Aiming at improving the sensitivity of optical fiber wavelength multiplexing acceleration sensor,the sensitization idea of fiber Bragg grating (FBG) acceleration sensor is analyzed and the sensitivity bottlenecks and inherent constraints of FBG acceleration sensor are clarified.The physical model of wavelength multiplexing fiber Fabry-Pérot (F-P) acceleration sensor is proposed,the principle of acceleration sensing is researched theoretically,and the analytical expressions of the resonance frequency and the sensitivity of the sensor are derived.The influence factors of structural stiffness,interference order,and cavity length of the system on the sensitivity and resonant frequency of the sensor are analyzed in depth,and the sensitivity response characteristics of the FBG and F-P acceleration sensors under different system stiffness are compared and analyzed.Because sensitivity and resonant frequency limit each other,furthermore,the figure of merit is introduced to compare and analyze the comprehensive performance of the FBG and F-P acceleration sensors.When the resonance frequency is 205 Hz,the sensor sensitivity is up to 198 nm/G,higher than FBG sensor sensitivity about 2 magnitude theoretically,and put forward the F-P type acceleration sensor wavelength multiplexing scheme.The theoretical analysis results show that the F-P type acceleration sensor has unique advantages,compared with the FBG type,which provides a new idea for the sensitization and wavelength multiplexing of the optical fiber type acceleration sensor,and lays a theoretical foundation.

A dynamic scene deblurring algorithm based on multi-scale dense connections and U-Net improvement is proposed to address the issues of texture detail loss, inability to suppress noise, and generation of ringing artifacts in existing motion blur removal networks during image restoration. First, the receptive field is effectively expanded by using the hole convolution downsampling in the U-Net network to avoid irreversible damage to the image without increasing the number of parameters, and the sub-pixel convolution is used to obtain clear image details with a small convolution kernel in the upsampling process, reducing the computational complexity; Secondly, a multi-scale dense feature extraction (MDFE) module is designed to enhance deep level feature extraction and reuse through densely connected convolutional layers. Spatial pyramid pooling (SPP) branches are used to guide the transfer and fusion of multi-scale features, promoting effective preservation of image details and textures; Finally, the ConvLSTM bidirectional connectivity structure is used to compensate for contextual features of simple cascading loss from the encoding path in a nonlinear manner, promoting cross stage interaction of deep features, and weakening edge artifacts and noise interference. Compared with existing advanced methods, the performance advantages of the algorithm proposed in this paper have been verified.

To address the shortcomings of the current white-box adversarial attacks against object detection about imperceptibility,this paper proposes a stealthy adversarial perturbations generation (SPG) method from the perspective of both generation process and the limitation of perturbations cost.First,the perturbations in the high-texture region of images which is hard to detected by human eyes are assigned a higher weight by texture information.Then,a perturbations position selection strategy is applied to reduce the number of improved perturbed pixels.Finally,these adversarial perturbations are decoupled to adaptively search for the best L2 norm metric.The proposed method and comparative methods are evaluated on the MS-COCO 2014 and PASCAL-VOC datasets against 4 dominant object detectors.Experimental results show that the metric value of imperceptibility of this method is greater than that of other methods.The mAP of the object detectors is degraded to below 9%.The L0 norm is less than 0.239,and the L2 norm of adversarial perturbations is less than 2.9×10-5 respectively.

Aiming at the problem that the existing image splicing detection network model has insufficient attention to edge information and is not good enough for pixel-level accurate localization effects,a DeepLabV3+ image splicing tampering forensic method incorporating a residual attention mechanism is proposed.The methods use an encoding-decoding structure to achieve pixel level image splicing tampering localization.In the coding stage,the efficient attention module is integrated into the residual module of ResNet101.The residual module is stacked to reduce the proportion of unimportant features and highlight the splicing tampering traces.Then,the spatial pyramid pooling module with hole convolution is used for multi-scale feature extraction.The obtained feature maps are stitched and then modelled by spatial and channel attention mechanisms for semantic information.In the decoding stage,the localization accuracy of the image splicing forgery region is improved by fusing multi-scale shallow and deep image features.The experimental results show that the localization accuracy of splicing tampering on CASIA 1.0,COLUMBIA and CARVALHO datasets reaches 0.761,0.742 and 0.745,respectively.The proposed method has better image splicing forgery region localization performance than some existing methods,and the network also has better robustness to JPEG compression.

