Multi-class segmentation of unlabelled living cells in time-lapse light microscopy images is challenging due to the temporal behaviour and changes in cell life cycles and the complexity of these images. The deep-learning-based methods achieved promising outcomes and remarkable success in single- and multi-class medical and microscopy image segmentation. The main objective of this study is to develop a hybrid deep-learning-based categorical segmentation and classification method for living HeLa cells in reflected light microscopy images. A symmetric simple U-Net and three asymmetric hybrid convolution neural networks—VGG19-U-Net, Inception-U-Net, and ResNet34-U-Net—were proposed and mutually compared to find the most suitable architecture for multi-class segmentation of our datasets. The inception module in the Inception-U-Net contained kernels with different sizes within the same layer to extract all feature descriptors. The series of residual blocks with the skip connections in each ResNet34-U-Net’s level alleviated the gradient vanishing problem and improved the generalisation ability. The m-IoU scores of multi-class segmentation for our datasets reached 0.7062, 0.7178, 0.7907, and 0.8067 for the simple U-Net, VGG19-U-Net, Inception-U-Net, and ResNet34-U-Net, respectively. For each class and the mean value across all classes, the most accurate multi-class semantic segmentation was achieved using the ResNet34-U-Net architecture (evaluated as the m-IoU and Dice metrics).

Two carbon fiber reinforced type IV pressure vessels are subjected to step-wise pressurization until burst, while monitored using acoustic emissions (AE). Unlike most prior studies, AE data is collected throughout the entire damage progression. The vessels, manufactured with differing parameters, failed in distinct composite layers – A-type in the hoop layers and B-type in the helical layers. The AE signals are evaluated to study material degradation and identify fiber breaks as signs of critical damage accumulation. The signals are distributed randomly across the surface, with localized accumulation only minutes before rupture, close to the rupture plane. The difference in manufacturing parameters did not result in any clear difference in the AE activity. Felicity and Shelby ratios show consistent decline with increasing pressure, suggesting potential for damage assessment and burst prediction. It is discussed how these ratios are affected by coupling quality of the AE sensors, the shape of the pressurization profile and prior loadings. Different signal features based on the amplitude and the frequency content are extracted for a classification into failure mechanisms. Based on previous studies, AE signals corresponding to fiber breaks have a characteristic high-frequency spectrum and show a delay in occurrence, with an increase in the number of breaking fibers towards the end of the experiment. Indeed, high frequency signals tend to occur later and signals in specific peak-frequency ranges (350 – 400 kHz, > 500 kHz) somewhat resemble the expected behavior. However, the dataset is too variable and too incongruent for any clear interpretations. Likely reasons are signal propagation effects, the complex composite structure, simultaneous occurrence of signals and measurement uncertainties. A review of relevant studies is provided to show that similar issues affect also previous works. Successfully identifying fiber breaks in large-scale, complex composite structures based on AE data, and turning this into an applicable health-monitoring technique, therefore remains a challenge.

In video-records, objects moving in intracellular regions are often hardly detectable and identifiable. To squeeze the information on the intracellular flows, we propose an automatic method of reconstruction of intracellular flow velocity fields based only on a recorded video of an unstained cell. The basis of the method is detection of speeded-up robust features (SURF) and assembling them into trajectories. Two components of motion—direct and Brownian—are separated by an original method based on minimum covariance estimation. The Brownian component gives a spatially resolved diffusion coefficient. The directed component yields a velocity field, and after fitting the vorticity equation, estimation of the spatially distributed effective viscosity. The method was applied to videos of a human osteoblast and a hepatocyte. The obtained parameters are in agreement with the literature data.