Poor Recycling Quality Due to Lack of Education Most people know that recycling is a good habit for the environment, but are often unsure of what goes into the recycling bin. This leads to:
Non-recyclable materials are put in the recycling stream (such as liquids or plastic bags) that contaminate recyclable materials and compromise recycling machinery.
Consumers turn to the trash bin in defeat.
What it does
The project will really decrease this problem and save a lot of resources and materials. Imagine that you have an empty bottle of water and you are distracted. Where will you throw it in the recycling bin or in the waste bin? The project will help you as it detects the item and by depending on artificial intelligence decides to which bin it belongs. The project shows the importance of artificial intelligence and that will save a lot of useful materials we need and will help in decreasing the cross-contamination problem which will show a lot of positive impacts on our environment.
How we built it
A convolution neural network algorithm is a multilayer perceptron that is a special design for the identification of two-dimensional image information. Always have more layers: input layer, convolution layer, sample layer, and output layer. In addition, in a deep network, the architecture convolution layer and sample layer can have multiple. CNN is not as restricted as the Boltzmann machine; it needs to be before and after the layer of neurons in the adjacent layer for all connections, and for convolution neural network algorithms, each neuron doesn’t need to feel the global image, just feel the local area of the image. In addition, each neuron parameter is set to the same, namely, the sharing of weights, namely each neuron with the same convolution kernels to the deconvolutional image. I wanted to use an algorithm that is known and effective at detecting if an object is recyclable or not. Using the Convolutional Neural Network algorithm, we were able to process and break down the images, so it would simplify them and let the computer do accurate predictions. I collected images of what is accepted in the curbside recycling bin and what’s not according to Republic Service®. During the course of 1.5 weeks, our team collected about 5,000 images within the time and started training the model using Google’s teachable machine. This helped us get the base for the project.
Challenges we ran into
At first, the program functioned when I fed an image to the system and it responded with a mini-matrix; which is really hard to comprehend for most members of the general population. I solved it by putting an if-else statement to make it has two greatly simplified outputs. But a major problem was that the detection was based on file upload instead of the video detection format we were looking for. After many trials and errors, I decided to let the user capture their webcam and save the image under the file name photo.jpeg. Then I was able to run the captured image through our original code and it gives an output I am trying to make the program have an alarm when the output is non-recyclable. Currently, it gives an error output I have tried multiple methods for sound . They all seem to have errors such as audio errors, number errors, etc. I attempted to use the Pygame method which gave me no available audio error. I am currently in the process of fixing it.
Accomplishments that we're proud of
After the collection of the pictures and the programming, the results were really astonishing. I tested the project. I showed a maximum of 50 materials to the webcam and it showed positive results. As a plastic bottle that is empty is made up of recyclable plastic and contains recyclable plastic. The program detected and showed it is 100% recyclable. By showing it a water bottle. the program detected the filled plastic water bottle and showed it is 100% non-recyclable.
What we learned
That AI is a great tool that can be invested in solving SDGs as I started to solve recycling problems using it. It is a small step for long progress. Also, YOLOv3 is extremely fast and accurate. In mAP measured at .5 IOU YOLOv3 is on par with Focal Loss but about 4x faster. Moreover, you can easily tradeoff between speed and accuracy simply by changing the size of the model, there's no retraining required. The unified architecture is extremely fast. The base YOLO model processes images in real-time at 45 frames per second. A smaller version of the network, Fast YOLO, processes an astounding 155 frames per second while still achieving double the mAP of other real-time detectors. Compared to state-of-the-art detection systems, YOLO makes more localization errors but is far less likely to predict false detections where nothing exists. Finally, YOLO learns very general representations of objects. It outperforms all other detection methods, including DPM and RCNN, by a wide margin when generalizing from natural images to artwork on both the Picasso Dataset and the People Art Dataset.
What's next for Green AI
The future implementation ideas for the project are making an effective and cheap monitor that will detect whether the material is recyclable or non-recyclable. Moreover, the future prospect is improvement in the design and structure of trash bins which plays a significant role in the smart world.