Inspiration
Do you know the journey fresh produce takes from seed to your fridge?
We’re all aware of food waste - It contributes to up to 10% of global greenhouse gasses - United Nations Environment Program (UNEP) (2021). Food Waste Index Report. We know grocery stores contribute to food waste by encouraging consumers to buy more than they need, overstocking shelves, inaccurately predicting shelf life or damaging products and restaurants also waste food by mismanaging inventory, poor menu choices or oversized portions. Despite this a majority of food waste in Australia comes from our homes (2.5 million tonnes) - The Food and Agribusiness Growth Centre (2021). National Food Waste Strategy Feasibility Study
Do you know the journey fresh produce takes from seed to the fridge?
Our team also had this question and want to target food waste at a household level through reconnecting consumers with the produce they’re eating. Our goal is to educate and raise awareness of the processes by which fruits and vegetables are grown. We seek to instil a deeper and empathetic connection with fresh produce. Interactions on LeafyLink encourage conscious consumer choices to reduce food waste and ultimately improve an individual’s contribution to a greener planet.
What it does
LeafyLink is an application that connects users back to the roots (literally) of what they’re consuming. We want to ignite a feeling of curiosity across the community with a social application for monitoring and learning about growing your own produce from seed to harvest. It is the perfect platform for novices to give gardening a shot. The app provides handy digital tools to make gardening accessible for people living in metropolitan areas. Regardless of how big your place is, we have a gardening solution for you. We want to foster a community of green thumbs who are aware of how fresh produce is grown and are responsible consumers.
Our app features:
- 🔢 Recommendation Algorithm - We take in your living conditions and lifestyle preferences and recommend you what to grow.
- 🌱Digital Garden Diary - Become acquainted with your plants, track how they grow through a daily diary and smart notification system. We help you catalog all your plants so you can have peace of mind.
- ❔ Plant Troubleshooting - Our image recognition system detect potential illnesses early on and recommend research based treatment and harvesting solutions.
- 👪 Encouraging Community - Socially connect with growers by watering other user’s plants near you. Discover what plant might be your next best friend!
How we built it
We built this application with a React frontend. Our plant recommendation algorithm takes in information including budget, location and living space while also considering how much experience the user has with gardening to match against our extensive list of plants. From inputted data, we infer information about seasonal sunlight hours and climate zones, matching each user’s input against 8 parameters (climate zone, available space, pot preference, garden preference, harvest time, required light levels, difficulty of growing plant, if in budget) of varied weighting to provide the best suggestions to our users. The backend of the application is built with Python and we’ve used Flask for the server and developed and trained our plant disease detection model using PyTorch.
Recommendation algorithm:
We have 8 parameters for which the user's input and the ideal conditions for a plant will be compared. Matches will increase the weighted score and misses will either decrease or not change the score. Some misses can either decrease or increase the score based on the distance of the user's parameter to the plant's ideal conditions. The higher a score, the more recommended the plant is. Our system will take the top 3 recommended based on input and display them to the user.
The parameters are as follows:
Climate: The Köppen climate classification of the user based on their latitude and longitude is matched against the ideal climate zones of each plant. Each plant may have more than one ideal climate zone.
Space: Plants may take up high, medium or low amounts of space. For example, an orange tree, carrots and basil would take up respectively high, medium and low amounts of space. Our algorithm will not suggest plants that our user does not have space for based on their property type.
Pot plant: We will preferentially score potted plants and factors such as daylight hours will have less effect on the weighted scores of potted plants.
In ground plant: We will preferentially score in ground plants and factors such as daylight hours will have more effect on the weighted scores of potted plants. (Note that some plants are suitable for both pots and in ground.)
Daylight hours: Based on the user's latitude and their current season, their approximate daylight hours will be calculated. This will be matched against the ideal sun requirements of plants - full sun, partial sun or shade. (Note that some plants can thrive in both partial sun and full sun.)
Difficulty of cultivating a plant: Some plants may require a more narrow range of conditions in which they thrive. We will match this parameter based on the user's experience. For example, nurturing an avocado tree is less novice friendly than growing a pot of kitchen basil.
Harvest time: Some plants will have longer times until they may first be harvested.
Budget: Some plants are more exotic and thus will be more expensive, especially for a novice to gardening who may not be willing to invest a high amount of money.
Each of the parameters have varying weighting which we tweaked until we got reasonable results.
Plant disease detection:
We use a convolutional neural network in order to predict the plant disease status. The model has 4 convolutional layers, and 3 fully connected layers. The architecture and model was reused from an existing open source project, which shortened development time. A separate server and backend was developed in order to host this model. In the future, a full model can be built and classify wider varieties of plants and diesases.
Challenges we ran into
We wanted a virtual AR garden so that users can see what their living space could look like with plants. We struggled to find a way to easily prototype AR, because none of our laptops had native AR support, and we were unsure whether we could complete it in 24 hours.
We trained and extended a model which classified whether plants had diseases. However, we were again unsure whether we could complete this within 24 hours. As a fall-back plan, we have a troubleshooting Q&A so that users can self diagnose their plants’ diseases.
Accomplishments that we're proud of
We’re immensely proud of the efforts of our team in building a functional prototype in under twenty four hours. We’re proud of the grit and determination our team members have put in to grow the idea of LeafyLInk to a reality.
What we learned
Our team gained the experience of working together to build LeafyLink in the span of twenty four hours from the conception of the idea to having a functional product. We're able to get hands-on experience building in languages we were not all familiar with - Javascript and python are certainly challenging languages to work with. We learnt how to host and deploy our site and integrate the front-end and back-end. We learned how to brainstorm empathetically, taking a human centered design approach and creating user flows that are seamless and logical. A vital learning was building software in a team - We gained practical experience working with git and utilizing version control to manage our files. Moreover, how to communicate ideas via written and visual means when rapid prototyping.
What's next for LeafyLink
In the future we’d like to expand the community aspect of the app by adding more social integrations such as swapping plants and sharing growing tips. We recognise the value of growing together to one's social wellbeing. We want to make the app more fun and include customisation such as skins and costumes for plants to incentivise continual use of the app. Further expansions include recommendations about composting and recipes which relate to the post-harvest lifecycle of our fresh produce. We’d love to partner with organizations such as OzHarvest which have initiatives targeting the mutual goal of preventing food waste and incentivise getting on the app. Together we propel towards greater awareness about conscious consumption.
Resources used: http://climateapi.scottpinkelman.com/
Log in or sign up for Devpost to join the conversation.