still ongoing, scientists don’t have a concrete idea of how different factors contribute to the disorder. Deep learning models such as SafeHive, however, train themselves based on pre-fed data, so they are often more effective than even humans experts in finding correlations and trends from different variables.
9. What’s Next: What are your ideas for taking your project to the next level?
As of now, I am using a publicly available dataset of bee health from Kaggle for the preliminary training of the model. However, this dataset is limited in its relevance to CCD factors, making it difficult to achieve a high accuracy. To expand this data, I plan on contacting and partnering with individual beekeepers as well as California environmental agencies to obtain a larger and more reliable training set that includes a greater variety in potential virus infestations and more realistic food sources. Once SafeHive is prepared for public release, these connections will also prove to be beneficial in helping the application reach a broader audience and, subsequently, maximize its positive impact.
13. Finances: If applicable, have you mobilized any of the following resources so far?
Babji Nallamotu, (510) 393-7987
Srilakshmi Nallamotu, (925) 858-4500
Bhanu Mummaneni, (408) 480-0827
Are you employed, or directly related (grand-parents, parents, sibling) to a GM or Ashoka employee?
How did you hear about this challenge?