Fig. 1.

The proposed implementation of the flow of artificial intelligence-based quantitative upconversion luminescence detection under small samples. The method enables rapid upconversion luminescence quantitative detection with high accuracy, ultra-sensitivity, and strong noise tolerance. (a) The actual developed portable device for upconversion luminescence quantitative detection. (b) Diagram of the hardware structure of the device. (c) Schematic diagram of UCNP-LFAs. (d) Implementation scheme for constructing a training database using a small number of samples. (e) Implementation of the data augmentation process. (f) The workflow for implementing transfer learning into the pre-trained network. (g) Deployment of trained AI models to local devices. (h) Majority Voting strategy, which aims at absolute accuracy of the final prediction results. (i) Fast transfer of prediction results to PC or mobile interfaces through real-time inference.