MechFusion AI has successfully completed a project titled “Real-Time Food Recognition and Pricing System Using Convolutional Neural Networks.” This project aimed to develop a comprehensive solution for addressing the challenges faced by restaurants and food stalls in Malaysia. By leveraging advanced Convolutional Neural Network (CNN) algorithms, the system efficiently recognizes food images and calculates accurate pricing, ensuring transparency and minimizing human error in the food pricing process
Project Overview:
Background and Problem Statement: The growing diversity of food items and the complexity of traditional food recognition methods necessitate a precise and efficient system. Traditional approaches are often slow and prone to inaccuracies, leading to customer dissatisfaction. This project sought to modernize the food service industry with a more reliable solution.
Objectives: The primary objectives included developing a CNN-based model trained on extensive datasets, integrating food detection and segmentation with a pricing system, and deploying the model on a microcontroller for real-life testing.
Methodology and Project Flow: Utilizing YOLO v8 and Mask-RCNN for initial model development, the project proceeded through a structured flow, including dataset preparation, model training, subsystem integration, and rigorous testing. The dataset comprised 10,000 images, with a split for training, validation, and testing. The model’s accuracy, loss, mAP, and recall were evaluated using Google Collaboratory.
Results and Evaluation: The final model achieved a 0.95 accuracy rate, with confidence set at 0.59, indicating robust performance without overfitting. The system was tested on hardware, including Nvidia GeForce RTX 3050 and various webcams, under different environmental conditions to ensure its reliability.
Future Plans: Suggestions for future work include refining the system’s algorithms, increasing the training dataset size, deploying the system on an edge device, and enhancing GPS mapping accuracy.
