The original neural network optimization algorithm (NNA) faces issues like weak global search capability, susceptibility to local optima in the later optimization stages, and inadequate convergence accuracy. To address these, we propose a chaotic adaptive search based on an improved neural network optimization algorithm (CASNNA). Firstly, we introduce a logistic chaotic mapping to mitigate uneven population distribution within the solution space, thereby expanding the search space and enhancing the likelihood of finding the global optimal solution. Secondly, a spiral search position update strategy is incorporated to balance global and local searches by dynamically adjusting the spiral shape, thus enhancing the algorithm's search efficiency and accuracy. Finally, an adaptive t-distribution strategy is employed, with a t-distribution mutation disturbance factor added during the later stages of algorithm iteration to escape local optima, thus improving the convergence speed. Simulation experiments were conducted using internationally recognized benchmark test functions to compare CASNNA with other meta-heuristic algorithms. The results demonstrate that the proposed CASNNA algorithm excels in terms of convergence speed and accuracy.

To address the shortcomings of the Weighted mean of vectors algorithm (INFO) in solving complex optimization problems, such as insufficient optimization accuracy and susceptibility to local optima, this paper proposes an Improved Weighted Mean of Vectors Algorithm (IINFO) based on multiple optimization strategies. Firstly, the vector combination stage is optimized with crossbar strategy to increase the diversity of population. Secondly, the algorithm is enhanced by incorporating the foraging phase of the Crayfish Optimization Algorithm, which accelerates the algorithm's convergence speed. Finally, the local search phase was improved by combining the predator avoidance phase of the walrus optimization algorithm to improve the optimization ability. Firstly, the IINFO algorithm was compared with six swarm intelligence algorithms, and then IINFO was compared with other improved INFO algorithms. Finally, the effectiveness of the three improved strategies was verified by ablation experiments. From the analysis of experimental data, the overall performance of IINFO algorithm is better than that of the six swarm intelligence algorithms, and the three improvements can improve the optimization ability and convergence speed of the algorithm to varying degrees.

A hybrid strategy, the improved Subtraction Average Optimizer (HSABO), is proposed to address the shortcomings of the Subtraction Average Based Optimizer (SABO) in handling complex problems such as poor convergence accuracy and susceptibility to local optimal. Firstly, Cubic mapping is applied for population initialization, thereby improving the overall search potential of the optimization algorithm. Secondly, adaptive T-distribution disturbance is introduced in the loop process. Finally, the golden sine strategy is used to correct the global optimum and population. Eight test functions with single-peak and multi-peak characteristics were used for comparison simulation experiments of various algorithms and improved ablation experiments of three strategies. Findings demonstrate that the three refined strategies effectively improve the performance of the conventional SABO algorithm, and the HSABO algorithm excels in performance compared to other algorithms.

To address the issues of slow convergence, low accuracy, and vulnerability to local optima in the Arithmetic Optimization Algorithm (AOA), this paper introduces the multi-strategy fused improved Arithmetic Optimization Algorithm (CLSAOA). Initially, the Circle chaotic mapping is utilized for population initialization, thereby augmenting diversity and balance within the initial population. Then, the lens imaging opposition-based learning strategy is incorporated into the optimization phase of AOA to improve the algorithm's exploration capabilities and its ability to escape local optima, thereby enhancing the convergence accuracy and speed. Finally, a sine-cosine strategy is added to perturb the particle positions, further improving the algorithm's convergence accuracy. Simulation experiments compare with other improved AOA algorithms, the original AOA algorithm, and other metaheuristic algorithms using international benchmark test functions for optimization. The findings indicate that the CLSAOA algorithm exhibits superior performance in terms of both convergence precision and speed.

