Background and Introduction

Nutriku is a Capstone project developed as part of the Bangkit Batch-1 Program. Collaborating with other participants in the program, we identified significant challenges related to nutritional awareness, food assessment, and access to healthy dining options in our communities.

Our Solution

Nutriku aims to address these challenges by providing:

• Comprehensive nutritional information about food items.
• Nutriscore parameter ratings for assessing the nutritional quality of food.
• Information and resources to raise awareness about the importance of food nutrition.

Nutriku’s Advantage

• Image Recognition of Indonesian Food: Utilizing advanced image recognition technology to identify and classify Indonesian food items accurately.
• Nutriscore Rating: Assigning Nutriscore ratings to food items to help users make informed decisions about their dietary choices.
• Expert Advice and Informative Articles: Offering expert advice and informative articles to educate users about nutrition and healthy eating habits.

Technical Explanation: Machine Learning

• Model: Nutriku employs the EfficientNet Lite V2 model for efficient and accurate image recognition.
• Library: We utilize the TensorFlow Lite Model Maker library for model training and deployment.
• Learning Type: Transfer learning is employed to leverage pre-trained models and adapt them to our specific task.
• Task Type: The primary task is object detection, enabling Nutriku to identify and classify different food items accurately.

Data Training

To train our machine learning model, we followed these steps:

• Data Collection: We collected images of 24 classes of Indonesian local food items such as soto, pempek, martabak, and more.
• Data Labeling: Roboflow was used to streamline the data labeling process, making it easier and more efficient.
• Data Source: The images were obtained from Bing using the Bing Image Downloader tool, ensuring a diverse and comprehensive dataset for training.