An image quality assessment method based on multi-task self-supervised learning is proposed to address the existing deep learning-based image quality assessment methods, which suffer from overfitting and insufficient generalization performance due to insufficient labeled data. First,17 distortion type images are synthesized by the algorithm and the full reference mean deviation similarity index (MDSI) score and distortion type are used as 2 labels for the synthesized distortion images. Subsequently, multi-task self-supervised learning on vision transformer (ViT) for predicting MDSI scores and distortion types. Finally, the trained model is fine-tuned on the downstream task to migrate the semantic features learned from the upstream task to the downstream task. The method in this paper is fully compared with mainstream no reference image quality assessment (NR-IQA) methods on several publicly available image quality assessment datasets, and the test results on LIVE, CSIQ, TID2013, and CID2013 are all improved by 1 to 2 percentage points compared with the best performing algorithms, which indicates that the proposed algorithm outperforms most mainstream unreferenced image quality assessment algorithms.

To enhance the multi-scale learning capacity of target detection algorithms, particularly for small targets, this paper proposes a super-resolution and multi-scale fusion target detection algorithm based on an improved YOLOv5 framework. Firstly, instead of the up-sampling operation of the original YOLOv5 model, the algorithm utilizes sub-pixel convolution to enhance the image resolution and preserve the information of small targets to the greatest extent possible. Secondly, the algorithm utilizes the parallel fast multi-scale fusion (PFMF) module to achieve two-way fusion of deep and shallow features. This upgrade from the original YOLOv5 algorithm's 3-scale prediction to 4-scale prediction improves the model's ability to learn multi-scale features and detect small targets. The experimental results demonstrate that compared with YOLOv5s, the improved model achieves a 2.8% and 3.5% increase in mAP @0.5 and mAP @0.5∶0.95, respectively, on the PASCAL VOC dataset. Similarly, on the MS COCO dataset, the improved model achieves a 4.3% and 5.2% increase in mAP @0.5 and mAP @0.5∶0.95, respectively. The experiments demonstrate the improved YOLOv5 model's enhanced capability in multi-scale detection, particularly for small targets, and indicate its potential practical value.

Blurred image restoration is an important task in the field of computer vision and image processing. In view of the problem of score function has relatively small differences and weak optimization function of mind evolutionary algorithm (MEA) in image restoration based on combination of MEA and wavelet neural network (WNN), an improved MEA-WNN image restoration model is proposed. The difference between score function is increased by using power law transformation and logistic regression function, therefore, the selecting function of MEA is enhanced significantly. Comparative experiments are conducted between improved and traditional WNN and MEA-WNN-based image restoration model, the improved model can increase the peak signal-to-noise ratio (PSNR) by 15% and 6.5%, and structural similarity (SSIM) by 6.1% and 5% respectively. Effectiveness and superiority of improved model is proved by some experimental results.

To further improve the brightness, contrast and definition of the low light image effectively, a low light image enhancement method based on weighted plateau histogram equalization is proposed. Taking full advantage of that the lightness component V is independent of the hue H and the saturation S in the HSV color space, this method converts the image to HSV color space, and by the bilateral filtering which has good edge-preserving ability, the lightness component V is decomposed into illumination image L and reflection image R with Retinex algorithm. The illumination image L is subjected to the weighted dual plateau histogram equalization, in which, the upper plateau threshold and the lower plateau threshold are determined adaptively by the principle of 3 in normal distribution, and the weighting coefficient is inversely proportional to the histogram frequency of the gray level. The experimental results show that compared with some existing methods, the effect of image enhanced by the proposed method is better, and the corresponding information entropy and average gradient are higher than the existing methods by more than 0.35 and 12, respectively, which proves that the proposed method has better low light image enhancement performance.

In order to improve the speed of lane line detection in autonomous driving, a method of feature extraction using convolutional neural network and classification network to realize the classification of virtual and solid lane lines is proposed. An efficient residual factorized ConvNet (ERFNet) is used to perform convolution operations and down sampling on images, the network adopts a bottleneck free one-dimensional convolution residual structure, utilizes vertical and horizontal one-dimensional convolution interpolation to enhance the generalization ability of nonlinear functions, obtains multi-scale contextual information based on variable fill ratios, to achieve feature extraction of images. After deconvolution and up sampling, the features are decoded and the image scale is restored, and finally the segmented image information is output. Compared to traditional semantic segmentation algorithms, this method can reduce a large number of feature parameters, enhance the learning ability of the model, and ensure detection accuracy while improving detection speed. The simulation experiments under conditions such as straight driving, turning, uphill, downhill, bumpy roads, and uneven lighting show that the detection accuracy of this method can reach 95.14%, and the detection speed is improved compared to mainstream algorithms.