At present, an all optical high-order tunable ordinary differential equation (ODE) solver is very difficult to implement. A novel all-optical first to third order linear ODEs solutions with tunable constant coefficients using double Sagnac rings coupled Mach-Zehnder-interferometer (DSMZI) on silicon waveguide chips are proposed. The structural composition and size of the DSMZI have been designed, and the working principles of its first to third order ODEs solutions have been derived. By varying the input electric heating power of the thermal-optical phase shifters of the individual arms of the MZI, the constant-coefficient of the differential equation can be simply tuned in large scope. It is demonstrated that the constant coefficient k ranges from 0.0015/ps to 0.092/ps for the first-order ODE. The constant coefficient p of the second-order ODE solver can be continuously tuned from 0.013/ps to 0.174/ps, correspondingly with the q varying from 0.00004225/ps2 to 0.007569/ps2. Three constant coefficients u, v, and w of the third-order ODE can be continuously tuned from 0.105/ps to 0.252/ps, 0.003675/ps2 to 0.021168/ps2, and 0.00004288/ps3 to 0.0005927/ps3, respectively. The all-optical ODE solvers with the DSMZI can be easily integrated with other optical components based on silicon on insulator, which can provide a path for future artificial intelligence or big data processing systems in optical computing on silicon waveguide chips.

针对常见人脸活体检测算法从单一RGB模态图像中提取的特征较为单一等问题,提出一种结合注意力机制的多模态双流活体检测算法SE-FeatherNet.首先,基于改进的FeatherNet网络,分别从Intel RealSense300 相机拍摄的Depth、近红外(Infrared Radiation,IR)图像中提取特征,然后将获取的特征图叠加在一起进行特征层融合,最后从融合的特征图中继续提取特征,并加入自注意力机制.针对单一评价标准存在偶然性的问题,使用多指标评价标准来验证模型的准确性.仿真结果表明,所提算法在CASIA-SURF数据集中的等错误率(equal error rate,EER)为 1.341%,半错误率(half total error rate,HTER)为1.537%,真正类率TPR@FRR=10e-2 为96.99%,并且参数量仅仅为0.58M.融合多模态的特征信息可以获取更低的错误率,改进的网络保证了算法的高效性和时效性以满足边缘设备算力有限的需求.

This paper proposes a multi-strategy Marine Predator Algorithm (MSMPA) based on the shortcomings of the Marine Predator Algorithm (MPA), such as slow convergence speed, low solving quality, and an easy fall into local optima. Firstly, during the initialization of the algorithm, Latin hypercube sampling is used to update the initialization position of the population. This method can make the position distribution of the entire population more uniform and increase the diversity of the population. Secondly, since the three optimization stages of the MPA algorithm are determined by the number of iterations, resulting in a slow convergence speed, using the elite grouping method to execute the three optimization stages of the MPA algorithm can effectively accelerate the convergence speed. Finally, in view of the fact that the algorithm is easy to fall into local optimum and low solution accuracy, the survival crisis awareness mechanism of the predator is proposed, and the algorithm modeling is conducted to change the search mode of the MPA algorithm and improve the overall quality of the algorithm. In this paper, the simulation results of 10 benchmark test functions and five other meta-heuristic algorithms are compared. Experimental results show that the above three improvements contribute to improving the performance of the algorithm.

For the original dingo optimization algorithm (DOA), such as slow convergence rate, easy to fall into local optimal solution, and reduction of population diversity, a dingo algorithm based on Tent chaotic mapping is proposed. Firstly, the uniformity and ergodicity of the Tent chaos mapping are used to ensure the difference of the population average and to improve the convergence rate of the algorithm; secondly, by introducing the scavenger strategy of the group average update position, the population update is not worse than the population average, so as to improve the convergence speed of the algorithm; finally, the weight factor is introduced into the population update strategy to improve the convergence speed and solution accuracy of the algorithm. It can be seen from the eight standard test functions that the multi-strategy improved dingo algorithm has faster optimization speed and stronger optimization ability.

针对传统哈里斯鹰优化算法(Harris Hawks Optimization,HHO)在处理庞杂问题易出现局部最优、收敛速度慢、寻优精度低的缺点,提出一种ELSHHO算法来对其进行改进.首先引入精英反向学习策略来对种群进行初始化,可以有效增强初始种群的多样性;其次在种群位置更新时加入精英反向学习策略可以提高算法探索解空间的能力和解的质量从而降低寻优难度加快收敛速度;最后,通过引入对数螺旋因子来增强算法的局部搜索性能,提高寻优精度.使用具有单峰和多峰特征的10 个测试函数来对改进的算法进行验证,通过实验得出,ELSHHO算法可以有效提高收敛速度和寻优精度.

