Evaluating sharpness functions for automated scanning electron microscopy

Automatic quantification and classification of microplastics in scanning electron micrographs via deep learning

Microplastics quantification and classification are demanding jobs to monitor microplastic pollution and evaluate the potential health risks. In this paper, microplastics from daily supplies in diverse chemical compositions and shapes are imaged by scanning electron microscopy. It offers a greater depth and finer details of microplastics at a wider range of magnification than visible light microscopy or a digital camera, and permits further chemical composition analysis. However, it is labour-intensive to manually extract microplastics from micrographs, especially for small particles and thin fibres. A deep learning approach facilitates microplastics quantification and classification with a manually annotated dataset including 237 micrographs of microplastic particles (fragments or beads) in the range of 50 μm-1 mm and fibres with diameters around 10 μm. For microplastics quantification, two deep learning models (U-Net and MultiResUNet) were implemented for semantic segmentation. Both significantly outmatched conventional computer vision techniques and achieved a high average Jaccard index over 0.75. Especially, U-Net was combined with object-aware pixel embedding to perform instance segmentation on densely packed and tangled fibres for further quantification. For shape classification, a fine-tuned VGG16 neural network classifies microplastics based on their shapes with high accuracy of 98.33%. With trained models, it takes only seconds to segment and classify a new micrograph in high accuracy, which is remarkably cheaper and faster than manual labour. The growing datasets may benefit the identification and quantification of microplastics in environmental samples in future work.

Image definition assessment based on Tchebichef moments for micro-imaging

This paper proposes a Tchebichef moment (TM)-based image definition assessment (IDA) method that employs the difference in the logarithmic spectra (DLS). To avoid the influence of the original image, the essential element point spread function (PSF) is extracted from the DLS to characterize the IDA function uniquely. The amplification of the PSF spot radius to the defocus amount in the micro-imaging system enhances the featural differences among the DLSs, thereby improving the sensitivity to the defocus amount. The DLS with an obvious geometric feature variation is described by a TM with a low order, which improves the anti-noise performance. The performed simulation and experiment verified the superiority of the proposed method.

Study of an image autofocus method based on power threshold function wavelet reconstruction and a quality evaluation algorithm

As a key component in optical microscopy imaging systems, autofocus technology has a significant effect on imaging quality. In this paper, an optical microscopy autofocus method that includes a wavelet denoising algorithm based on a power threshold function and a Brenner image quality evaluation algorithm is presented. Experimental results show that the power threshold function wavelet denoising algorithm, which can be adopted to obtain more realistic optical images, is superior to the traditional soft, hard, hyperbolic, and exponential threshold functions in terms of peak signal-to-noise ratio, signal-to-noise ratio, mean squared error, and histogram indicators; moreover, compared to the Roberts, sum modulus difference (SMD), and energy gradient functions, the Brenner image quality evaluation algorithm can be used to quickly and accurately lock onto the focal plane. By integrating and applying these two core algorithms in the autofocus image acquisition system of a microscope, the image sharpness and focusing quality are greatly improved, which benefits the further evaluation of images.

Manipulations of Wavefront Propagation: Useful Methods and Applications for Interferometric Measurements and Scanning

Phase measurements obtained by high-coherence interferometry are restricted by the 2π ambiguity, to height differences smaller than λ/2. A further restriction in most interferometric systems is for focusing the system on the measured object. We present two methods that overcome these restrictions. In the first method, different segments of a measured wavefront are digitally propagated and focused locally after measurement. The divergent distances, by which the diverse segments of the wavefront are propagated in order to achieve a focused image, provide enough information so as to resolve the 2π ambiguity. The second method employs an interferogram obtained by a spectrum constituting a small number of wavelengths. The magnitude of the interferogram's modulations is utilized to resolve the 2π ambiguity. Such methods of wavefront propagation enable several applications such as focusing and resolving the 2π ambiguity, as described in the article.

Autofocus on moving object in scanning electron microscope

The sharpness of the images coming from a Scanning Electron Microscope (SEM) is a very important property for many computer vision applications at micro- and nanoscale. It represents how much object details are distinctive in the images: the object may be perceived sharp or blurred. Image sharpness highly depends on the value of focal distance, or working distance in the case of the SEM. Autofocus is the technique allowing to automatically adjust the working distance to maximize the sharpness. Most of the existing algorithms allows working only with a static object which is enough for the tasks of visualization, manual microanalysis or microcharacterization. These applications work with a low frame rate, less than 1 Hz, that guarantees a low level of noise. However, static autofocus can not be used for samples performing continuous 3D motion, which is the case of robotic applications where it is required to carry out a continuous 3D position measurement, e.g., nano-assembly or nanomanipulation. Moreover, in addition to constantly keeping object in focus while it is moving, it is required to perform the operation at high frame rate. The approach offering both these possibilities is presented in this paper and is referred as dynamic autofocus. The presented solution is based on stochastic optimization techniques. It allows tracking the maximum of the sharpness of the images without sweep and without training. It works under uncertainty conditions: presence of noise in images, unknown maximal sharpness and unknown 3D motion of the specimen. The experiments, that were performed with noisy images at high frame rate (5 Hz), were conducted on a Carl Zeiss Auriga 60 FE-SEM. They prove the robustness of the algorithm with respect to the variation of optimization parameters, object speed and magnification. Moreover, it is invariant to the object structure and its variation in time.

Combining gradient ascent search and support vector machines for effective autofocus of a field emission-scanning electron microscope

Autofocus is an important issue in electron microscopy, particularly at high magnification. It consists in searching for sharp image of a specimen, that is corresponding to the peak of focus. The paper presents a machine learning solution to this issue. From seven focus measures, support vector machines fitting is used to compute the peak with an initial guess obtained from a gradient ascent search, that is search in the direction of higher gradient of focus. The solution is implemented on a Carl Zeiss Auriga FE-SEM with a three benchmark specimen and magnification ranging from x300 to x160 000. Based on regularized nonlinear least squares optimization, the solution overtakes the literature nonregularized search and Fibonacci search methods: accuracy improvement ranges from 1.25 to 8 times, fidelity improvement ranges from 1.6 to 28 times, and speed improvement ranges from 1.5 to 4 times. Moreover, the solution is practical by requiring only an off-line easy automatic train with cross-validation of the support vector machines.