In order to solve the traditional image fusion deficiencies,such as excessive edge smoothing,loss texture details,low contrast,non-prominent target and missing source image information,this paper proposes an infrared and visible dual-band image fusion algorithm based on non-subsampled shearlet transform (NSST).Firstly,the source infrared and visible images are enhanced through adaptive guided filter (AGF).Secondly,the infrared and visible images are decomposed into low and high frequency components by NSST,respectively.Then,the low frequency components are fused by using the local adaptive intensity (LAI) rule,while high frequency components are fused by using dual channel adaptive pulse coupled neural network (DCAPCNN).Finally,the fused image is reconstructed by using the inverse NSST.Experimental results show that the proposed method has advantages in appropriate contrast,reserving the infrared target characteristic,including more background edge and texture detail information,and fusion image with high signal-noise ratio,the infrared and visible image advantage are effectively combined,compared with the existing traditional and deep learning fusion algorithms,the proposed algorithm achieves better experimental results,with superior performance in both subjective visual perception and objective indicator evaluations.

In order to extract the text information from fuzzy inspection images efficiently and accurately,this paper proposes a combination of maximally stable extremal regions (MSER) algorithm based on edge enhancement and multi-feature adaptive weight fusion with immunogenetic (IGA) optimization support vector machine (SVM) to extract text from fuzzy inspection images.The edge-enhanced MSER algorithm is used to detect the image text,and merge the detected MSERs to get the text candidate regions.In order to filter out non-text region in the candidate regions,the feature fusion formula is used to fuse the three features of the image,then the IGA is used to optimize the SVM classifier to find the optimal parameters,and finally the candidate regions are fed into the trained classifier to filter out non-text.The experimental results show that the algorithm has higher true positive rate and lower false positive rate compared with other algorithms,higher accuracy for fuzzy inspection image text extraction.

Agricultural diseases can cause early defoliation of crops and weakened photosynthesis,which can affect crop quality and reduce farmers′ incomes.Aiming at the problem of target miss detection exception caused by small targets, complex background and unstable natural light during the initial occurrence of the diseases,this paper proposes a YOLOv4 detection algorithm that integrating lightweight networks.Firstly,the trunk network is simplified and multi-scale group convolution is enhanced to improve the anti-interference ability of the mode in the complex backgrounds.Secondly,the lightweight space channel expand (SCE) attention mechanism is designed to reduce the impact of detail information loss in the deep network.Finally,the pyramid structure with the feature of skip connection is applied for the replacement of integration module with path aggregation network (PAnet) feature to further realize the model lightweight.Experimental results show that the improved algorithm reaches 84.17% of mAP50 and the detection speed is 50 FPS in the dataset of this paper,which is 0.71% and 10 FPS higher than that of YOLOv4 detection algorithm,that meets the requirements of the detection accuracy and speed of agricultural diseases on the mobile devices.

Polarization dependent of laser-induced nanostructures is an effective technique to realize nano-pattering and has been favored by researchers.The polarization-dependent characteristics of laser-induced nanostructures on 6H-SiC crystal surface are investigated by the method of femtosecond laser micromachining.The spherical nanoparticles with a diameter of about 150 nm,the elliptical nanoparticles and the periodic ripple structures with the period of 150 nm have been fabricated on sample surface by controlling the laser machining parameters of polarizations and time delays.The results show that the polarization properties of incident laser directly affect the induced microstructure morphologies,and the femtosecond laser with preferential incident plays a decisive role in the final surface micro-nano structures.The physical mechanism of polarization-dependent formation of laser-induced nanostructures is preliminarily discussed,the surface plasmon polariton (SPP) plays an important role in the generation of surface micro-nano structures,and the research results are of great significance for the controllable fabrication of laser-induced periodic surface structures (LIPSS).