To solve the problems of the AOA algorithm, such as the weak ability of global exploration and local development, low accuracy, and easily falling into local optimum, a multi- strategy fusion improved Archimedes optimization algorithm (MAOA) is proposed. First, two kinds of chaotic mapping and reverse learning are used to initialize the population, so that the population has a good initial solution. Secondly, the transfer operator is improved to make the conversion between global exploration and local development not monotonous and enhance the convergence ability. Finally, the adaptive Gaussian mutation strategy is introduced in the local development stage to jump out of the local optimum as much as possible and improve the convergence accuracy. The simulation experiment compares the MAOA algorithm with the standard AOA algorithm and other meta-heuristic algorithms under eight benchmark test functions. The experimental results show that the MAOA algorithm has good comprehensive performance in terms of solution accuracy and convergence speed.

In order to overcome the shortcomings of the sparrow optimization algorithm, which include limited population diversity, low high-latitude solution accuracy, and early convergence, this paper proposes to improve the sparrow search algorithm (FASSA) by using multiple strategies. Firstly, the improved sine chaotic map is used to initialize the population position, which results in a more uniform distribution of the population's initial solution position. Secondly, the position update of the discoverer is improved by introducing the golden sine algorithm with a curve adaptation, which effectively accelerates the speed of convergence and coordinates the capabilities of global and local search. Finally, the firefly algorithm is introduced to combine all sparrows with the optimal sparrows using the firefly perturbation method, so that the quality of the individuals after each iteration is improved. This paper selected eight benchmark functions for testing, and the simulation results of FASSA and the other four meta-heuristic algorithms are compared. According to the experimental findings, the improved algorithm has a great improvement in global search, overcoming local optimization, convergence speed, and convergence accuracy.

The TLTHHO algorithm is proposed to address the shortcomings of Harris Hawks optimization in dealing with certain complex problems that are prone to local optimality, slow convergence, and inability to find the global optimal solution. Firstly, a teaching-learning-based optimization algorithm is used to improve the population's low-quality solutions, enhance communication between populations, and improve solution quality. Secondly, to perturb the global optimal solution and improve the algorithm's ability to jump out of the local optimal solution, a dynamic adaptive t-distribution variation strategy is introduced. Finally, a greedy strategy is used to select between the perturbed individuals and the hawk individuals, thus improving the optimization ability and enhancing the optimization accuracy of the hawk population. The performance of TLTHHO is validated by 10 test functions with unimodal and multimodal characteristics, and the simulation experiments show that the proposed algorithm's performance is effectively improved, and the convergence speed and accuracy are faster and more accurate than the comparison algorithms.

传统人脸检测算法在复杂环境背景下一直存在着检测准确率及效率低等问题.近年来,得益于人脸数据集的增长以及计算机硬件的极速发展,使用深度神经网络的人脸检测算法在准确度方面已有很大提升,但使用的模型结构越来越复杂,检测速度也相对变慢.本文提出一种改进的多任务卷积神经网络(Multi-task convolutional neural networks,MTCNN)算法.在制造数据集时更改IOU阈值参数,来获取更多、更精确的人脸样本;对与置信度损失有关的交叉熵损失函数和与偏移量损失有关的均方差函数求均值,使得整个网络收敛得更加平稳.经在AFW、PASCAL以及FDDB数据集上实验,与传统算法相比,该算法在保证实时性的同时提升了检测准确率,可应用于追求更高准确率的人脸检测系统.

