Inspiration
While visiting the Pisgah National Forest in North Carolina I noticed a posting about the Spotted Lanternfly, an immigrated species that is invading North Carolina and causing irreparable damage to the ecosystem. I wondered how anyone would ever remember what such a tiny bug looked like, or even how someone would even know about this pest without seeing warning signs for it.
What it does
This project sets to solve two problems.
The impossible problem of training average people on the details of every invasive plant, animal, and insect in North America (there are approximately 4,300 of them in the US according to the Fish and Wildlife Service).
Properly identifying invasive species and determining them against native species.
How I built it
The Invasive Species Detector relies on IBM's Watson Visual Recognition. Training Watson on several images of invasive species allows the AI to detect objects in any image uploaded to it. To interface with Watson we built a website which allows users to upload images which are transferred to the AI for classification. Watson analyzes the image contents and returns information about the objects, which we display to the user.
Challenges I ran into
Properly training the AI to get accurate results from a variety of images. The AI can pick up "bad habits" where it will detect false positives. Making sure to avoid teaching it the wrong things and even worse, breaking those bad habits was a real challenge.
Accomplishments that I'm proud of
Creating a multi-part, full-stack web application using several tools, APIs, and frameworks.
What I learned
Many things. I have never used Python for webdev, I really enjoyed how straightforward and easy the Flask framework makes things. IBM's cloud services are super user friendly as well, without such a plug-and-play AI service I doubt this project would have been possible, especially in less that 24 hours!
What's next for Invasive Species Detector
Right now it can only recognize 3 species, and those species could definitely use more training. I would love to keep training the AI on more species and provide it with better images for learning.
Built With
- ai-object-recognition
- css3
- flask
- html5
- ibm-watson
- python
Log in or sign up for Devpost to join the conversation.