Inspiration
The current coffee cherry sorting techniques are primarily manual and result in poor water waste management, wastage of water, prolonged emissions due to coffee production machines running for a longer time due improper sorting resulting from fatigue and lack of automation in coffee industries. All these culminate in a higher carbon footprint due to increased emissions and water pollution due to the water being used for density sorting being released into rivers and lakes untreated.
manual cherry sorting using color image link
manual density sorting using water example image link
emissions in the coffee industry sample image link
Worth noting is that a study carried out by the Natural Resources Institute Finland found that coffee farming and processing makes up 68% of a coffee’s climate impact, compared to 11% for brewing and less than 4% for transportation, roasting and packaging combined.
What better way to contribute to sustainability (especially in industrial emissions) than to reduce emissions in coffee industries through providing a greener, sustainable, faster, economical and more accurate solution - with a view of expanding to other industries in future?
What it does
The AI-powered coffee cherry sorting machine automates the process of sorting cherries based on colour, size, and density. These are the primary categories for sorting coffee cherries in use currently in the industry. It uses computer vision to identify ripe, underripe, overripe, halfripe or damaged cherries, and sorts them accordingly. The system provides real-time feedback on the quality of the cherries, allowing farmers and cooperatives to make data-driven decisions early in the supply chain. By doing so, it improves the consistency of premium-quality coffee, reduces waste, optimises water use in density sorting, and minimises energy consumption and emissions by improving sorting efficiency.
How we built it
We used a combination of machine learning algorithms for image recognition and computer vision for sorting the cherries based on colour and size and also tested the object detection model on edge impulse as a back-up. Both models performed well with the edge impulse output having a slightly lower loss overall. We ended up going for the edge impulse model due to ease of integration since the custom-made model failed to deploy. Ultimately, the model was trained on a custom-made dataset of coffee cherries that covered different stages of ripeness and defects. For density sorting, we utilized image bounding boxes data, the focal length data, the width and height data in implementing the coffee cherries' sorting according to density. The entire system is integrated with a real-time feedback mechanism that collects data through cameras mounted on the sorting machine and relays the data to a real-time database. This data is then visualized in a dashboard-based website design. The model is able to generalize well even when exposed to new data - this makes it very adaptable. The hardware system whose servo-mechanism runs on an Arduino Nano and uses a Raspberry pi for capturing images using a raspberry pi camera, running the sorting model, computation and relaying the data to the cloud. This system is currently being powered by a standard power supply but we are looking to integrate solar technology and rechargeable batteries in its powering system.
Challenges we ran into
- Data collection: Finding a large and diverse dataset of coffee cherries with proper labels for colour, size, and density was difficult. We had to spend a significant amount of time collecting actual coffee cherries and pre-processing the images for model training.
- Hardware integration: Ensuring the AI model runs efficiently on hardware with limited computational power was challenging. Real-time processing required optimising the machine learning models to run on low-power devices like Raspberry Pi.
- Power supply: Despite starting with a view of using rechargeable batteries and solar panels, we ended up going for a standard power supply in demonstrating our solution. This was attributed to the inavailability of the solar panels during the Hackathon period.
Accomplishments that we're proud of
- Successfully building a prototype that automates the sorting of coffee cherries, significantly reducing manual labour hence the time taken during coffee cherry processing.
- Improving the precision of sorting, ensuring a higher percentage of premium-quality cherries are selected, while also reducing waste.
- Reducing water usage in the sorting process by introducing efficient density-sorting mechanisms which involved mathematical calculations rather than using free-flowing water.
- Developing a system that is both scalable for larger cooperatives and adaptable for smallholder farmers, making it accessible to various stakeholders in the coffee industry.
What we learned
- Importance of data: The success of AI-powered systems heavily depends on the quality and diversity of the data used. We learned how to collect and the importance of contributing to the curation of better datasets for training our models.
- Hardware-software integration: Working with hardware for real-time applications taught us about optimising AI models for efficiency and low power consumption, especially when working with IoT devices.
- Sustainability impact: We gained a deeper understanding of how small technological improvements, like optimising energy use, and reducing water waste, can have a significant impact on sustainability in agriculture and reduction of emissions in industries especially coffee industries.
What's next for AI-powered coffee cherry sorting machine
- Refinement and scaling: We plan to refine our model further with more diverse data and improve the hardware’s efficiency and overall look. Our goal is to scale the solution for both smallholder farmers and large cooperatives, making it accessible globally.
- Integration with renewable energy sources: To enhance sustainability, we aim to integrate solar power or other renewable energy sources to power the sorting machine, reducing its carbon footprint even further.
- Expansion to other crops: We also envision expanding this AI-powered sorting technology to other agricultural crops and products that require similar sorting processes, such as juice, cigarettes industries, etc.
- Partnerships: We are exploring partnerships with coffee cooperatives, industry stakeholders, and sustainability organisations to test and implement the system at a larger scale.
Log in or sign up for Devpost to join the conversation.