Challenges & Mitigation:
This week, the biggest change was switching from using an API to web scraping for collecting environmental data. The API we planned to use was unreliable (down, buggy, outdated) and lacked important data. Paid backup APIs weren’t an option, so we switched to scraping data from a website focused on houseplants. The challenge is that this site only covers houseplants, which may limit our scope. We’ll either narrow our focus to houseplants or find another source.
We also switched from AWS to Replit due to the lack of AWS credits and other limitations. This change required us to adjust the web app code, which has already been done. We now need to explore different options for user notifications, such as Twilio, since we no longer have access to AWS services.
We’re still facing challenges with getting enough plant data for the ML model. To solve this, we’re collecting our own data and looking into data augmentation techniques. We’re also prioritizing setting up sensors and cameras to collect data in the next few weeks.
Progress:
- ML Framework: The main framework for plant health classification has been set up. We’re testing multiple models and using online image datasets to find the best-performing model.
- API & Web Scraping: The API code has been set up, and we’ve started web scraping from the chosen website. We’re in the process of collecting a more comprehensive dataset.
- Frontend & Backend Development: New frontend pages have been added to the web app, and we’ve completed implementing WebSockets for real-time communication between hardware and software. Additionally, we’ve switched the database from SQLite to MySQL after changing from AWS to Replit.
Next Steps:
- Finish training the ML models for plant health classification
- Set up the RPi camera and begin collecting our own image data for training the ML model
- Finish the web scraping process to gather a full dataset of plant environmental conditions
- Integrate the Raspberry Pi with the web app
- Start integrating the various components we’ve been working on separately