For the application requirements of multi-band and perfect absorber, a tri-band absorber based on metasurface is designed in this paper. The proposed absorber consists of a metallic pattern layer, an intermediate dielectric layer and a metallic bottom plate. The results show that the absorber has absorptivity of 99.98%, 99.94% and 99.98% at three resonant frequencies of 5.16 GHz (C-band), 7.27 GHz (C-band) and 8.32 GHz (X-band), which can achieve almost perfect absorption. The origin of three resonant frequencies is analyzed by impedance matching theory. The analysis of the surface current, electric field and magnetic field distributions are simulated at resonant frequency to explain the tri-band absorption mechanism. The regulation of single band, dual band and tri-band can be obtained by changing the number of metallic patterns on the top layer of the absorber. Moreover, the dependence of the absorptivity on polarization angle and incidence angle is studied, and the results show that the absorber has good polarization insensitivity and wide-angle stability (0°—60°). The proposed tri-band absorber based on metasurface has the characteristics of easy design and thin and light, and has great application potential in the fields of electromagnetic absorption, stealth, and sensing.

When segmenting urban street view images in parital Transformer network, multi-scale features and context information in the network are not fully utilized, leading to defects such as holes in large targets and imprecise edge segmentation of small targets. In this paper, a Trans-AsfNet method based on Transformer architecture is proposed to extract multi-scale features and aggregate context information to solve this problem. The segmentation method introduces Swin Transformer as a new feature extraction network to strengthen the long-distance dependence of information. An adaptive subspace feature fusion module (ASFF) is proposed to strengthen the network's ability to extract multi-scale features, and an effective global context aggregation module (EGCA) is designed to improve the context information aggregation capability of the network, and uses rich multi-scale information for feature decoding and information compensation. Then, the context information of different scales is aggregated to strengthen the semantic information of the understanding target, so as to eliminate the holes of large targets and improve the edge segmentation accuracy of small target pixels. The Trans-AsfNet method is verified and tested by the CamVid urban street view dataset, and the experimental results show that the network can basically eliminate the segmentation hole defects and improve the segmentation effect of small target edges, and the MIoU reaches 69.5% on the CamVid test set.

To better understand the interior walls of the water conveyance tunnel, panoramic images of underwater structures' surface defects are obtained at the cost of resolution. However, the lower resolution often falls short of meeting monitoring requirements. To address the conflict between resolution and image acquisition, a bio-inspired S-FREAK underwater image stitching algorithm is proposed. By simulating the vision system of the underwater creature "horseshoe crab," the algorithm enhances image with adaptive lateral inhibition, highlighting its architectural features, considering the characteristics of low signal-to-noise ratio and low contrast of underwater images . Additionally, the algorithm introduces the fast retina keypoint (FREAK) module, emulating human retina characteristics through scale-invariant feature transform (SIFT), to improve the resolution of key feature points. Finally, random sample consensus (RANSAC) feature filtering and fade in and out fusion methods correct the stitching images. Experimental results show that the enhanced adaptive lateral inhibition mechanism increases the matching logarithm of effective feature points, significantly improves stitching accuracy, and optimizes the final outcome.

Object tracking is often difficult to achieve good tracking performance in complex scenes, such as brightness changes, background interference and fast movement. Therefore, we propose an object tracking algorithm that combines infrared and visible light with adaptive feature fusion and an attention mechanism to improve tracking performance. By leveraging the complementary strengths of infrared and visible light, we enhance the performance of traditional object tracking algorithms in complex scenes. To achieve this, we first employ an attention mechanism in the initial three convolution layers to select relevant features from both the infrared and visible modalities. Simultaneously, we dynamically allocate weights to the features of different channels, enabling adaptive feature fusion. Subsequently, the features from different channels are fused, and the object is tracked using the instance classification module. Experimental results obtained from the GTOT dataset and RGBT234 dataset demonstrate the effectiveness of our proposed algorithm. The accuracy and success ratio achieved 90.4% and 73.2% on the GTOT dataset, and 79.6% and 56.1% on the RGBT234 dataset, respectively. These results surpass those of current mainstream algorithms.

Aiming at the problem of stereoscopic image quality prediction bias, a lightweight stereoscopic image quality assessment method combining peripheral visual information is proposed based on the human eye vision model. First, a binocular perception model is constructed to acquire the central concave visual area and the peripheral visual area, and a symmetrical stereoscopic information fusion (SSIF) module is used to enhance the parallax information. Then, the binocular quality perception features are obtained by the lightweight feature extraction (LWFE) module. Finally, the relationship between the subjective and objective stereoscopic image quality evaluation values maps is realized in the fully connected layer. An adaptive multi-loss strategy is introduced to guide the model training, while the performance tests are conducted in LIVE 3D and the Waterloo IVC stereoscopic image databases. The results show that the proposed algorithm performs well comprehensive and maintains a high level of consistency with human subjective quality perception.