将物理层网络编码引入光网络中,研究了在3Gb/s的传输速率下基于差分相移键控(DPSK)调制的物理层网络安全编码的光网络.将两路同频信号通过光纤传输到中继,在中继处混合在一起后同时再通过光纤传给两个用户,在接收端将混合信号通过解码出来得到两个用户信号比特的异或,最后通过与其中一个用户的信息进行异或就可以得到另一个用户的信息.由于两路信号为同频信号,混合后必会发生干涉,利用DPSK将两路信号占空比分别调制为33％与67％,解决了信号混合后的干涉问题.由于在中继处两个信号混合后,发出的是两个用户的混合信号,具有一次一密效果,在利用物理层网络编码提高网络的吞吐量的同时,也增加了从中继到用户的链路的安全性.

基于虚拟仿真实验教学中心的建设,提出虚实结合的计算机实验教学方法和手段,介绍相应的实验技术、项目以及方案,给出实验考评方法,最后说明虚实结合的实验教学效果与意义。

To address overfitting of recognition results in complex Convolutional Neural Network(CNN) on small and medium face databases,this paper proposes a face recognition algorithm based on improved CNN and ensemble learning.Combining the characteristics of planar networks and residual networks,the improved CNN replaces the fully connected layer with the average pooling layer to make the network structure simple and highly portable.Based on this improved CNN,the voting-based ensemble learning strategy is used to implement convex combination for results of all individual learners and obtain the final result,so more accurate face recognition could be realized.Experimental results show that the recognition accuracy of the proposed algorithm reaches 98.89%,99.67% and 100% respectively on Color FERET,AR and ORL face databases with a high convergence speed.

虽然人脸识别技术已经取得了很大的成就,但是如何提高不同姿态、光照、表情识别的准确性也面临着很大的挑战.为了进一步提高精度,神经网络模型被设计得越来越复杂,这也直接导致在反向传播时会出现梯度消失等现象.为了缓解这些问题,本文提出了一种复杂条件下参数较少的双通道卷积残差网络模型.该模型由多个输入通道组成,共同学习输入图像的不同特征.将原始图像作为第一通道的输入,然后利用Sobel算子提取原始图像的一阶导数特征,并将其作为第二通道的输入,来自两个通道的人脸特征信息经过融合后送入到一个残差模块,经过平均池层,最后用于识别.该网络模型结构简单、参数少、速度快、准确性高.模型在FERET、AR和FEI数据集上进行了训练和测试.实验结果表明,本文的方法优于当前的一些先进方法.

本文目的在于解决任意不同新的四维超混沌系统的同步问题.本文优先设计两个不同结构的新的四维超混沌系统,进行同步分析.再添加外界干扰,研究新型四维超混沌系统的混沌同步.采用自适应滑模同步进行控制,利用simlink设计模块进行仿真.结果表明,超混沌系统能够达到有效同步.

为了解决当前图像轮廓识别算法中由于区域标记和轮廓标记性质不同,导致难以将多标记融合识别技术应用于图像轮廓识别中的问题,本文提出了一种基于环绕数约束的能量最小化模型,用以精确识别目标轮廓.在这种模型中,区域标记(如颜色和纹理均匀性)和轮廓标记(如局部对比度和连续性)通过一个目标函数进行描述,实现多标记融合识别.首先,将环绕数作为约束,将其引入到能量最小化模型中,得到区域标记与轮廓标记的线性约束;然后,对区域标记、轮廓标记以及曲率标记进行融合,实现对图像中目标轮廓的识别;最后,将能量最小化模型与标记相结合,通过比率能量函数对算法进行实例应用分析,验证算法的有效性.实验结果表明:与传统轮廓识别算法相比,所提算法具有更高的轮廓识别精度.

The physical layer network coding (PNC) technology is introduced into the optical network in this article. The proposed scheme employs all-optical XOR logic gates that rely on cross-phase modulation-based saganac interferometer fiber optic rings to perform encoding and decoding. The encoding and decoding operations are respectively applied to the coupler and the server of the data center passive optical interconnect structure, which increases the throughput of information exchange. Complex signal processing operations are replaced by optical XOR, which effectively reduces the delay. The scheme has been tested for error-free simulation of synchronization and asynchronous in transmission rate of 20-Gb/s.