Real-Time Detection of Fruits and Vegetables with Calorie Estimation Using Deep Learning

Abstract

This paper presents a comprehensive
approach to real-time fruit and vegetable detection
integrated with calorie estimation using deep learning
techniques. With the rise of health awareness and the need
for accurate dietary assessment tools, the proposed
system addresses a practical challenge by enabling
automated nutritional analysis through computer vision.
A custom YOLOv5 object detection model was trained on
a robust dataset containing 32 classes of commonly
consumed fruits and vegetables. The model demonstrates
high accuracy in diverse environments, including
variable lighting and background noise, making it
suitable for real-world deployment. Detected items are
mapped to a locally maintained calorie dictionary based
on standardized nutritional data, enabling calorie
estimation per item and cumulative calculation for
multiple items in a single frame. The system supports both
image uploads and real-time webcam input and is
deployed as a lightweight web application built using
Flask. It offers fast inference, requiring no external APIs,
ensuring data privacy and offline usability. The solution
has potential applications in fitness tracking, diet
planning, educational tools, and mobile healthcare
solutions. Experimental results demonstrate strong
performance in terms of detection accuracy and response
time, making the system a reliable aid for smart food
analysis and calorie tracking.