Deep learning-based halal food recognition
Halal is the term used for permissible food according to Islam. Indicators such as Halal logo have been used to guide Muslims in identifying Halal food. Department of Islamic Development (JAKIM) in Malaysia has introduced a standard Halal logo for locally manufactured products. Problem arises when M...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2020
|
| Subjects: | |
| Online Access: | http://eprints.utm.my/id/eprint/93023/1/MohamadSyafiqMazliMSKE2020.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Halal is the term used for permissible food according to Islam. Indicators such as Halal logo have been used to guide Muslims in identifying Halal food. Department of Islamic Development (JAKIM) in Malaysia has introduced a standard Halal logo for locally manufactured products. Problem arises when Muslims in Malaysia are travelling overseas, especially to non-Latin language country. It is difficult to find the ingredients used in the product, as it is written in non-Latin language, to determine whether it is a Halal or non-Halal product. Thus, this paper proposed the use of an image recognition system in overcoming the problem by classifying between Halal and non-Halal food based on deep learning algorithm, as the number of Malaysians who travel overseas is increasing every year. Convolutional Neural Networks (CNN) deep learning method is used to recognize and classify the images into Halal and non-Halal products due to its higher accuracy on image classification. The images of product packaging are downloaded from the Google Image, augmented, resized into a 100 x 100 pixels dataset, and injected into the model by using python with package of TensorFlow. The images are taken from various products, which are available in Malaysia, to train the CNN model as a prototype. The images of overseas product packaging are expected to be added into the model for further development. A testing set, independent from training set, which are taken by camera phone, is used to test the accuracy of the trained CNN model. Multiple CNN models have been trained by tuning the number of layers and other related parameters to reach the optimum architecture for this project. CNN architecture used in this project is compromises of three convolution layers and three max pooling layers, then followed by a fully connected layer and a softmax layer. The prototype has achieved more than 90% accuracy on classifying the images of the food packaging. |
|---|