In this paper, we overview the previously reported underwater signal detection system using 1D integral imaging convolutional neural networks (1DInImCNN). The 1DInImCNN system comprises cameras arranged in a one-dimensional configuration for optical signal collection and the 1DInImCNN approach for signal detection. The 1D camera array is used to capture the spatial and temporal information, encoded using Gold code and transmitted by a Light-emitting Diode (LED). Various turbidities and occlusions are created in a water tank to test the performance of the proposed method under such degradations. The 1DInImCNN method is compared to the previously proposed 3D integral imaging (3D InIm) with Convolutional neural network (CNN) and Bi-Long Short-term memory (Bi-LSTM) approach. The results suggest that the 1DInImCNN-based approach outperforms the previously proposed 3D InIm with the CNN-BiLSTM approach in terms of computation costs and detection performance.
KEYWORDS: Fiber optic gyroscopes, Object detection, Sensors, Active remote sensing, 3D image processing, Integral imaging, LIDAR, Thermal sensing, Education and training, 3D surface sensing
This paper presents an overview of a previously published work on the performance comparison of different sensors
(Visible, LWIR, and LiDAR-based imaging systems) for the task of object detection and classification in the presence of
degradation such as fog and partial occlusions. Three-dimensional integral imaging has been shown to improve the
detection accuracy of object detectors operating in both visible and LWIR domains. As fog affects the image quality of
different sensors in different ways, we have trained deep learning detectors for each sensor for 2D imaging as well as 3D
integral imaging to compare the performance of sensors in the presence of degradation such as fog and partial occlusions.
We present an overview of previously reported Single Random Phase Encoding (SRPE) and Double Random Phase Encoding (DRPE) optical bio-sensing systems. In contrast to traditional imaging modalities that rely on lenses to capture and magnify subjects, SRPE and DRPE employ phase masks to modulate the light field emanating from an object. This modulation results in a pseudo-random optical signal to be received at the sensor, which is then classified by an appropriate classification algorithm. This lensless paradigm not only reduces the physical bulk and expense associated with optical components but provides wide field of view, and enhanced depth of field in comparison with lens-based imaging system. In biomedical imaging, the application of SPRE and DRPE systems has significant promise in the context of distinguishing between various types of red blood cells (RBCs) for disease diagnosis. Specifically, these imaging systems have demonstrated remarkable efficacy in identifying horse and cow RBCs, as well as differentiating between sickle cell-positive and negative RBCs with high accuracy and robustness to noise. The integration of Convolutional Neural Networks (CNNs), when trained directly on captured opto-biological signature (OBS) images show significant robustness to noise. Training a CNN on the Local Binary Patterns (LBP) of captured OBS images has shown not only improved classification performance but also maintained accuracy under conditions of significant data compression.
In this work, we present an overview of previously published work on the identification of COVID-19 red blood cells (RBCs) and sickle cell disease based on the reconstructed phase profile using a deep learning framework. The video holograms for thin blood smears were recorded using a compact, low-cost, and field portable, 3D-printed shear-based digital holographic system. Individual cells were segmented from the holograms and then each frame was reconstructed to extract spatio-temporal signatures of the cells. Morphology-based features along with motility-based features extracted from reconstructed phase images, were fed to a bi-LSTM to classify between COVID-19 positive and healthy red blood cells. Based on the majority of the cell subjects were classified as healthy or diseased.
Over the last years structured illumination digital holographic microscopy (SI-DHM) has been experimentally proved to double the resolution limit in conventional DHM. In SI-DHM, the underlying specimen is illuminated using a spatially varying structured illumination (SI) pattern, which enables super-resolution (SR) images to be retrieved using the proper computational reconstruction process. All these reconstruction methods require the acquisition of at least a couple phase-shifted DHM images. In particular, for a pure sinusoidal pattern, there is a need of recording two phase-shifted DHM images per orientation of the pattern (e.g., 6 images per isotropic SR improvement). Taking advantage of the simultaneous recording of the virtual (e.g., conjugated) image in the raw DHM image, here we present a novel computational method to reconstruct an isotropic SR image using one acquisition per pattern’s orientation (e.g. total 3 images per isotropic improvement). Because our proposed method shows a 50% reduction in the data acquisition and, therefore, acquisition time, we believe that our method should increase the utility of SI-DHM in live-cell imaging. We have validated our method using simulated and results.
We overview our recently published multi-dimensional integral imaging-based system for underwater optical signal detection. For robust signal detection, an optical signal propagating through the turbid water is encoded using multiple light sources and coded with spread spectrum techniques. An array of optical sensors captures video sequences of elemental images, which are reconstructed using multi-dimensional integral imaging followed by a 4D correlation to detect the transmitted signal. The area under the curve (AUC) and the number of detection errors were used as metrics to assess the performance of the system. The overviewed system successfully detects an optical signal under higher turbidity conditions than possible using conventional sensing and detection approaches.
In this keynote address paper, we overview recently published works on the current techniques and methods for automated cell identification with 3D optical imaging using compact and field portable systems. 3D imaging systems including digital holographic microscopy systems as well as lensless pseudorandom phase encoding systems are capable of capturing 3D information of microscopic objects such as biological cells which allows for highly accurate automated cell identification. Systems based on digital holography enable reconstruction of the cell’s 3D optical path length profile. The reconstructed 3D profiles can be used to extract morphological and spatio-temporal cell features from biological samples for classification and cell identification. Similarly, pseudorandom encoding techniques such as single random phase encoding (SRPE) and double random phase encoding (DRPE) can be used to encode 3D cell information into opto-biological signatures which can be used for cell identification tasks. Recent advancements in these areas are presented including compact and field-portable 3D-printed shearing digital holographic microscopy systems, integration of digital holographic microscopy with head mounted augmented reality devices, and the use of spatio-temporal features extracted from cell membrane fluctuations for sickle cell disease diagnosis.
We overview a previously reported system for automated diagnosis of sickle cell disease based on red blood cell (RBC) membrane fluctuations measured via digital holographic microscopy. A low-cost, compact, 3D-printed shearing interferometer is used to record video holograms of RBCs. Each hologram frame is reconstructed in order to form a spatio-temporal data cube from which features regarding membrane fluctuations are extracted. The motility-based features are combined with static morphology-based cell features and inputted into a random forest classifier which outputs the disease state of the cell with high accuracy.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.