In order to study the effects of different dispersion models on the ellipsometry spectra analysis of TiO2 films containing pores, TiO2 thin films are prepared by sol-gel method, and five dispersion models are used to fit the ellipsometry spectra of TiO2 films in the wavelength range of 1.55—4 eV. The fitting results of each model are verified by the oblique reflection spectra. The results show that the selection of different dispersion models has an effect on the fitting results of film thickness and porosity, and the fitting results of refractive index dispersion are obviously affected by the dispersion model. New-Amorphous, Tauc-Lorentz and Adachi-New Forouhi models are suitable for the ellipsometry spectra fitting of sol-gel TiO2 film on the whole test band. However, the Cauchy Absorbent and Sellmeier Absorbent models cannot get a good ellipsometric spectral fitting for the whole test band, and the applicable band is narrow. The results provide a reference for the ellipsometry spectra analysis of sol-gel TiO2 thin films containing pores.

Marine risers are important structures in marine engineering. In order to ensure their normal operation, this paper proposed a method that uses weak fiber Bragg grating sensing cables to complete shape reconstruction of the riser and uses the finite element simulation experiments to verify it. Firstly, the sensing cables were installed along the axis of the riser in the form of a backpack-style pipeline. Based on the Frenet-Serret framework, an algorithm for reconstructing the shape of the riser was designed. Then, a finite element model for monitoring risers was constructed, and strain data inside the sensing optical fiber cable various deformation conditions were extracted by solving the model. By combining algorithms, the three-dimensional shape of the riser was reconstructed. Finally, an analysis and calculation of the shape reconstruction error were performed. The results indicate that the end error of the designed riser shape reconstruction method was controlled within 1.7%. The method has a good effect on the reconstruction of the riser shape, with a simple monitoring form and certain engineering application value.

A prediction model for electric load based on an immune support vector machine (SVM) algorithm is proposed to address the issues of high randomness and poor stability in the electric load of residential areas. Considering various factors that affect the electric load of residents, the historical electric consumption and relevant climate data of residential areas are used as the processing objects. The principal component analysis (PCA) algorithm is utilized to preprocess the historical data of the power grid, and the immune algorithm is combined to preprocess the data by forming data clusters and defining labels for training the prediction model. To improve the accuracy of the model, the biological immune optimization algorithm is used to optimize the parameters of the SVM model. In the load prediction process, the prediction error is used as the basis for feedback tuning of the prediction model. The prediction performance of the immune SVM algorithm load prediction model is compared with that of the commonly used back propagation (BP) neural network and SVM algorithm model. The short-term and medium-term prediction accuracies of the immune SVM algorithm load prediction model are both above 98%, demonstrating good accuracy and robustness.

A phase reconstruction method using a rhombic-Gaussian combined filtering window is proposed in order to solve the problems of susceptibility to noise, spectral leakage, and crosstalk from other order spectra in the extraction of differential phase using conventional filtering windows in quadriwave lateral shearing interference (QLSI) phase reconstruction. By combining a rhombic window on the spectrum plane with a series of two-dimensional Gaussian windows perpendicular to the spectrum plane, the differential phases in two orthogonal directions are extracted from the QLSI interferogram. Finally, the phase to be measured is reconstructed from the two differential phases by a least square based Fourier transform method. Experimental measurements are implemented under a standard sample to compare the proposed rhombic-Gaussian composite window with four conventional filtering windows on the effects of the reconstructed phase in terms of the accuracy of the phase difference, root mean square error (RMSE), and peak-to-valley (PV) error. The results demonstrate that the phase difference of the phase retrieved by the proposed method is closest to the nominal value of the sample, and the RMSE error and PV error of the reconstructed phase are both the minimum, thus effectively improving the quality of phase reconstruction.

The existing unsupervised deep learning algorithms based on auto-encoders and generative adversarial networks have problems such as poor generalizability, high missed and false detection rates in the defect detection task of yarn-dyed fabric. To address these issues, a yarn-dyed fabric defect detection algorithm based on U-shaped attention gate auto-encoder (UAGAE) is proposed. Firstly, the light weight network EfficientNet-B6 is employed as the feature extraction module to capture more representative features from input images. The introduced attention gate (AG) mechanism is used to suppress feature responses in non-target regions, leveraging decoder features as a reference to eliminate redundant information in skip connections, thereby aiding in image reconstruction. Subsequently, during the training phase, a combined loss function is utilized to preserve both the structure and details of the reconstructed images. Finally, during the detection phase, the ultimate detection results are obtained through adaptive threshold segmentation and mathematical morphology operations. The proposed algorithm achieves a precision (P) of 53.45%, recall (R) of 61.58%, F1-measure (F1) of 53.63%, and mean intersection over union (IoU) of 40.83% on the public dataset YDFID-1. Notably, it attains the highest F1 and IoU metrics across 14 different fabric patterns. The comparative experimental results indicate that the UAGAE algorithm, in comparison to several other defect detection algorithms, exhibits a superior capability in effectively performing yarn-dyed fabric defect detection and localization.

