Deep Learning for Supply Chain Optimization | Using Automated Robots to Sort Packages

Deep Learning for Supply Chain Optimization | Using Automated Robots to Sort Packages

How I made an autonomous robot that can help you get your online orders faster

A utomation has been the main trend over the last few years. And now, with the ever growing demand for e-commerce,and Amazon handling 5.76 million orders every single day (in the US alone!), the supply chain industry is facing a new optimization problem.

Sorting is an important step in the delivery process and traditionally, this process is carried out manually by hand. A simple Google search for “ Package Sorting Jobs ” will show you thousands of companies recruiting manpower for this task. Needless to say, manual sorting by hand is slow, inefficient and leads to delays. In a fast paced industry like Supply Chain, every minute of delay leads to loss of revenue for the company.

Thus, companies are looking for a faster, more efficient and more reliable system. This problem can be solved using Machine Learning.

So, during the Coronavirus Lockdown , with no access to electronics and hardware shops, I decided to make my own “automated sorting machine” using whatever scrap materials I could find at home. The machine is capable of sorting packages into different categories according to their final destination.

This video shows the functioning of the machine.

How It Works!

• A camera is placed above the conveyor belt.
• The camera sends a snapshot of the parcel to the computer.
• The computer processes the input and runs a Deep Learning algorithm (Faster RCNN) on the image.
• The Deep Learning model determines the appropriate destination for the package and automatically sorts it.

The Technical Stuff

• I used Tensorflow Object Detection API to train a deep learning model based on Faster RCNN Architecture.

• I trained it on my own dataset by clicking hundreds of photos of the packages to be sorted.
• After training the model using Tensorflow Object Detection API, OpenCV performs the task of classification using the inference graph and labelmap generated during training.