In this work, we present a deep-learning architecture to augment the sensing capabilities of a micro electro mechanical system (MEMS) for fluid mechanical applications. The MEMS sensor is composed of a polyvinylidene fluoride flexible piezoelectric flag and a bluff body, converting three-dimensional fluid mechanical energy flow into a timedependent voltage signals. The developed deep learning method allows for extracting accurate wind speed and classifying turbulences. The bluff body generates vortexes which are not only the functions of the bluff body’s geometry but also related to the fluid speed. These vortexes induce a mechanical vibration into the attached piezoelectric flag and hence generate charge displacement and an electrical voltage signal. By placing the mentioned setup in a wind tunnel, we excited the structure with various wind speeds while different combinations and geometries of the bluff body and piezoelectric flag were considered. An unsupervised autoencoder was used to extract the continuous manifold in Fourier spectra of time domain voltage generated by the piezoelectric sensor when the wind speed is continuously changed from 0 to 33 meters per second. We found that this manifold is highly correlated with wind speed. By adding another feed-forward network in parallel to the decoder network of the autoencoder we also incorporated the measured wind speed as the data’s label and could successfully use the system to extract the wind speed, despite the sensor was placed under the strong turbulence generated by the bluff body. We also investigated the ability of our deep learning method to classify different bluff bodies from the voltage harvested from the piezoelectric flag, finding that this unique cability is resilient to the wind speed in the range. Such a system can be turned into a system that fingerprints different turbulence and uniquely differs them for various applications.

Machine learning has had a significant impact on the value of spectroscopy-based characterization tools, particularly in biomedical applications, due to its ability to detect latent patterns within complex spectral data. However, it often requires extensive data preprocessing, including baseline correction and denoising, which can lead to unintentional bias during classification. To address this, we present a deep learning-based signal preprocessing method capable of handling all the defects of raw Raman spectroscopy data without any need of human intervention. To achieve this, a novel deep convolutional neural network (CNN) architecture was trained on randomly generated spectra with various defects. We demonstrate that the proposed network results in faster training and that it can perform complete spectral preprocessing in a single step with more accuracy, speed, and defect tolerance than conventional methods. These improvements make it an ideal candidate for hyperspectral imaging applications in which tens of thousands of raw spectra may need to be processed rapidly. The superiority of this method is demonstrated for simulated Raman spectra, surface-enhanced Raman spectroscopy (SERS) imaging of chemical species, classification of low resolution Raman spectra of human bladder cancer tissue, and finally classification of SERS spectra from human placental extracellular vesicles (EVs). These findings encourage the future use of deep learning as a rapid and unbiased method of preprocessing spectroscopy data and may be particularly useful in biomedical applications involving large data sets from highly heterogeneous samples and signal defects of complex nature.

Machine learning has shown great potential for classifying diverse samples in biomedical applications based on their Raman spectra. However, the acquired spectra typically require several preprocessing steps before standard machine learning algorithms can accurately and reliably classify them. To simplify this workflow and enable future growth of this technology, we present a unified solution for classifying biological Raman spectra without any need of prepossessing, including denoising and baseline establishment. This method is developed based on a custom version of a convolutional neural network (CNN) elicited from ResNet architecture, combined with our proposed data augmentation technique. The superiority of this method compared to conventional classification techniques is shown by applying it to Raman spectra of different grades of bladder cancer tissue and surface enhanced Raman spectroscopy (SERS) spectra of various strains of E. Coli extracellular vesicles (EVs). These results show that our method is far more robust compared to its conventional counterparts when dealing with the various kinds of spectral baselines produced by different Raman spectrometers.

Bacterial extracellular vesicles (EVs) are nanoscale lipid-enclosed packages that are released by bacteria cells and shuttle various biomolecules between bacteria or host cells. They are implicated in playing several important roles, from infectious disease progression to maintaining proper gut health, however the tools available to characterise and classify them are limited and impractical for many applications. Surface-enhanced Raman Spectroscopy (SERS) provides a promising means of rapidly fingerprinting bacterial EVs in a label-free manner by taking advantage of plasmonic resonances that occur on nanopatterned surfaces, effectively amplifying the inelastic scattering of incident light. In this study, we demonstrate that by applying machine learning algorithms to bacterial EV SERS spectra, EVs from cultures of the same bacterial species Escherichia coli can be classified by strain, culture conditions, and purification method. While these EVs are highly purified and homogeneous compared to complex samples, the ability to classify them from a single species demonstrates the incredible power of SERS when combined with machine learning, and the importance of considering these parameters in future applications. We anticipate that these findings will play a crucial role in developing the laboratory and clinical utility of bacterial EVs, such as the label-free, noninvasive, and rapid diagnosis of infections without the need to culture samples from blood, urine, or other fluids.