Quantum positioning sstem (QPS) has the characteristics of high positioning accuracy and high safety factor. However, sandstorm can cause great changes in optical quantum entanglement degree, which will further affect the positioning error of quantum positioning system. In this case in order to reduce the positioning error, an optimal entanglement degree adaptive (OEDA) strategy to resist sandstorm interference is proposed based on switching strategy of dual-satellite satellite-ground link. The relationships between each parameter of the sandstorm, transmission distance and the entanglement degree are established. And the positioning error caused by the global feature of sandstorm is compared before and after the adaptive adjustment of the system. The simulation results show that when the overall characteristic aggregation factor of sandstorm is 10 and the transmission distance is 10 km, the system positioning error is reduced from 0.14 m to 0.02 m by using OEDA algorithm. Thus it is known that the accuracy of QPS in sandstorm can be improved by using OEDA strategy.

To investigate the motion and radiation backward symmetry of high-energy electrons in linearly polarized ultrashort ultrastrong laser pulses, the electron trajectory, the angular distribution of spatial radiation energy, and the histogram of the spatial frequency distribution is plotted with the aid of numerical simulation software, based on the framework of classical nonlinear Thomson scattering. According to the research, nonlinear Thomson scattering is significantly influenced by the initial phase of the laser when the laser pulse is only a few cycles long. Also, its moving trajectory and spatial energy distribution which have apparent correlations have a "triple-symmetric" property concerning the initial phase of driving laser. The cutoff of the high harmonic spectrum can reach 1×1060, and the symmetry of high harmonic amplitude peaks provides a feasible idea of the selection of laser parameters in actual experiments. The study also provides a viable approach to obtain the desired spatial angle distribution of radiation energy and high harmonic amplitude peak of inverse nonlinear Thomson scattering by initial envelop phase modulation of the laser pulse.

The modulation of vortex light field is essential for the application of vortex light in laser communication, biological manipulation and other fields. Based on Fresnel diffraction theory, we numerically simulate the intensity, phase and spiral spectrum distribution of Laguerre-Gaussian (LG) beams focus field with various lens spherical aberration coefficients in order to reveal the aberration effect on the propagation characteristic of LG vortex beams. The results show that the increase of the primary spherical aberration coefficient of the optical system deteriorates the intensity distribution and spiral phase of LG beam focus field, and disperses the orbital angular momentum (OAM). Therefore, the aperture compensation scheme is proposed, in which the lens aberration is eliminated by placing an appropriate-size circular aperture in front of a lens. After the aperture compensation, the focus field of the LG beam is restored from the star-shaped hollow distribution to the hollow ring distribution, and the OAM state remains single and stable. This work provides a method for improving the lens aberration on vortex light transmission effect.

Keratoconus is a progressive corneal disease that mostly occurs in adolescence and can cause irregular astigmatism and vision loss. Late-stage blindness requires corneal transplantation. Therefore, early and accurate screening of keratoconus is necessary to prevent the progression of the disease and avoid deterioration. As a classic algorithm, neural network is an effective tool for keratoconus diagnosis. However, as the data of keratoconus cases grows day by day, in order to make full use of the new data, it is often necessary to retrain all samples, which will consume a lot of time. In order to solve the above problems, this article proposes an incremental learning algorithm integrating neural networks to achieve intelligent diagnosis of keratoconus. In addition, this article also introduces the ideas of undersampling and cost sensitivity to solve the problem that existing incremental learning algorithms cannot handle imbalanced data. Experimental results show that the recognition accuracy of the algorithm proposed in this article reaches 97%, and requires short training time and less storage space. Therefore, this algorithm can assist in the diagnosis of keratoconus more efficiently.

The Er3+ doped laser gain medium around 2.7—3 m with unique advantages has important applications in the fields of biomedicine, environmental detection and nonlinear optics, etc. However, the lower-level lifetime 4I13/2 is longer than that of upper-level lifetime 4I11/2, resulting in a high laser threshold, severe thermal effect and other problems. Some researchers devote to exploring laser gain media with low threshold and high beam quality by doping deactivate ions, thermal bonding, single crystal fiber (SCF), cascade laser, low photon energy and so on. Compared with other gain mediums in the same waveband, the Er3+ doped laser gain mediums have been rapidly developed and laser performance has also been improved obviously, which can be expected to further develop in its application field. In this paper, we summarize the research progress of the Er3+-doped laser crystal, such as YAG (Y3Al5O12), YSGG (Y3Sc2Ga3O12) and YAP (YAlO3), the Er3+-doped ceramics such as Y2O3, Lu2O3, the Er3+-doped zirconium fluoride glass such as ZABLAN (ZrF4-BaF2-LaF3-AlF3-NaF), and the relevant results and experimental protocol are highlighted. At the same time, we also look forward to the future development direction.