The current research on simultaneous localization and mapping (SLAM) in academia mostly assumes static scenes, but dynamic objects are inevitable in real-life scenarios. Integrating deep learning into visual SLAM systems can collaboratively eliminate dynamic objects from the scene, effectively enhancing the robustness of visual SLAM in dynamic environments. This paper first introduces a classification of deep learning-based visual SLAM in dynamic environments and then provides a detailed overview of visual SLAM systems based on object detection, semantic segmentation, and instance segmentation. A comparative analysis of these approaches is also presented. Finally, considering the recent trends in the development of visual SLAM, the paper analyzes the main challenges of deep learning-based visual SLAM in dynamic environments and summarizes potential future directions.

As a graphene-like structural material, hexagonal boron nitride (h-BN) possesses many superior properties. Owing to their enhanced qualities such as simple device structure, low power consumption, and good scalability potential, h-BN memristors are receiving increasing attention and are currently considered to hold great application prospects in the fields of computing and storage, artificial neural networks, and neuromorphic computing. In this paper, a classification of memristors is introduced, the resistive switching mechanism of the h-BN memristor is discussed, and the research status of the h-BN memristor is reviewed. Finally, the current challenges of the h-BN memristor are pointed out, and the paper concludes with an outlook of the future direction of development.

In this paper, a terahertz beam splitter based on an open ring structure is proposed. Its structural unit is a typical metal-dielectric-metal structure. The top metal pattern has the form of an open ring, and the bottom layer is a continuous metal plate. An 8×8 phase gradient metasurface is formed by changing the opening size and radius of the opening ring. When the terahertz wave is vertically incident along the -z and +z axes, the electromagnetic wave will be divided into four beams that are symmetrical along the x and y axes but have different energy distributions. The two incident modes can obtain two different beam splits. At 0.7THz, when the terahertz wave is vertically incident along the -z and +z axis directions, the beam splitting ratios are 0.8∶1 and 1.9∶1, respectively, realizing the tuning of the beam splitting ratio. This beam splitter has the advantages of small size and low cost, and it can be applied in the fields of terahertz communication, terahertz imaging, and terahertz stealth.

GaAs/AlGaAs surface-illuminated uni-traveling carrier (UTC) photodiodes (PDs) are critical devices for short-range optical links. However, a conflict exists between bandwidth and responsiveness. We report a GaAs/AlGaAs UTC-PD enhanced by a distributed Bragg reflector (DBR) that consists of 20 cycles of high/low Al component AlxGa1-xAs and has a reflectance ＞0.9 in the wavelength range of 830~870nm. The thickness of the GaAs absorption layer is reduced to 1040nm, which compromises the light absorption of the PD and transit time of photogenerated electrons. The UTC-PD device is fabricated with a double-mesas structure, polymer planarization, and a coplanar waveguide electrode. The device has a -3dB bandwidth of 19.26GHz, a responsivity of 0.4926A/W at a wavelength of 850nm, and a bias of -2V.

Due to size reasons, current common flow rate sensors can only be inserted into pipes with larger diameters to measure liquid flow rate. To be placed into a micro pipeline, a cylindrical thermal-resistance micro-electromechanical system (MEMS) flow sensor with temperature compensation was designed and fabricated. It is composed of a heating resistor and a temperature-measuring resistor to measure the flow rate through the working principle of heat loss. The experimental results show that the temperature coefficient of resistance (TCR) is 0.2236%/℃ and the sensitivity was 4.25mV/(cm·s-1). The designed circuits demonstrate an excellent effect of temperature compensation for accurate measuring results. The proposed flow sensor can measure the flow rate in a pipe with an inner diameter of 2mm, and is expected to be applied to the industrial, biomedical and other fields.

