Abstract:
Visually impaired individuals face signifcant challenges in identifying food items and understanding their nutritional content, often relying on others for assistance in daily dietary decisions. This dependency not only limits personal independence but also increases the risk of consuming inappropriate or unhealthy meals. To address this gap, this project presents an AI-Augmented Food Analyzer, a mobile application designed to identify food items and their calorie content using image based analysis. The system utilizes state of the art deep learning models, including ResNet101, EfcientNet-B4, and ConvNeXt-Small, trained on a saliency enhanced dataset to improve classifcation accuracy and model focus. Through this approach, the models achieved an accuracy exceeding 90%, ensuring reliable performance in real world food recognition scenarios. The application leverages visual saliency to highlight key regions within an image, enabling more robust feature extraction despite challenges such as inconsistent plating, varied lighting conditions, and mixed food compositions. Once the food item is identifed, its calorie information is estimated and communicated to the user through an accessible interface designed specifcally for visually impaired individuals. Features such as audio prompts, simplifed navigation, and nonvisual feedback ensure ease of use and independence. By integrating advanced deep learning techniques with a user centered design philosophy, this project provides a meaningful solution that enhances accessibility, supports healthier dietary choices, and contributes to the growing feld of AI driven assistive technologies
