ECOBIN: Artificial intelligence-based dustbin that sorts recyclable and non-recyclable materials from waste.

This project is a component of a waste management system that sorts recyclable and non-recyclable materials from trash automatically. It uses the latest technology of Artificial intelligence to separate different kinds of recyclable materials like metal, glass,

Project Title

ECOBIN: Artificial intelligence-based dustbin that sorts recyclable and non-recyclable materials from waste.

Project Area of Specialization

Artificial Intelligence

Project Summary

This project is a component of a waste management system that sorts recyclable and non-recyclable materials from trash automatically. It uses the latest technology of Artificial intelligence to separate different kinds of recyclable materials like metal, glass, plastic, paper, cardboard, and trash at the source of its generation.

In this project, more than 5000 images of recyclable and non-recyclable materials are collected from different parts of the city of Karachi as input data to the deep learning algorithm. The algorithm of the convolutional neural network trains a model which we will use to classify our real-time data from our bin. We use Jetson Nano which is a developer kit that lets us run multiple neural networks in parallel for our computer vision applications. The bin has a camera, sensors, LCD screen, motors all controlled by Jetson Nano.

When the user throws trash in the bin it kept in a compartment initially (A in figure 3), where the camera capture an image, this image then goes into Jetson Nano where it pre-processes that image and infers the image using a pre-trained deep learning model whether it is recyclable or not. For example, If a thrown trash is paper then Jetson nano identifies it as paper and gives a command to the motors to rotate the rotor so that the paper bucket comes underneath the compartment, then doors (D in figure 3) drop the paper into the paper section.

Project Objectives

Improper waste management will have enormous adverse impacts on the economy, public health, and the environment. Effective waste recycling is both economic and environmentally beneficial. It can help in recovering raw resources, preserving energy, mitigating greenhouse gaseous emission, water pollution, reducing new landfills, etc.

In Karachi, solid waste recycling relies on local collectors who trade for profit. This collection process is mostly done by hand and most of the recyclable materials are disposed of with the non-recyclable materials and end up in landfills.

The main goal of this project is to design a system to classify waste into two basic categories, recyclable and non-recyclables. Recyclable materials are further classified as paper, plastic, metal, cardboard, and glass, and more, which will be separated by a robotic system into their respective bin.

Our main goals are :

1. To design an efficient way to manage and sort recyclable materials.
2. Prototype a robotic bin that sort the waste category wise and store it.
3. Train a deep learning model to classify images with high accuracy.
4. To find out those waste materials in Karachi which are an environmental hazard and try to give a solution to collect and dispose of them.
5. To study the current system of garbage collection, recycling plants, generating materials, and landfills of  Karachi.

Project Implementation Method

First, we need to collect data which consist of different types of waste images from different parts of the city, Karachi. Those images then will be labeled group wise. To enhance image features and to remove unwanted noise, images captured by the camera are preprocessed under the Keras framework. During training, augmented images, including image rotation, height/width shifting, size rescaling, zooming, etc are generated for each data instance to enhance the universality of the training model. We implement this project in parallel with the software and hardware side of the project shown in the Gantt chart in figure 2.

Convolutional neural networks (CNN) are widely applied in analyzing a visual image. We will use CNN to train a model on pre-processed waste images. Generally, CNN takes images containing investigated items as inputs and classify images into different categories. The capability of CNN can be controlled by varying dimensional parameters and local architecture structure. In recent years, different CNN architecture variations emerge. In considering the computational cost and in-field application limitations.

Our goal of this project is to achieve the maximum accuracy rate of the model. The higher the accuracy of the model, the more it will sort the waste precisely. After successful training and testing of the model, the model then will be deployed on Jetson Nano.

The dustbin body is to be designed on Computer-Aided Design software “SOLIDWORKS” and developed using acrylic material. All the hardware assembling and testing will be done in parallel with the software side. Integrating all the motors, sensors, cameras, Jetson Nano and deep learning model is the most critical part of this project. To avoid any problem, in the end, we will use the divide and conquer algorithm in which we divide a big problem into several small problems then try to solve those small problems one by one.

Once the integration and synchronization of all the parts are done, the bin will be able to sort the user’s trash automatically into its respective bin.

Benefits of the Project

1. This project can help in recovering raw resources, preserving energy, mitigating greenhouse gaseous emission, cope water pollution, reducing new landfills.
2. It can revolutionalize the waste management sector by separating the recyclable and non-recyclable materials at the user end very efficiently.
3. It has the ability to collect more recyclable waste that can increase the production of local recycling plants.
4. It can help in identifying the waste that is harmful to the environment.
5. We can also make this project to collect disposable surgical masks into a separate partition which then can be shredded and use to make products such as composite lumber for shipping pallets, railroad ties, and composite decking.
6. The sorted recyclables can be directly sent to their respective recycling plant.
7. It can be connected to an online cloud system to monitor the current level status of each bucket of the bin, and a request is generated when any of them is full.

Technical Details of Final Deliverable

It is a prototype of a bin that consists of a small computer Jetson Nano, cameras, motors, and a metal sensor. Jetson Nano Developer Kit is a small, powerful computer that let us run multiple neural networks in parallel for image classification and object detection. The motorized rotor (B in figure 3) is of cylindrical shape having six equal partitions for each category of recyclables. High torque geared Motor will be used to rotate the rotor. All this system is powered by a power supply of 12v dc. An object detection sensor detects any object in a trash hole(A in figure 3), a metal sensor(inductor) used to sense metal in the trash, a camera fitted on the walls of the trash hole capture an image of the object and send it to the Jetson nano which has pre-trained deep learning model embedded that infer the waste category to process it. By using the results from the model Jetson nano then send the command to the motor to rotate the rotor to a certain degree so that the resulted category bin in the rotor comes underneath the trash hole. The doors (D in figure 3) will open and drop the object in its respective portion.

For example, a paper is placed at A, an object sensor detects(shown in figure 4) the presence of paper and sends the signal to Jetson nano which then sends a command to the camera to capture an image of the paper. The captured image then goes to the pre-trained model embedded in Jetson Nano, which classifies that image as a paper. Jetson Nano process these results to calculate the required position of motors to rotate the rotor so that the paper bucket comes underneath the trash hole. Doors (D in figure 3) then opens to drop paper into its bin. A precise optical encoder sensor is continuously synchronized with Jetson and sends the current position of the rotor. All the status and process displays on LCD. If the object is metal then the metal sensor (inductor shown in figure 4) senses the magnetic changes that occurred by the object and sends it to Jetson nano to further process it to drop the metal in its respective bucket.

Final Deliverable of the Project

HW/SW integrated system

Core Industry

IT

Other Industries

Others

Core Technology

Artificial Intelligence(AI)

Other Technologies

Robotics

Sustainable Development Goals

Sustainable Cities and Communities, Climate Action, Life on Land

Required Resources

If you need this project, please contact me on contact@adikhanofficial.com