Single-pixel imaging, which allows imaging with a single-pixel detector and a correlation method, can be accelerated by combining machine learning. In addition, the accuracy of the estimation was improved using the uncertainty of the estimated value by machine learning. The machine-learning algorithm was constructed from a physical perspective based on errors in the measurement system. On the other hand, to improve the reliability of the machine learning estimates, the uncertainty of the estimates was evaluated using standard deviation values derived by data augmentation. By using the value with the lowest uncertainty as the final estimate, we improved machine learning and achieved measurements with a small number of illuminations.
Recently, there has been a high demand for high-quality imaging from weak light in measuring micro-defects. Deep Learning Ghost Imaging (DLGI) has been proposed as a fast and sensitive imaging method for defect inspection. However, measurement with deep learning has a problem evaluating the prediction uncertainty. The predicted value from deep learning is distributed close to the true value in the feature, while the traditional measurement value is physically distributed close to the true value. Then, applying the conventional uncertainty evaluation method based on statistics is difficult. To overcome this problem, we propose the evaluation method of the prediction uncertainty based on the feature map in the middle layer of the CNN. By adding random numbers to the middle layer, several close estimates of feature values can be obtained. The standard deviation of these estimates is defined as prediction uncertainty. This paper shows the numerical comparison of the proposed method with evaluation by data augmentation, which evaluates the prediction uncertainty by adding fluctuations to the input data. The data augmentation method can estimate the uncertainty of changes in measurement conditions. Although the data augmentation method does not provide enough change for low SNR data, which makes uncertainty evaluation difficult, the proposed method offers constant fluctuation even for low SNR data. We have numerically confirmed that the proposed method can accurately evaluate the prediction uncertainty even for low SNR.
In demand for minute defect inspection, it is required to detect weak scattered light caused by small defects. Ghost imaging (GI) is
known for its high sensitivity and high noise resistance method. However, it requires many measurements to obtain a high-quality image
because GI is the correlation-based imaging method. Reducing the number of measurements, a method combined with deep learning has
been proposed. In order to improve the estimation accuracy using CNN, we propose to parallelize the convolutional layers. Parallel
convolutional layers can efficiently extract both local and global features, which contributes to the improvement of estimation accuracy.
In this report, we show that parallel CNN is more accurate than conventional CNN by experiments.
We propose a novel imaging method using the Ghost Imaging (GI) in a photon limited imaging. In this report, in order
to obtain image using few photons, the First Photon-detection Time (FPT) is applied to the GI. The proposed signal detection
method is able to obtain a signal with only 1 photon. The FPT has variations that are due to a shot noise. Using a correlation,
effect of the noise on images is removed. As a result, the GI with FPT (FPGI) was able to obtain high quality image than a
conventional imaging method using same photon number. Furthermore, to improve the detection time, we modified
machine learning to reduce the measurement number in the view point of noise influence.
This paper reports for high-sensitivity imaging based on Ghost imaging (GI), which is one of the single-pixel imaging. Although the GI is correlation-based imaging between structured illumination lights and detected signals, there is an advantage in detecting weak light intensity such as fluorescence of molecules. Especially, in the case of using extream weak light intensity, a photon signal is useful for imaging. Therefore, we focused on the arrival time of the first photon and used the time as the intensity of the signal. Furthermore, to improve the detection time, we applied machine learning to reduce the measurement number. In this paper, we have proposed the principle and some experimental results.
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.