Waste Sorter using Images

Inspiration

The world population produces between 7 and 9 billion tons of waste every year, of which 70% is mistreated, ending in landfills with the risk of polluting natural environments. This could hugely affect the environment. In order to manage this and protect our environment for future generations, the idea for this project was born.

What it does

The software takes in images and identifies what type of waste it is, i.e., glass, metal, paper, or plastic, and classifies it accordingly.

How we built it

We needed to use artificial intelligence for this purpose. So, we made a CNN, i.e., Convolution Neural Network model for analyzing and classifying the images.

While building the model we made multiple layers. The first few layers enhance the image making it easier for classifying. These layers were made of convolution layers of 32, 64, 128, and 256 filters, and the images coming out after the layer was normalized and passed through a max pooling layer. So, only the important features of the images were retained while the rest was discarded or ignored.

Next was dense layers. These layers are responsible for classifying the image. There were multiple dense layers of 1024, 512, 256, 128, and 4 decision trees in each one separately. Like in the earlier layers, the data coming out of each of the dense layers was also normalized.

The data was split in a ratio of 3:1 for the training and validation sets respectively. Even though it was restricted to 40 epochs in the code, upon analyzing and studying the training graph, the model can yield an accuracy of up to 85-90%.

Challenges we ran into

Since it is a CNN model, images are loaded and as such, it takes a lot of training time. The only way to reduce this is to integrate it with an extremely good processor and a GPU, i.e., Graphics Processing Unit, or a TPU, i.e., Tensor Processing Unit.

Accomplishments that we're proud of

This is the first-ever CNN model we have created with such a huge application in the environment. This model can also yield up to 85-90% accuracy when 100 epochs are used.

This software can also help in making sure of a sustainable environment.

Studies have also shown that the coronavirus can survive for 7-9 days on non-biodegradable wastes. With physical waste management methods, this could lead to secondary transmission on contact. Hence, automatizing waste classification by using our software can contribute to developing solutions for these problems.

What we learned

We learned how to build a CNN model from the ground up. We also got a clear insight into how human activities have and are affecting the environment badly.

What's next for Waste Sorter Using Images

This software must get a bigger dataset to learn from with more categories for the waste types. It must also be used on a processor with extremely high speed to train the model.

For this software to be applied in the real world, it must be integrated with a robotic arm or a conveyor belt system to segregate the waste.