First Scientific statement and report

Current scientific research have proved that Raman spectrometry can be used to reliably test the composition of various materials, including food products. SFF team has decided to kickstart the project by scanning a few selected food products that have gain the high publicity in recent years — pure and mixed samples of meat of various animals, fish products and formaldehid, and various vegetable oil types.

To ensure the integrity of the research, the samples are collected from well known and reliable sources, like national laboratories, carefully prepared at the university laboratory and scanned with high precision scientific Raman spectrometer. In the process, multiple samples are prepared from pure and mixed specimen and multiple scans are conducted on each single sample to make measurements as statistically reliable as possible. The data bookkeeping system is designed so that all the possible specimen and sample metadata is preserved along with the spectra.
Once the Raman spectra is ready, statistical and data pre-processing analysis are carried out. Our team of AI researchers design a number of Machine Learning models in order to train them to recognize the slightest food product patterns, for pure type and quantity in a mixed product recognition. Models are carefully verified by using the testing dataset, which was not used during the training in order to simulate the actual production environment.

At the moment of this article, AI models are able to recognize 5 vegetable oil types with the accuracy of more than 98% (error rate is less than 2%), meat types of 8 animals with more than 96% accuracy (error rate is less than 4%) and meat mixes at the rate of 90% with precision of 20%. Samples are being prepared and scanned 24/7 and the model performance is subject to improvement as it absorbs more data.