Transforming Agriculture With AI
Meet Amarnath De
Machine Learning Engineer @ Kisan Network
Amarnath De is currently working as a Machine Learning Engineer at Kisan Network Pvt. Ltd. on the application of ML in the agricultural industry. Detection and analysis of the quality of the crop is his main objective.
He has a strong inclination towards research in AI for its applications in the field of neuroscience.
Amarnath has published 2 papers in well-renowned conferences (Springer and IEEE), attending many different workshops and meetups and having a hobby towards robotics.
What inspired you to pursue a career in Artificial Intelligence?
Before I joined my college I was inspired by how robots used to function. During the course of my bachelor’s, I released the different areas of engineering that were involved in making a robot and how they worked/behaved the way they do eg. the motors, the microcontrollers, the code.
As I dug deeper into it during my Masters at Jadavpur university (Intelligent Automation and Robotics), I came across the different branches that were under Artificial Intelligence.
Then I realized that in front of me was an ocean full of unexplored regions. Although I love everything about robotics as a whole, it was the intelligence that inspired me the most.
After doing research and a few courses on machine learning I knew that Artificial Intelligence was my career choice.
Explain the functions of Machine learning in the Supply Chain?
“AI is the new electricity”- Andrew Ng
Machine learning (a subset of AI) is playing a significant role in every field. The supply chain is one of them. Machine Learning is used mainly in optimizing the flow of the process in this sector. A few examples are below:
Vision-based inspection: Many supply chain companies are resorting to a vision-based inspection of the product of interest. Checking of broken packages, sorting of products based on their size, detecting anomalous packages, using computer vision with X-rays to detect illegal objects without external help.
Route optimization: Package delivery is handled by humans. But to deliver a package it takes time as they have to deliver multiple packages to multiple paces in one go. Here time is a big constraint as it increases the cost due to multiple obstacles like traffic, road clearance, driver’s performance, etc.
Demand prediction and analysis: Demand depends on a number of parameters. It can be quantified based on the time of the year and location. It helps the supply chain companies to optimize their resources to those places where demand is high.
Why AI and deep learning are useful for agriculture?
In order to accelerate the process so as to meet up with global demand for produce, deep learning is used in areas such as quality/health assessment and logistics.
It gives the farmer the best deal and the retailers the quality they want, supply chain companies use object( crop ) detection and quality classification to measure the different parameters that are used to measure the quality of the crop.
Recently a lot of sophisticated object detection and classification algorithms have been invented to tackle the different challenges.
Measuring these parameters using traditional image processing techniques ( features of the crop, color, texture, environment to name a few) is very difficult. Measuring the size, condition of the crop can be done in order to assess how good or bad the yield is.
This reduces the time from sowing the seed to putting it on the shelves of the market significantly without compromising the quality of the output.
There's any development limitation of AI and machine learning in the supply chain?
Deep learning models are big correlation machines. But correlation not necessarily causations If we make a detector that is made to detect and classify some object another object with similar features will also be classified as the same.
Building a model that is able to identify features properly is the challenge for an ML engineer.
Certain crops have infinite ways it can shape (eg. ginger), if we want to make a segregation machine that counts the number of gingers in yield, the challenge is mostly in how we prepare the dataset as we have to determine the shape pattern it takes.
Where do you see yourself in 5 years in the field of AI?
To do some hardcore research and exploring new boundaries was my all-time favorite.
Since the early days, I have been fond of robots as a whole ( mechanical, electronics and most importantly the intelligence that is coded under the hood) So 5 years from now I want to see myself doing deep level research with an equivalently dedicated team in an organization that focuses on developing AI to empower us rather than overpowering us.
Currently, I have many potential areas under my radar.
Artificial General Intelligence (AGI) Building systems that are closer to human behavior.
The fusion of neuroscience and AI ( eg. nuralink)
Build robust systems that can work in any environment given. (eg. Boston Dynamics)
Most importantly fighting the environmental risk like climate change.
How do you see the future of AI implemented on the supply chain in the next 10 years?
So in the supply chain, I can easily see it being fully automated.
Crops being picked from farms by robots, sorting and processing them with the least human intervention, delivery of it to the retailer by driverless vehicles.
Many are working in making this dream come true in the respective fields
A challenge you had faced implementing ML to detect and assess the quality of crops on agriculture?
India is an agriculture-dependent country with a population of 135 crores (1.35 billion) people. Moreover, it is a country filled with a diversified landscape.
In order to build good ML models, the dataset has to have a certain pattern and also cover variation as much as possible. That pattern may be large but still deterministic. Otherwise, it makes it an NP-hard problem.
Crops have different varieties and dimensions. Using computer vision in such highly varying objects can be a challenging task thus making inspections difficult.
In some cases using computer vision is not the most robust way to determine the quality of certain crops. (eg. Pineapple)
What role models have been your greatest inspiration to get into AI?
I have read a lot of papers of authors who have shaped AI the way it is today. But honestly, at a more personal level there are few mentors who have helped me immensely in shaping my career and giving me the proper guidance:
Prof. Amit Konar: I am very fortunate to have done my M.Tech in Intelligent Automation and Robotics under his guidance. His methodologies of solving problems are one of a kind. Being very passionate about what he does and the discipline that he follows inspires me to this day. His approach to solving any problem from first principles has helped me shape the way I solve any given problem.
Amritansh Kumar: Without his initiative of organizing meetup sessions I don't think I would have gained knowledge about what AI really is. During my Internship in DRDO, on weekends he used to give us research papers to read and analyze and make a critique on it. The weekend sessions in which we had to discuss it has helped me know about the different fields in AI. Most importantly he has taught us why Knowledge about AI should be free for all.
Dr. Narayan Panigranhi: My internship at CAIR, DRDO was a big stepping stone for my career in this field. His guidance has helped me a lot. Publishing 2 papers would have been very challenging without his guidance.