The trap inside the organic-inorganic lead halide perovskite (OLHP) semiconductor materials is crucial factor for its photoelectric properties.In order to understand the effect of trap on carrier recombination process in polycrystalline methylamine bromide perovskite ((Methylammonium)PbBr3,MAPbBr3) films,time resolved microwave conductivity (TRMC) technology is applied.The experimental results show that both free carrier recombination and trapped carrier thermally emission recombination occur to polycrystalline MAPbBr3 films.The energy level related to trapped carrier thermal emission recombination isolate from continuous band, and their central energy depth and distribution width are 0.6 eV and 89.2 meV,respectively.Excitation wavelength varying TRMC experiments are also used to differentiate shallow trapped electrons and electrons in conduction band.The experiments confirm that shallow trapped electrons are more likely to transition to deep trapped states compared with electrons on coduction band.

Aiming at the problem of the high computational complexity of the cross-component linear model (CCLM) in versatile video coding (VVC) intra prediction, this paper proposes an improved algorithm,quick cross-component linear model (QCCLM) based on the CCLM technology.First,the position and number of sub-sampling samples are fixed according to copy adjacent available samples to fill unavailable samples,and remove the redundant processes and additional calculation steps;Then,the luminance down-sampling optimization process is used to reduce the types of down-sampling filters;Finally,the derivation process of the linear model parameter β is improved so as to make the prediction model more accurate.The experimental results show that compared with the standard algorithm of H.266,the algorithm saves 0.14% of the code rate on the chrominance component of the test image sequence in the all intra frames configuration,and the total coding time is reduced by 4.05% on average.The algorithm improves coding performance while reducing coding complexity.

In order to balance the relationship between structure and performance in chaotic map and ensure the security of the encryption system,a block image encryption algorithm based on cosine-exponential chaotic map is proposed.Firstly,a new cosine-exponential chaotic map is constructed by modulating the cosine map with the introduction of the Tent seed map through the non-linear exponential term,and the SHA-256 function is used to generate the key associated with the plain-text,generating chaotic sequences with strong randomness to achieve one-time pad.Then,based on the Latin square and bit-level transformation,the double Latin square and extended bit algorithm are designed through two rounds of Latin square indexing and bit stitching,respectively,and combined with two-dimensional Josephus sequences,to scramble the inter-block pre-scrambled plain-text within blocks,achieving differential scrambling of different blocks.Finally,based on the Zig-Zag transform,the cross Zig-Zag transform method is designed using a circular imitation Zig-Zag transform,which diffuses the intermediate cipher-text with the chaotic sequence nonlinearly in both directions to achieve simultaneous change of pixel position and pixel value to complete the image encryption.The experimental results show that the algorithm has a large key space and can effectively resist typical attacks such as differential analysis and statistical analysis,and has a good encryption effect.

Aiming at the problems of blurred details, incomplete energy preservation and long running time in traditional medical image fusion,a medical image fusion method based on hybrid filtering and improved edge detection pulse coupled neural network (PCNN) in non-subsampled shearlet transform (NSST) domain is proposed.Firstly,the YUV model is used to perform a color space conversion to separate the luminance channel Y,and then compound filter is used to enhance the source MRI image and the grayscale image of the luminance channel in different degrees.Secondly,the grayscale images of the enhanced magnetic resonance imaging (MRI) and luminance channels are decomposed using NSST to obtain the high and low frequency subbands.The low-frequency subband uses a fusion strategy with a modified Laplace energy sum and a local area energy weighted sum,and the high-frequency subband uses an improved edge detection PCNN fusion strategy.Finally,the fused images are obtained by NSST inverse transformation.By comparing with other six fusion methods,this method can effectively improve the detail extraction and energy preservation in the process of image fusion,and the overall algorithm operates with high efficiency and good visibility.