Abstract:
"Existing food recognition systems often struggle with the cultural complexities and regional
variations inherent to Sri Lankan cuisine. This results in inaccurate calorie estimations,
hindering individuals' efforts to maintain healthy dietary habits. This research addresses this
challenge by proposing a novel Sri Lankan food recognition model based on the YOLO8
algorithm.
A custom dataset of labelled Sri Lankan food images was created to train the model. This
dataset incorporates diverse dish, ensuring robust recognition capabilities. Furthermore, the
model undergoes fine-tuning and hyperparameter tuning, optimizing its ability to distinguish
between visually similar dishes.
Initial results demonstrate promising performance. The model exhibits high overall accuracy
with specific misclassifications primarily occurring between visually similar curries. Analysis
through Confusion metrics further validates its effectiveness in identifying various curries. This
translates to accurate calorie estimation, empowering individuals to make informed dietary
choices and promote healthier lifestyles."