In order to improve the recognition rate of the traditional local binary pattern (LBP) algorithm when extracting target image features, a feature extraction method based on mask iterative region of interest (ROI) to improve the LBP algorithm is proposed. The extraction method using mask iterative ROI reduces the processing of interference information or invalid regions and shortens the extraction time of defective region. Based on the LBP, the circular area of the said central pixel point is determined according to the preset radius, the gray value size relationship between the neighboring sampling points is added into the consideration, and together with the central threshold, it is used as the influence factor to decide the LBP coding situation, and the directional features hidden between the neighboring points are fully utilized to further improve the accuracy of image recognition, and the experimental results show that using the PASCAL VOC gear defect dataset as the validation sample, the defect images captured in the experiment show a 2% improvement with SVM recognition accuracy compared with traditional LBP algorithm, with a maximum recognition rate of 99.32%. The Manhattan recognition accuracy improves by 0.67% compared with traditional LBP algorithm, with a maximum recognition rate of 98.54%. The European recognition accuracy improves by 0.44% compared with traditional LBP algorithm, with a maximum recognition rate of 97.87%.

The principal component analysis network (PCANet) is a kind of deep subspace network based on the simplified architecture of convolutional neural network. To address the issue that PCANet cannot process image samples in real-time during the convolutional kernel extraction process, this article proposes an incremental sequential row-column 2DPCA network (IRC2DPCANet). This method can process training samples on time in the process of filter training, which can improve the efficiency of network training. The experiments on three typical face datasets, which is PIE, AR and Yale, indicate that this method has good classification performance. Finally, the influence of the filter number and filter size on classification rate is also investigated.

In order to improve the X-ray intensity of wire target driven by picosecond laser, a method of backlighting of static step wedge target by X-ray source generated by picosecond laser irradiating side radiation of self-sustaining independent wire target is proposed firstly, and experimental research is carried out on picosecond beam of the XGIII laser facility, China Academy of Engineering Physics. In this method, a picosecond laser is used to incident at 30° away from the target normal, and the X-ray source is generated by interacting of a laser with molybdenum and gold wire targets with a diameter of 10 m. The X-ray source focal spot is measured by a multi-hole array pinhole and a CCD camera. The minimum transverse size of the focal spot of the X-ray source obtained is about 20 m, and the longitudinal size is about 10 m. The optimal spatial resolution of static target imaging is close to ~11 m, which is comparable to the diameter of the wire target. The experimental results show that using the self-sustaining independent wire target side can obtain strong penetrability microfocus X-ray source. This method can provide reference for the research and application of high-energy density material dynamic process diagnosis.

The hybrid asymmetrically clipped optical orthogonal frequency division multiplexing (HACO-OFDM) technique only utilizes the odd subcarriers and the imaginary part of the even subcarriers for symbol transmission, leading to the problem of low spectral efficiency. To address this problem, an adaptively biased HACO-OFDM (AB-HACO-OFDM) method is proposed. In the proposed method, a pulse-amplitude modulated discrete multitone (PAM-DMT) signal is designed for the unused real part of the even subcarriers, and is further superimposed on the HACO-OFDM signal for hybrid transmission to enhance the subcarrier utilization. Meanwhile, a power-efficient adaptive bias signal is introduced to guarantee the non-negativity. The simulation results demonstrate that compared to HACO-OFDM, the proposed AB-HACO-OFDM is capable of effectively improving the spectral efficiency of the transmission, while at the same time, achieving better performance in bit-error rate (BER). Additionally, compared to the conventional orthogonal frequency division multiplexing (OFDM) methods, the proposed method requires much lower signal-to-noise ratio to achieve the BER of target, thus exhibiting higher power efficiency.

In order to solve the optical power limitation problem of the transmitter of the radio-over-fiber (RoF) communication system and to improve the reliability of information transmission, a LDPC-PS-64QAM scheme is proposed by combining low-density parity-check (LDPC) codes and probabilistic shaping (PS) techniques to optimise the transmission characteristics of RoF systems. The scheme uses the sign reversal method to probabilistically shape the 64-quadrature amplitude modulation (QAM) signal, while the LDPC code specified by 3GPP is used as the channel coding scheme. The simulation results show that at a symbol rate of 10 Gbaud, the bit error rate (BER) of uniform distribution (UD) and PS signals are compared, and the transmitting power of PS signal is reduced by 1.4 dB compared with UD signal, and the transmitting power of the transmitter is further reduced by using LDPC code; at the fiber length of 20 km, the optical signal-to-noise ratio (OSNR) of PS signal is reduced by 2.1 dB compared with that of UD signal at the same BER, and it has a better noise resistance. The result shows that the LDPC-PS-64QAM system has higher reliability by changing the transmission length of the fiber.

