Objective:
The obesity epidemic is widespread not only in adulthood, but also in childhood; in Europe, especially in Italy and Greece, despite being native countries of the Mediterranean diet, the rate of childhood obesity is extremely high. In order to better understand and address this problem, it is essential to collect real-world data on eating habits, a process which can be facilitated by smartphone image collection, annotation and analysis.

Methods:
Children aged 9 to 18 years from 23 public schools in Thessaloniki used the “myBigO App” in order to capture their dietary intake, by taking pictures of most of their meals for two weeks. For this study, 2744 pictures of snacks were processed and annotated using VGG Image Annotator for determining the food category and whether they meet healthy eating guidelines. The quality of snack consumption (ultra-processed vs natural foods) was also evaluated through the NOVA classification system.

Results:
40.7% of children consumed snacks that meet healthy eating guidelines. An additional 18.1% of the snacks was determined as healthy depending on context and, more importantly, quantity. Analysis of the types of snacks, showed that the most consumed snacks were fruit and/or natural fruit juices (37%), followed by consumption of baked products (27%), sweets (23%) and salty snacks (13%). An alarming amount (50.4%) of snacks were classified as ultra-processed.

Conclusions:
These results can contribute to improve national food and nutrition education policies, with the aim of creating effective programs for reducing the prevalence of childhood obesity and overweight. Additionally, the resulting food image dataset with annotated food categories and processing level can be used towards creating new automated image annotation processes.