Engineering and Architecture
Domenec Puig Valls
Computer Science and Mathematics of Security
Intelligent Food Ingredients Quality Control System
Quality of foods has a direct impact on human health. Providing a high quality food is impossible without good ingredients. Human is able to assess quality of food ingredients using intrinsic senses such as Olfactory, Eyesight and Sense of taste. However, using human resources to control quality of food ingredients is time consuming and costly. We aim to develop a system for automatically improving quality of daily foods by checking quality of food ingredients. Specifically, we propose a food quality evaluation system which is able to assess food ingredients using their images. To this end, images of ingredients are acquired using the camera of a cellphone. Then, the system analyses the image using computer vision techniques and classifies the ingredients into different levels of qualities. The objective of quality control system is to estimate values of vitamins, minerals, protein and carbohydrate of food ingredients.
In the first step, images captured by the cellphone are analysed using computer vision methods. In this step, features such as shape, texture and color are extracted from the image. These features are essential for evaluating quality of food ingredients. For instance, features of a fresh tomato are different from an old tomato. Contrary to an old tomato which is red-to-brown, color of a fresh tomato is red. Also, the texture of a fresh tomato is softer compared to an old tomato. These features can be extracted using hand crafted descriptors such as Histogram of Oriented Gradients (HOG), Bank of Gabor Filters, Color Histogram and Local Binary Patterns (LBP). Nonetheless, a better scheme for extracting visual features of food ingredients is to incorporate feature learning methods. In this scheme, features of images are learnt using deep neural networks.
In the second step, image descriptors are classified using machine learning methods. The system categorizes food ingredients into four quality levels including Fresh, Partially old (safe to eat), Totally old (unsafe to eat) and Rotten. Classification stage can be done using Random Forests, Support Vector Machines (SVM) or Artificial Neural Networks. Here, we have focused on deep learning methods which are able to learn a descriptor and a classifier simultaneously from data. Previous studies have showed that deep neural networks are able to provide a richer representation compared to hand crafted techniques. This has helped us to develop systems which have already surpassed human performance in some tasks.
In the last step, nutrient level of food ingredients are estimated using regression models. The value of vitamin, minerals, protein and carbohydrate of food ingredients varies in the four classes and it is reduced from Fresh to Rotten. However, after classifying food ingredients into one of the mentioned classes, the system estimates value of nutrition facts.
37.5 hours a week
|This project has received funding from the European Union's Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement No. 713679|