The company uses computer vision technology to detect damage to fruits and vegetables

New Delhi: Amazon India plans to use advanced computer vision, machine learning (ML) and artificial intelligence (AI) technologies to manage the quality assurance of fruits, vegetables and other agricultural products sold on its platform.

In an interview, Rajeev Rastogi, vice president of Amazon India ML, said that the company has developed a computer vision program that can identify defects such as cuts and scratches on tomatoes and onions to determine when they have deteriorated.

The system uses a mixture of convolutional neural network (CNN) and visual transformer (ViT) algorithms. CNN is a deep learning algorithm that can take image input and assign importance to all aspects of the image, while ViT is a dedicated version of the transformer algorithm that can weigh the importance of each part of the data it obtains.

"In our grocery business, product quality is the most important customer input and the primary driver of repeat purchases," Rastogi said. "Currently, quality is handled manually, which can’t really scale. It’s also prone to errors. The cost is high and the repeatability is not high. Therefore, we have developed a computer vision system to classify the quality of fresh produce by analyzing agricultural product images," he said.

Amazon uses computer vision technology to detect cuts, cracks, and pressure damage. The technology has been deployed on tomatoes and onions in Amazon stores, but Rastogi said it is building an AI-enabled machine Auto Grader to automatically grade products moving on conveyor belts.

"Compared with manual grading, it will improve product quality and consistency, while reducing grading costs by nearly 78%," he said.

Rastogi said: "In the future, we also hope to use infrared sensors to detect attributes such as sweetness and maturity, which cannot be detected in RGB (red, green, and blue) images captured by traditional computer vision algorithms." He pointed out. Infrared can help avoid "destructive methods" of quality assurance, such as maturity and sweetness that often require people to actually eat fruits or vegetables.

"The near-infrared images of desserts and non-desserts will be very different. Because the sugar content will be different, the near-infrared characteristics will also be different," he said.

The system has been deployed in stores in India, and some pilot projects have also been deployed in Europe. Rastogi acknowledged that the system is still in its early stages, and it will take some time for the company to determine when to launch the system on a large scale. However, he said that the system is more than 90% accurate in finding defects. "It varies from product to product. Just like we have better results with tomatoes, we can do this with high precision. Our onions are slightly lower, and we are working hard to (improve this)."

He said that the difficulty of determining the quality of fresh produce may also vary from product to product. "Most of the defects we are looking at (now) can be detected by human vision. I think anything that can be detected visually can also be easily detected by computer vision," he said.

In the future, Amazon can use such a system to classify and categorize products to identify the types of products offered by sellers. For example, if the seller provides a specific type of tomatoes, the system can classify the appearance of the tomatoes and determine in advance whether the correct product has been delivered.

The company must also customize the system based on India's product quality. Since the quality of Indian fresh produce is different from similar products in Western countries, the method of determining quality must also be changed. “Tomatoes in India are different from tomatoes in the United States. So you can’t just say that the methods of processing tomatoes and onions are the same in every country. Due to agricultural practices, the nature of the product may be very different,” Rastogi said.

Theoretically, the company can also bring Auto Grader to the front end. Customers only need to open the Amazon app to test the quality of vegetables or fruits themselves.

Don't miss a story! Keep in touch with Mint and understand the situation. Download our app now! !

Log in to our website to save your bookmarks. It only takes a moment.

Oops! It looks like you have exceeded the limit of bookmarking images. Delete some to bookmark this image.

Your session has expired, please log in again.

You are now subscribed to our newsletter. If you can’t find any emails on our side, please check your spam folder.