Easy-garden

Machine Learning model to take care of your plants easily.

Inspiration

Most of the people like having plants in their home and office, because hey are beautiful and can connect us with nature just a little bit. But most of the time we really don´t take care of them, and they could get sick. Thinking of this, a system that can monitor the health of plants and tell you if one of them got a disease could be helpful. The system needs to be capturing images at real time and then classify it in diseased or healthy, in case of a disease it can notify you or even provide a treatment for the plant.

What it does

It is a maching learning model that takes an input image and classify it into healthy or disease, displaying the result on the screen.

How I build it

I use datasets of plants healthy and disease found in PlantVillage and developed a machine learning model in Tensorflow using keras API, getting either healthy or diseased. The data set consisted in 1943 images are of category diseased and 1434 images are of category healthy. The size of each image is different so the image dimension. Most of the images are in jpeg but also contains some images in .png and gif. To feed the maching learning model, it was needed to convert each pixel of the RGB color images to a pixel with value between 0 and 1, and resize all the images to a dimension of 170 x 170. I use Tensorflow to feed the data to neural network, and created 3 datasets with different distributions of data, Training: 75%, Valid: 15% and Testing: 10%.

Challenge I ran into

Testing a few differents models of convolution neural networks, there were so many different results and was a little difficult to adapt to another architecture at first, ending with the model Vgg16, that was pre-trained on imagenet but we still change the architecture a little bit it was posible to retrain the model just a bit before it ends. App building continue to be a challenge but I have learned a lot about it, since I has no experience on that, and trying to combine it with AR was very difficult.

What I learned

I learned a lot more neural networks models that haven't used before as well as some API's very useful to develop the idea, some shortcuts to display data and a lot about plant diseases. I learn some basic things about Azure and Kinvy platforms.

Built With

• colab
• jupyter-notebook
• kinvy
• python