InvaCV Monitoring

Inspiration

Nature is a fascinating topic. The interactions between animals, plants, and their surroundings is exceeding complex. However, invasive species — as made worse by extensive human activity and changing environments caused by global warming — threaten this balance. Thus, we were motivated to work to help alleviate this problem; to change the world.

What it does

A map shows users areas around them containing invasive species. They can take a picture and using machine learning InvaCV suggests the most likely invasive species in the picture.

How we built it

We built the app inside android studio using java and kotlin. We used pytorch for the machine learning component and did transfer learning on a mobilenetv3 model for image classification. For the training images we used a python web scraper to collect images of the invasive species.

Challenges we ran into

At first we wanted to use google colab for training the image classification network however we soon ran out of gpu credits and instead pivoted to training the model directly on our laptop using cpu. We encountered many problems with android studio as none of us had ever used it before. One prominent problem we spent hours trying to solve was opening the classification model and the problem ended up being that it was not inside of the assets folder and that we were using a relative path instead of an absolute path. Another major issue was getting the google maps api up and running and we spent a considerable amount of time getting the api key to work

Accomplishments that we're proud of

We are proud of the fact that we trained a custom classification model in such a time constrained setting and on such low end hardware. We are also proud of the UI design and how much we were able to learn about android development.

What we learned

We learned so much about android development and how difficult a seemingly simple task can be. We also learned a lot of coordinating a github repository between multiple people at the same time.

What's next for InvaCV Monitoring

We hope to develop out the firebase server and the response form so that users responses can be reviewed by experts. We also want to increase the accuracy and breadth of species that we can detect and expand to having species in all of the United States.