ABSTRACT
In-situ irradiation transmission electron microscopy (TEM) offers unique insights into the millisecond-timescale post-cascade process, such as the lifetime and thermal stability of defect clusters, vital to the mechanistic understanding of irradiation damage in nuclear materials. Converting in-situ irradiation TEM video data into meaningful information on defect cluster dynamic properties (e.g., lifetime) has become the major technical bottleneck. Here, we present a solution called the DefectTrack, the first dedicated deep learning-based one-shot multi-object tracking (MOT) model capable of tracking cascade-induced defect clusters in in-situ TEM videos in real-time. DefectTrack has achieved a Multi-Object Tracking Accuracy (MOTA) of 66.43% and a Mostly Tracked (MT) of 67.81% on the test set, which are comparable to state-of-the-art MOT algorithms. We discuss the MOT framework, model selection, training, and evaluation strategies for in-situ TEM applications. Further, we compare the DefectTrack with four human experts in quantifying defect cluster lifetime distributions using statistical tests and discuss the relationship between the material science domain metrics and MOT metrics. Our statistical evaluations on the defect lifetime distribution suggest that the DefectTrack outperforms human experts in accuracy and speed.
ABSTRACT
Deep learning models can recognize the food item in an image and derive their nutrition information, including calories, macronutrients (carbohydrates, fats, and proteins), and micronutrients (vitamins and minerals). This technology has yet to be implemented for the nutrition assessment of restaurant food. In this paper, we crowdsource 15,908 food images of 470 restaurants in the Greater Hartford region on Tripadvisor and Google Place. These food images are loaded into a proprietary deep learning model (Calorie Mama) for nutrition assessment. We employ manual coding to validate the model accuracy based on the Food and Nutrient Database for Dietary Studies. The derived nutrition information is visualized at both the restaurant level and the census tract level. The deep learning model achieves 75.1% accuracy when compared with manual coding. It has more accurate labels for ethnic foods but cannot identify portion sizes, certain food items (e.g., specialty burgers and salads), and multiple food items in an image. The restaurant nutrition (RN) index is further proposed based on the derived nutrition information. By identifying the nutrition information of restaurant food through crowdsourced food images and a deep learning model, the study provides a pilot approach for large-scale nutrition assessment of the community food environment.
