1 of 13

Image Classification App to Detect Allergens in Indian Dishes

TITLE.HTML

Anya Parekh | Computer Systems

2 of 13

/MY PROJECT

RELEVANCE.HTML

  • Using image classification to identify Indian dishes
  • Kaggle for datasets
  • TensorFlow Lite Model Maker for building the model
  • Android Studio for developing the app

3 of 13

/IMAGE CLASSIFICATION’S APPLICATIONS

/MEDICINE

/AGRICULTURE

/FASHION

APPLICATIONS.HTML

The most common use of image classification. Has been used to identify lesions.

In the field of agriculture, image classification has been used to identify soil, plants, fruits, etc.

The most recent expansion of image classification. It was used to separate between apparel and accessories.

4 of 13

/GATHERING DATA WITH KAGGLE

/GATHERING DATA WITH KAGGLE

DATA.HTML

  • Determined feasibility of project
  • “Food: Allergens and Allergies”
  • “6000+ Indian Food Recipes Dataset”

​

5 of 13

/IMAGE RETRIEVAL

RETRIEVAL.HTML

Different ways to retrieve images

/DuckDuckGo

/Bing Image Search API

/Chrome Extension

Can be done with code

Can be done with code

Has to be done manually

Simple implementation

Harder implementation, but would provide better images

Inefficient

Can only work with a limited amount of data

Can only work with a limited amount of data

Can work with unlimited amounts of data

6 of 13

/IMAGE CLASSIFICATION MODEL

Approach #1: fast.ai

  • Had previous knowledge
  • Simple, easy-to-use library
  • BUT couldn’t be integrated into Android Studio

Approach #2: TFLite

  • Completely new
  • Relatively easy-to-use API
  • Could be integrated into Android Studio

7 of 13

<BUILDING THE MODEL>

/BUILDING THE MODEL

BUILDING.HTML

8 of 13

FORMATTING.HTML

TensorFlow Lite Model Maker only takes .jpg images

/FORMATTING IMAGES

9 of 13

CREATING.HTML

Training and evaluating the model

/CREATING A MODEL

10 of 13

INSTANTIATING.HTML

Creating a new instance of the model

/IMPLEMENTING IT IN ANDROID STUDIO

11 of 13

/USER INTERFACE

12 of 13

/FUTURE STEPS

  • Improve accuracy of model
  • Make app more aesthetically pleasing → marketable
  • Expanding app use
    • More Indian dishes
    • Other cuisines

YOUR LOGO HERE

INDEX.HTML

13 of 13

ANY QUESTIONS?

QUESTIONS.HTML