Aramid fiber-reinforced polymer (AFRP) has the advantages of high modulus, high specific strength, and light weight, and has important applications in the aerospace field. In order to overcome the problems in AFRP machine, abrasive water jet and continuous laser processing, such as hair drawing, delamination and large heat-affected zone (HAZ), we carried out the experimental study of ultra-short pulse laser processing AFRP. The ablation threshold of AFRP samples was measured experimentally, and the effects of pulse flux, laser frequency and scanning speed on the HAZ and taper were discussed. The results show that the single-pulse ablation threshold of AFRP processed by 1 030 nm and 230 fs femtosecond laser is 0.815 3 J/cm2, and the pulse accumulation factor is 0.598 7. The HAZ increases and the taper decreases with the increase of pulse flux. When the scanning speed is 500 mm/s, the pulse flux is 4—6 J/cm2, and the pulse frequency is 100—200 kHz, the HAZ and taper can be balanced, and the precision machining effect of HAZ ≤ 20 m, taper ≤2 °and smooth layer without hair drawing, delamination on the inner wall can be obtained. This study provides a feasible technical approach for precision machining AFRP and has potential application value.

Delayed optical feedback semiconductor lasers are the most widely used chaotic light sources in recent years. However, its unpredictability evaluation is still an urgent issue to be addressed. From the information-theoretic point of view, the Kolmogorov-Sinai (KS) entropy calculated from the linearized differential equation can accurately evaluate the unpredictability of chaotic laser signals. However, evaluating the KS entropy of experimental data is very difficult, if not impossible. In this paper, we demonstrate that the Cohen-Procaccia (CP) entropy can be used as a practical alternative method for quantifying the unpredictability of chaotic signals generated from delayed optical feedback semiconductor lasers experiments after optimizing the embedding dimension d, the normalized vertical resolution N, and the sampling time .

Alzheimer's disease (AD) is an irreversible degenerative disease of the central nervous system. In view of the shortcomings of AD diagnosis methods, such as strong subjectivity, time-consuming, invasive, non-invasive, high accuracy, real-time diagnosis of near-infrared spectroscopy (NIRS) analysis has great advantages. In this paper, the spectral information of 265 serum samples that have been diagnosed as healthy and AD was collected, spectral analysis and principal component analysis (PCA) were carried out, and finally combined with aquaphotomics analysis. The results showed that in the first three PCA scores of patients with AD, PC1 tended to be negative, which was different from that of healthy people. In combination with aquaphotomics analysis, it was found that the absorbance of AD patients at the place where the water spectral pattern (WASP) was multi-hydrogen bond and OH stretching was higher than that of normal people, and the probability of the existence of macromolecular groups of water in serum was higher. On the contrary, the absorbance of normal people in free water with single hydrogen bond, hydration and no hydrogen bond is significantly higher than that of AD patients, and the probability of small molecular groups of water in serum is higher. Therefore, the diagnosis method of AD based on NIRS combined with aquaphotomics is feasible.

Brain tumor magnetic resonance imaging (MRI) segmentation is an important step in the diagnosis and treatment of brain tumors. In this paper, a brain tumor MRI segmentation algorithm that integrates multi-scale features is proposed to address the low segmentation accuracy caused by the limited receptive field size of the U-Net network structure and the gap in contextual information. Firstly, a multi-scale aggregation module (MAM) is designed to replace the conventional convolutional layers in the original U-Net network. This increases the depth and width of the network to capture detailed boundary information of the feature maps. Secondly, the context atrous spatial pyramid module (CASP) is used in the skip connection instead of direct concatenation operation. This expands the network's receptive field and enhances the extraction ability of lesions at different scale sizes. Finally, a multi-level aggregation attention module (MAA) is designed at the bottom of the U-shaped network. This module enables the network model to focus on effective features in the image segmentation region and exclude background noise. The improved algorithm is experimentally validated on the Cancer Genome Atlas (TCGA) (brain tumor data) database. The results demonstrate that the proposed algorithm achieves the following metrics: mean intersection over union (mIoU) of 91.39%, Dice coefficient of 92.81%, sensitivity of 89.14%, specificity of 99.95%, and accuracy of 95.78%.