Location: Children's Nutrition Research CenterTitle: Automatic food detection in egocentric images using artificial intelligence technology Author
|Jia, Wenyan - University Of Pittsburgh|
|Li, Yuecheng - University Of Pittsburgh|
|Qu, Ruowei - University Of Pittsburgh|
|Baranowski, Tom - Children'S Nutrition Research Center (CNRC)|
|Burke, Lora - University Of Pittsburgh|
|Zhang, Hong - Beihang University|
|Bai, Yicheng - University Of Pittsburgh|
|Mancino, Juliet - University Of Pittsburgh|
|Xu, Guizhi - Hebei University|
|Mao, Zhi - University Of Pittsburgh|
|Sun, Mingui - University Of Pittsburgh|
Submitted to: Public Health Nutrition
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 2/9/2018
Publication Date: 3/26/2018
Citation: Jia, W., Li, Y., Qu, R., Baranowski, T., Burke, L.E., Zhang, H., Bai, Y., Mancino, J.M., Xu, G., Mao, Z.H., Sun, M. 2018. Automatic food detection in egocentric images using artificial intelligence technology. Public Health Nutrition. http://dx.doi.org/10.1017/S1368980018000538.
Interpretive Summary: The first step in processing all day images from a passive approach to assessing diet (i.e. requires no effort on the part of a participant, as in wearing a camera all day that takes images of everything in front of the person every few seconds) is to identify those images that include images of food. Artificial intelligence (AI) is a set of methods that can be used to identify food in images, but the AI algorithms must be trained with large data sets of images that have been pre-identified as having food or not, and then tested for accuracy against another set of images. This is often done by splitting a set of pre-identified images into random halves and training the AI on one half and testing it on the other. In one such set of 3900 images, the AI attained a level of 91.5% and 86.4% accuracy in identifying just foods. In a second data set with 29,515 images, the AI attained levels of 85% sensitivity (i.e. it correctly identified 85% of with food) and 88.8% specificity (i.e. correctly identified 85.8% of the images as not having food) in identifying food. These are important first steps, but new types of AI or more training images will be needed to improve AI accuracy.
Technical Abstract: Our objective was to develop an artificial intelligence (AI)-based algorithm which can automatically detect food items from images acquired by an egocentric wearable camera for dietary assessment. To study human diet and lifestyle, large sets of egocentric images were acquired using a wearable device, called eButton, from free-living individuals. Three thousand nine hundred images containing real-world activities, which formed eButton data set 1, were manually selected from thirty subjects. eButton data set 2 contained 29 515 images acquired from a research participant in a weeklong unrestricted recording. They included both food- and non-food-related real life activities, such as dining at both home and restaurants, cooking, shopping, gardening, housekeeping chores, taking classes, gym exercise, etc. All images in these data sets were classified as food/non-food images based on their tags generated by a convolutional neural network. A cross data-set test was conducted on eButton data set 1. The overall accuracy of food detection was 91.5 and 86.4%, respectively, when one-half of data set 1 was used for training and the other half for testing. For eButton data set 2, 74.0% sensitivity and 87.0% specificity were obtained if both 'food' and 'drink' were considered as food images. Alternatively, if only 'food' items were considered, the sensitivity and specificity reached 85.0 and 85.8%, respectively. The AI technology can automatically detect foods from low-quality, wearable camera-acquired real-world egocentric images with reasonable accuracy, reducing both the burden of data processing and privacy concerns.