An adaptive CMOS detector that automatically adjusts gain based on light intensity was designed. The detector can automatically adjust the gain based on the intensity of light during integration, enabling adaptive sensitivity and dynamic range readouts in both low-light and high-light conditions. In this study, a new comparator circuit is introduced alongside the traditional CTIA circuit to control the size of the CTIA integration capacitor. By comparing the output voltage from short exposure with the reference threshold, the results are used to fine-tune the long exposure integration gain for each pixel. The design and simulation of a 128×1 linear array CMOS detector were conducted using 0.5μm 5V-CMOS technology. The simulation results demonstrate that the CMOS detector can adaptively adjust four different integration gains within a range of five orders of light intensity, covering photocurrents from 2pA to 100nA while maintaining excellent signal readout performance.

Single-photon avalanche diodes (SPADs) have found widespread application as efficient photon detection devices in quantum communication and 3D imaging technologies. In this study, a simulation program with an integrated circuit emphasis (SPICE) model was established for SPAD within the Cadence environment. Employing the Verilog-A language, a combination of two exponential functions was utilized to continuously describe the dynamic variation in the equivalent resistance of the SPAD in the Geiger mode region. These two exponential functions encompass the equivalent resistance characteristics of the high- and low-resistance regions, addressing the issue of non-convergence in the segmented resistance models during simulation. The SPICE model simulates the dynamic behavior of the SPAD device in the "photon reception-avalanche pulse generation-quenching-reset" operational process, as well as the static current-voltage (I-V) characteristics of SPAD from forward bias to secondary breakdown. The effectiveness and stability of this model were verified through simulations involving four quenching circuits.

The use of devices that emulate biological synapses has gained recognition as a highly suitable alternative to the Von Neumann architecture. Artificial retinal devices provide advantageous contributions to the modeling of machine vision and image recognition. The goal of this research is to create artificial synaptic devices that are light-activated. This is achieved by using poly (vinylidene fluoride-trifluoroethylene) (P(VDF-TrFE)) as the ferroelectric layer and copper phthalocyanine(CuPc) as the semiconductor film. The researchers performed experiments aimed at replicating synapses, resulting in the generation of excitatory postsynaptic currents (EPSC), which allowed them to observe the synaptic plasticity characteristics and simulate memory responses to optical inputs. The experimental findings demonstrate that the polarization state of a ferroelectric material is influenced by the gate voltage, leading to a modulation of the decay rate of the photoresponsive current. The implementation of a high-pass filtering function was successfully accomplished. This device also enabled the realization of ‘AND’ and ‘OR’ Boolean logic functions by using gate voltage and light illumination as dual input logic signals.

The asymmetry and narrow linewidth properties of the Fano line shape are conducive to the realization of refractive-index sensing with high sensitivity and low detection limit. In this study, the coupled structure of the microring resonator and the Fabry-Perot cavity was designed based on the slot waveguide, using the three-dimensional time-domain finite-difference method. The Fano line shape resonance was realized and utilized to improve the performance of refractive index sensors. In contrast to reported complex structures such as multiple micro-ring cascades, a sensitivity of 500nm/RIU and detection limit of 4.00×10-5RIU were achieved under the wavelength tracking scheme based on the Fano lineshape resonance of the single-slot waveguide microring resonator. The sensitivity was 7.24×104dB/RIU for a detection limit of 5.52×10-6RIU under the intensity tracking scheme.

A polarization-selective dual-band tunable terahertz metamaterial absorber is proposed and investigated. The absorber is composed of square split graphene rings, a SiO2 dielectric layer, and a Au reflection layer. The finite-difference time-domain simulation results show that the absorber can achieve dual-band high-efficiency absorption under different polarization light incidence. When the incident light is x-polarized, the absorption rates at 7.86 and 12.63THz are 97.9% and 91.2%, respectively; when the incident light is y-polarized, the absorption rates at 6.30 and 10.52THz are 94.1% and 93.2%, respectively. By changing the Fermi energy of graphene, the wavelengths of the dual-band absorption peaks of the two polarizations can be tuned. In addition, the effects of the thickness of the dielectric layer and the physical parameters of the graphene split ring on the resonant absorption peaks are investigated. Because dual-band high absorption can be generated in both polarization states, this absorber has important potential applications in terahertz polarization imaging, terahertz sensing, selective spectral detection, and polarization multiplexing.

