Rapid and high accuracy identification of culture medium by CNN of Raman spectra

Culture media are widely used for biological research and production. It is essential for the growth of microorganisms, cells, or tissues. It includes complex components like carbohydrates, proteins, vitamins, and minerals. The media's consistency is key for predictable outcomes in biology applications. However, traditional methods of analyzing media are costly and time-consuming by using chromatography or mass spectrometry. This study introduces an innovative approach using optimized convolutional neural networks (CNN) combined with Raman spectroscopy to identify culture media. Samples of culture media from different models and batches are prepared for identification experiment. Raman spectra of each culture media samples are captured with unique molecular vibrations and rotations by Raman spectrometer rapidly. After preprocessing of sample data, Raman spectra are input to CNN for identification training and validation. An optimized CNN with more layers is designed to enhance the identify ability for Raman spectra. In experiment, it compared the performance of PCA-SVM, the original CNN, and an optimized CNN for media identification. The PCA-SVM achieved high accuracy and precision rates of 99.19% and 98.39% respectively. The original CNN achieved an accuracy of 71.89% due to limited training dataset. The optimized CNN model achieved a perfect accuracy rate of 100% in identifying different culture media. To avoid overfitting risk, additional external test is performed with optimized CNN. The result confirmed that optimized CNN offering effectiveness in identifying media from different models and batches, with strong generalization ability. The findings in study may offer an efficient and cost-effective method for pharmaceutical companies, to ensure the consistency of culture media.