Advancements in machine learning and deep learning have led to increased automation in various aspects of daily life, including retail. Automation helps optimize supply chains and improve inventory management, demand forecasting, and logistics coordination. However, there are challenges when it comes to automating tasks like identifying produce without barcodes.
To address this issue, researchers from Skoltech and other institutions have developed a computer vision-based approach to distinguish weighted goods at supermarkets. They collected different types of images, including natural images taken in gardens, grocery stores, and labs. They also used top-view container images with multiple objects. By combining and augmenting these images, they were able to train a neural network that can accurately identify different types of produce.
Prior to this approach, classical computer vision systems had limitations. These systems had to be retrained every time a new variety of produce was introduced, which was time-consuming and required manual labeling of a large amount of data. However, the new approach developed by the researchers eliminates the need for retraining and achieves high accuracy even with a small number of training images.
Benefits of the New Approach
The researchers found that their approach is highly effective, particularly when the number of training images is low. In their experiments, they categorized five different types of fruits and demonstrated that the approach can achieve 98.3% accuracy without using natural training images. This means that the system can accurately identify and categorize produce without the need for extensive training data.
Automation in retail, particularly in tasks like identifying produce, is becoming more efficient and accurate with the help of computer vision techniques. The approach developed by Skoltech researchers eliminates the need for retraining and achieves high accuracy even with a small number of training images. This advancement has the potential to improve efficiency and customer experience in supermarkets.
To learn more about this research, you can read the full paper.
If you enjoy our work, you’ll love our newsletter, so don’t forget to subscribe!