The miniaturized fiber Bragg grating (FBG) demodulation system is a hot research topic in the field of fiber optics sensing. The development of photonic integration technology has given rise to a series of compact FBG demodulation devices. Arrayed waveguide gratings (AWGs) serve as the core components in photonic integrated demodulation systems. Their spectral characteristics have a significant impact on the demodulation performance of photonic integrated FBG demodulation systems. In this study, we designed, simulated, and fabricated an arrayed waveguide grating (AWG) based on a silicon on insulator (SOI) substrate. Test results indicate that this AWG has an on-chip insertion loss of 3dB, crosstalk of less than -20dB, and a 3dB bandwidth of 2.3nm. We constructed a demodulation system based on this AWG, and experimental results show that the system achieves a demodulation accuracy of 34.2pm within a 0.8nm range, with a wavelength resolution of 6pm.

The radiation-mode light radiated from an M-Z optical waveguide may couple with the fiber, and this is the main factor leading to failure of the high-precision coupling between a chip and the fiber array. This study used fiber to scan the waveguide and radiation-mode light with a high-precision coupling method and studied the center position of the radiation-mode light regions, the characteristics of the radiation-mode light power that changed with the bias voltage, and the power distribution characteristics of the waveguide and radiation-mode light scanned by the fiber. The effective coupling ranges of the waveguide in the X and Y directions were 14~15.5 and 14~16μm, respectively, and the effective coupling ranges of the radiation-mode light in the X and Y directions were 89~92 and 92~96μm, respectively. The study investigated the characteristics of the radiation-mode light to formulate the correct solution and solve the coupling failure caused by radiation-mode light.

In this study, an intelligent high-speed LED driver circuit is designed using bipolar technology, which can be used as the input driver of an optical isolated IGBT gate driver chip. This design uses a new logic gate structure to realize the functions of signal transmission, fault feedback, and external setting and to reduce the delay time of signal transmission. The test results show that under normal temperature conditions, the input signal transmission delay time of the LED driver circuit is 69.99ns, and the delay time from the effective reset signal to fault elimination is 4.59μs. The designed driver circuit can meet the requirements of the optocoupler isolated IGBT grid driver chip.

Two-dimensional heat-dissipation models of common devices were established using the finite element method to improve the heat-dissipation capacity of quantum-cascade lasers (QCLs). By setting the heatsink temperature to 293K, wavelength to 8.3μm, waveguide width to 8μm, and thermal power to 12.4W, the temperature, heat flux distribution, and heat dissipation capabilities of QCLs with different device structures were studied. The results show that the highest temperatures of the epilayer-up-bonded double-channel ridge device without and with electroplated gold were 546K and 409K, respectively, while that of the epilayer-down-bonded device was 362K. For buried heterostructure (BH) devices, the highest temperatures of the epilayer-up-bonded device without and with electroplated gold were 404K and 401K, respectively, while that of the epilayer-down-bonded device was 361K. Compared with the copper submount, the highest temperature of the buried heterostructure epilayer-down-bonded to a diamond submount device was 355K. Analysis of the heat flux distribution of the models shows that the heat flux of the BH devices is more uniform, and the temperature of the core area is lower, indicating that BH structures are more suitable for high-power devices.

Simulating the neural synapses of the brain is a key step to realize the next generation computer, that is, the brain-like neuromorphic computing system. To mimic the plasticity of neural synapses using photons and develop all-optical artificial synaptic devices, we carried out an experimental study on the chalcogenide amorphous semiconductor artificial synaptic device based on controllable light-induced inhibition. The control of the material chemical composition and pumping optical power on the artificial neural synapse was studied, and the plasticity of the artificial neural synapse was described. The results show that As-S planar waveguides with different impurities have different light-induced inhibition processes, and different pumping optical powers correspond to different suppression depths. Based on these characteristics, the artificial neural synapse exhibits short-term plasticity (STP), long-term plasticity (LTP), and paired-pulse depression (PPD), demonstrating that it has good plasticity.

