Non-Halal Ingredients Detection of Food Packaging Image Using Convolutional Neural Networks

Halal food is absolute for Muslims. Unfortunately, many packaged foods have not been certified halal. One way of looking at the ingredients is listed. However, often the code of composition is unfamiliar. Current image processing and machine learning technologies can help read the image of the packaged food composition for the detection of non-Halal compositions. The development supports this and using Android smartphones. However, it is often constrained by the size of the font and position of packaged foods uneven. Deep learning one of the developments in machine learning that has a high level of recognition. It was used to detect the Halal composition of packaged food using convolutional neural networks (CNNs). Image of packaged food had been segmented each character before recognized by CNN's. The CNN models using the 250 epoch and 64 batches tested to obtain optimal parameters and configuration, including the use of an adaptive learning rate method (AdaDelta) and Stochastic Gradient Descent (SGD) configuration. The accuracy of each character is 98.08% if using SGD and 98.02% if using AdaDelta. Non-Halal detection combined character to the word and compared with the list of non-Halal compositions. The segmentation process significantly determined the accuracy. It made the overall word accuracy of only 50%, whose most significant error lied in the wrong segmentation process.