A Machine Learning Approach to Recognize Junk Food

Abstract

People are naturally food grains. So they are always looking for appetizing food and junk foods are the largest source of it. Recent time the most observable point is that peoples are attracted to outdoor foods than homemade foods. As a result our dietary is changing and we are leaning to junk food day by day which causes a bad effect on our health and increases the risk of health disorders. Machine learning facts are being used in every inch of our life and recognizing object through image processing is one of those. Although, the reason of foods being different in nature make this process critical where traditional approaches will be the cause of a poor accuracy rate. Deep Neural Network has outperformed all of these problems. In this study, we tried to recognize local junk foods based on a new dataset consist of 2000 data belonging 5 junk food classes. All the data in the data set were collected using Smart-phone camera and believed to be unique in every sense. Convolution Neural Network (CNN) technology was used to reach the goal which is renowned for image processing. Throughout the study we achieve an accuracy of 90.47% which turned out to be satisfying. Furthermore we did test based on real-life scenario and the result was out of the mark. Our ambition is to take this study to the next level which will be implemented on another study later on. Designing such system that will prevent the society not to take junk food and be conscious about health is our final goal.