Batch Normalization Free Rigorous Feature Flow Neural Network for Grocery Product Recognition

Abstract

Automatic product recognition is crucial in advancing economic and social fronts due to its superior reliability and time-saving nature compared to manual operations. The precise organization of products on store shelves is essential for boosting sales and ensuring customer satisfaction. However, verifying that the physical arrangement aligns with the ideal plan is a costly and time-consuming task for store personnel. In the computer vision domain, detecting products in scene images poses a considerable challenge, particularly when dealing with grocery items displayed on store shelves. The arrangement of products often presents crowded environments with numerous identical objects placed closely together. This study illustrates the ongoing challenge of identifying specific objects in complex situations despite using advanced object detection systems. The proposed framework consists of a three-stage pipeline. The initial stage incorporates a cutting-edge product detection algorithm, YOLOv5, to locate multiple grocery objects. The proposed OD-Refiner layer in the second stage identifies the missed retail object and rectifies overlapping bounding boxes of YOLOv5. The OCR-based object recognizer called Batch Normalization Free Rigorous Feature Flow Neural Network (BNFRNN) is proposed in the final stage of the pipeline. The performance of the proposed framework was evaluated using a benchmark dataset, WebMarket. The proposed framework outperforms current state-of-the-art approaches by achieving a precision score of 92.56%, recall of 85.64%, and F-score of 88.97%.

Keywords

• Convolutional neural network
• deep learning
• object detection
• object recognition
• text recognition