Table 1.
Descriptions of major food certifications used in the United States and selected for training in the study.
Fig 1.
Process of an object detection model development for certification mark recognition on front-of-package images.
The overall process can be divided into two distinct stages: model training and inference. During the training stage, the training dataset is collected, pre-processed, labeled with the certification marks, and used to train the model. During the inference stage, pre-processed infant food products’ images are used as the test dataset to grasp the certification status.
Fig 2.
Process of nutrition and health-related text analysis for the front-of-package images using optical character recognition.
From pre-processed infant food product images, the texts are extracted using OCR, then processed and filtered using nutrition and health-related databases. Following this, the nutrition and health-related texts are used for frequency analysis, and the association of co-occurring health-related texts is further identified by network analysis.
Table 2.
Performance of the certification mark recognition model based on object detection developed using Google’s AutoML Vision.
Table 3.
Status of using certifications for infant food products.
Table 4.
Frequencies of nutrient-related texts used for the front of packages identified by optical character recognition-based text analysis.
Table 5.
Frequencies of health-related texts used for the front of packages identified by optical character recognition-based text analysis.
Fig 3.
Co-occurrence network of health-related categories for baby food products.