Amazon India on Wednesday introduced a shelf monitoring solution — a Machine Learning (ML) powered farm-to-fridge quality assurance system for fresh produce.

With this launch, the tech giant aims to enable its sellers to meet the consumer demands for high-quality fresh fruits and vegetables while shopping on Amazon Fresh.

“The shelf monitoring solution will bolster the capabilities of Amazon Fresh sellers to fulfil the commitment of delivering the finest grocery services in India. With seamless automation, the solutions ensure top-notch quality of fresh produce, enhancing customer satisfaction,” Rajeev Rastogi, Vice President, Machine Learning, Amazon, said in a statement.

Store shelf monitoring solution is powered by state of art computer vision models and Wi-Fi-enabled IoT cameras to detect pre-determined defects in fruits and vegetables using the image of the crate, as an input.

The solution detects the count of visible items of produce and identifies specific visual defects such as cuts, cracks, and pressure damage among others, the company said.

Amazon has developed two types of models — one for detecting each item in the crate and counting the total number of items, and a second to identify the defect classes present in each item.

Both these models are trained using annotated defects in millions of produced images. “Our focus has always been customer backwards and we are providing our Amazon Fresh customers consistent and superior quality of fresh produce across India,” said Harsh Goyal, Director and Head of Everyday Essentials, Amazon India.

Moreover, the tech giant said that the shelf monitoring solution currently supports manual monitoring through a mobile app and automated monitoring using cameras installed on top of produce shelves.

In manual monitoring, operators use the Johari app to submit a produce crate image taken freely from their smartphone. The shelf monitoring solution assesses the image for quality and if acceptable analyses the image to detect defects and uses grading logic to highlight the items that don’t meet the quality criteria and need to be culled.