Learning a global underwater geolocalization model with sectoral transformer.

Polarization-based underwater geolocalization presents an innovative method for positioning unmanned autonomous devices beneath the water surface, in environments where GPS signals are ineffective. While the state-of-the-art deep neural network (DNN) method achieves high-precision geolocalization based on sun polarization patterns in same-site tasks, its learning-based nature limits its generalizability to unseen sites and subsequently impairs its performance on cross-site tasks, where an unavoidable domain gap between training and test data exists. In this paper, we present an advanced Deep Neural Network (DNN) methodology, which includes a neural network built on a Transformer architecture, similar to the core of large language models such as ChatGPT, and integrates an unscented Kalman filter (UKF) for estimating underwater geolocation using polarization-based images. This combination effectively simulates the sun's daily trajectory, yielding enhanced performance across different locations and quicker inference speeds compared to current benchmarks. Following thorough analysis of over 10 million polarization images from four global locations, we conclude that our proposed technique significantly boosts cross-site geolocalization accuracy by around 28% when contrasted with traditional DNN methods.

Supercontinuum intrinsic fluorescence imaging heralds free view of living systems.

Optimal imaging strategies remain underdeveloped to maximize information for fluorescence microscopy while minimizing the harm to fragile living systems. Taking hint from the supercontinuum generation in ultrafast laser physics, we generated supercontinuum fluorescence from untreated unlabeled live samples before nonlinear photodamage onset. Our imaging achieved high-content cell phenotyping and tissue histology, identified bovine embryo polarization, quantified aging-related stress across cell types and species, demystified embryogenesis before and after implantation, sensed drug cytotoxicity in real-time, scanned brain area for targeted patching, optimized machine learning to track small moving organisms, induced two-photon phototropism of leaf chloroplasts under two-photon photosynthesis, unraveled microscopic origin of autumn colors, and interrogated intestinal microbiome. The results enable a facility-type microscope to freely explore vital molecular biology across life sciences.

Angle of polarization calibration for omnidirectional polarization cameras.

Polarization cameras quantify one of the fundamental properties of light and capture intrinsic properties of the imaged environment that are otherwise omitted by color sensors. Many polarization applications, such as underwater geolocalization and sky-based polarization compass, require simultaneous imaging of the entire radial optical field with omnidirectional lenses. However, the reconstructed angle of polarization captured with omnidirectional lenses has a radial offset due to redirection of the light rays within these lenses. In this paper, we describe a calibration method for correcting angle of polarization images captured with omnidirectional lenses. Our calibration method reduces the variance of reconstructed angle of polarization from 76.2 ∘ to 4.1 ∘. Example images collected both on an optical bench and in nature, demonstrate the improved accuracy of the reconstructed angle of polarization with our calibration method. The improved accuracy in the angle of polarization images will aid the development of polarization-based applications with omnidirectional lenses.

Optical inspection of nanoscale structures using a novel machine learning based synthetic image generation algorithm.

In this paper, we present a novel interpretable machine learning technique that uses unique physical insights about noisy optical images and a few training samples to classify nanoscale defects in noisy optical images of a semiconductor wafer. Using this technique, we not only detected both parallel bridge defects and previously undetectable perpendicular bridge defects in a 9-nm node wafer using visible light microscopy [Proc. SPIE9424, 942416 (2015)], but we also accurately classified their shapes and estimated their sizes. Detection and classification of nanoscale defects in optical images is a challenging task. The quality of images is affected by diffraction and noise. Machine learning techniques can reduce noise and recognize patterns using a large training set. However, for detecting a rare "killer" defect, acquisition of a sufficient training set of high quality experimental images can be prohibitively expensive. In addition, there are technical challenges involved in using electromagnetic simulations and optimization of the machine learning algorithm. This paper proposes solutions to address each of the aforementioned challenges.