-
Next, We capture the Image
-
After Capturing image we will get two option. Either to upload or to re capture
-
This Pic consists of Header and Introduction section
-
This section consists of Key, Features, out Impact, and Frequently asked questions
-
The next section encourages to read blogs and educate youself
-
About Us Section
Inspiration
The inspiration for EcoSort AI came from witnessing overflowing waste bins filled with improperly sorted trash, which leads to contamination and prevents effective recycling. We wanted to address the confusion people face when deciding where waste belongs—compost, recycling, or landfill. Our goal was to create a solution that empowers individuals to make better waste disposal decisions effortlessly, using AI to make the world a cleaner place.
What it does
EcoSort AI simplifies waste disposal by automatically classifying items into compost, recycling, landfill, or hazardous categories. Users can upload or capture an image of the item they want to dispose of, and our AI model identifies the type of waste and suggests the appropriate bin. The solution aims to reduce waste contamination, improve recycling efficiency, and educate users on sustainable practices.
How we built it
We built EcoSort AI using a combination of front-end and back-end technologies: Front-End: We used HTML, CSS, Bootstrap to create a simple and intuitive interface where users can capture an image of the waste item. Back-End: We developed the back-end using Python Flask to create RESTful API endpoints for image processing and classification. AI Model: Our AI model is powered by TensorFlow and built using MobileNetV2 through transfer learning. This approach allowed us to leverage a pre-trained model and fine-tune it to classify waste effectively. Data Management: We stored our data using MongoDB Atlas and deployed the entire system on an AWS EC2 instance for scalability and accessibility. Model Training: We employed data augmentation and online data compression tools to maximize the effectiveness of our limited dataset, allowing us to train the model more effectively.
Challenges we ran into
Training AI with Limited Dataset: With only a small dataset of waste images, it was challenging to achieve high model accuracy. We addressed this by using data augmentation to expand the dataset and transfer learning to leverage pre-existing model knowledge. Accessing Uploaded Files in Code: Ensuring that images uploaded from the front-end could be seamlessly processed by our AI required us to create and refine a robust Flask API, thoroughly testing it with tools like Postman. Achieving High Model Accuracy: Differentiating between visually similar waste types like certain plastics and metals was challenging. Through iterative optimization and tuning the model’s hyperparameters, we significantly improved its accuracy.
Accomplishments that we're proud of
High Model Accuracy: We successfully achieved over 85% accuracy for waste classification, allowing the model to provide meaningful assistance in real-world situations. Seamless User Experience: We’re proud of the smooth interaction between our front-end and back-end, ensuring users can easily capture images and receive accurate waste classification results in real time. Environmental Impact: By providing an easy-to-use solution for waste sorting, we are making an impactful contribution toward reducing contamination in recycling processes and promoting sustainable habits.
What we learned
Collaboration is Key: Building EcoSort AI taught us the importance of teamwork and interdisciplinary skills, combining AI, web development, and cloud infrastructure. Data Augmentation and Optimization: We gained deeper insights into how data augmentation and transfer learning can overcome data limitations, leading to better model performance. Balancing Complexity and Usability: We learned how to simplify a complex AI-driven solution to make it user-friendly and accessible for a general audience, which is crucial for ensuring high adoption rates.
What's next for EcoSort AI
Localization and Geolocation: We plan to add features that provide users with nearby recycling centers based on their location, making the solution more practical and actionable. Gamification: To encourage engagement, we aim to introduce a points-based system where users earn rewards for correctly sorting waste, promoting environmentally responsible behavior. Crowdsourced Data Collection: We want to expand our dataset by crowdsourcing images from users, which will help the AI model become more accurate over time and adapt to different types of waste items globally.
Log in or sign up for Devpost to join the conversation.