FOOD RECOGNITION SYSTEM USING CONVOLUTIONAL NEURAL NETWORK (CNN)

Artificial intelligence and computer vision have changed how digital systems understand and analyze images. Food recognition is a key application, important for healthcare, nutrition tracking, and the food industry. Still, it is difficult because food can look different depending on how it is presented, the lighting, and how it is prepared. These differences make it hard for traditional image classification methods that use handcrafted features to be accurate. This study aims to create a Food Recognition System using Convolutional Neural Networks (CNNs) that can classify food images accurately and efficiently. The research focuses on solving the issues of unreliable manual food tracking and the need for automated tools in diet and restaurant management. TensorFlow and Keras were used in the development of the system, which used transfer learning to implement the EfficientNetB0 architecture. The benchmark for training and assessment was the Food-101 dataset, which could be accessed through TensorFlow Datasets. To enhance generalization, the dataset underwent preprocessing, normalization, and augmentation. Accuracy, precision, recall, and F1-score were used to gauge the model's performance, and visualizations like latency analysis, confusion matrix, and accuracy/loss curves were used. To facilitate image uploading, prediction, and performance visualization, a Streamlit web application was created. The model demonstrated its suitability for real-world applications like restaurant automation and nutrition tracking by achieving low inference latency and high classification accuracy, according to experimental results. To sum up, the system demonstrates that CNNs with transfer learning are capable of efficiently identifying a variety of food types and offers a solid basis for upcoming additions that include portion estimation, multi-dish recognition, and mobile deployment.