Inspiration
Home gardening and green roof helps reducing the carbon footprint for individuals! However, for many people, it is hard to raise a plant child from beginning to the end as there are may times disease that are hard to identify. This is why PlantDoctor comes in handy to identify whether your plant is doing alright or is it suffering.
What it does
Plant Doctor first takes a picture of the plant using a PiCam, which sends to the Raspberry Pi for processing. The board is loaded with the a convolutional network using transfer learning of ResNet152, trained with an open source health/diseased plant dataset (Link: https://www.kaggle.com/datasets/vipoooool/new-plant-diseases-dataset). As such, PlantDoctor then identifies whether the plant is disease or not, and outputs the classification result on the screen for user.
How we built it
The main hardware components the team used is Raspberry Pi board and Arduino board to allow the use of bluetooth module and the taking image/processing of image. Furthermore, the team developed the deep learning model during the duration of the hackathon and trained with different parameters to improve the model accuracy, which resulted in: batch_size=114, num_epochs=5, learning_rate=6e-4, decay_rate=0.9, threshold=0.7.
Challenges we ran into
Due to time constraint, the team was having trouble integrating all parts together and running out of time for training the model. In the future, we will research ahead of time about the specifications for the components, and develop a more detailed plan to ensure that our progress follows the timeline of the hackathon.
Accomplishments that we're proud of
We are really proud to have used Raspberry Pi for the first time and be able to integrate the use of PiCam to take pictures and integrate it with the deep learning model. We are also proud of ourselves to be able to develop the full transfer learning neural network and achieve a high training and validation accuracy during the short period of time.
What we learned
We as a team has learned much more about the use of computer vision, Raspberry Pi, use of Arduino bluetooth modules, etc! Most importantly, we improved on our hardware developing skills and the integration between hardware components with deep learning model.
What's next for PlantDoctor
For the next steps, PlantDoctor aims to include more variety of plants and wider range of diseases to be able to classify more commonly seen garden plants! Furthermore, the team aim to have the setup mounted onto autonomous small "rover", which the robot can identify objects that resembles the shape of a plant, and move itself over, then classify the diseases it may have. Long way to go!
Log in or sign up for Devpost to join the conversation.