To reduce the bias drift of fiber optic gyroscopes due to the temperature effect, a temperature compensation model of the fiber optic gyroscope was established using an ensemble learning algorithm based on a least squares polynomial model and back propagation (BP) neural network model optimized by a genetic algorithm (GA-BP). A temperature compensation experiment of the fiber optic gyroscope was conducted after online compensation. Experimental results show that the model reduces the bias drift of the fiber optic gyroscope by more than 85% in an environment with a temperature change of -40~+60℃, and the average bias output of the compensated starting section is closer to the zero position.

Indium phosphide (InP) polycrystals were synthesized using the horizontal gradient-freeze method. The influence of different temperature gradients on the ratio of polycrystals was analyzed, The results show that the crystals are indium-rich with a ratio of less than 97% when the temperature gradient is lower than 4℃/cm, and the crystals are stoichiometric with a ratio of more than 99% when the temperature gradient is above 5℃/cm. The impurities and electrical properties of the polycrystalline samples were analyzed using glow discharge mass spectrometry (GDMS) and Hall tests. The purity of stoichiometric InP polycrystals was greater than 99.99999%; the carrier concentration was less than 8×1015cm-3; the mobility was greater than 3900cm2·V-1·s-1. The impurities in the polycrystals primarily included Si, S, Fe, Cu, Zn, and As. The sources of the impurities and their effects on the properties of the materials were analyzed.

Cadmium zinc telluride (CdZnTe) is the most critical room-temperature semiconductor nuclear radiation detector material. A portion of Te was substituted by Se in the CdZnTe lattice to obtain cadmium zinc telluride selenide (CdZnTeSe); this led to an increased composition of the ionic bonds in the lattice, which improved the hardness of the crystal and reduced the concentration of Cd vacancies and Te inclusion defects, thereby enhancing the quality of the material. In this study, to obtain CdZnTeSe crystals suitable for the manufacture of nuclear radiation detectors, the growth of CdZnTeSe crystals under Te-rich conditions was investigated using the vertical Bridgman method, and Cd0.9Zn0.1Te0.97Se0.03 single-crystal ingots with a diameter of 21mm and a length of more than 70mm were successfully prepared. The Cd0.9Zn0.1Te0.97Se0.03 crystals thus obtained had an X-ray diffraction wobble half-peak width of 0.104° in the (110) plane, and the size of the Te intercalation phase was ＜5μm, indicating that the crystals had good crystallinity. The energy bandgap and infrared (IR) transmittance of the Cd0.9Zn0.1Te0.97Se0.03 ingot tails were lower than those of the head and middle of the ingot. This can be attributed to the decrease in crystallinity in the subsequent growth stages of the crystals due to the insufficient release of latent heat during the growth of Cd0.9Zn0.1Te0.97Se0.03 using the vertical Bridgman method. The electrical performance indices of the head and middle of the CdZnTeSe ingots satisfied the requirements for the preparation of room-temperature radiation detectors.

To develop a low-cost and high-sensitivity device for detection of aflatoxin B1 (AFB1), an optical fiber immunosensor based on the functionalization of polyaniline for the detection of AFB1 was developed. A sensitized optical fiber sensor for the detection of AFB1 was designed by combining the excellent optical properties of polyaniline (PAni) with the soft and bending-resistant properties of the plastic optical fiber (POF). First, PAni was used to carboxylate the POF surface in the sensing region, and second, the sensing region was cross-linked with glutaraldehyde to immobilize the AFB1 antibody molecule, which could then be obtained by the sensitized optical fiber sensor for detecting AFB1. The sensor can detect the concentration of AFB1 through the change in the light intensity signal of the optical fiber output caused by the immune response. The sensitizing effect of the polyaniline coating on the sensor was experimentally investigated, and the specificity, immunity to interference, and detection limit of the sensor were tested. The results showed that there was a linear relationship between the output signal of the sensor and the AFB1 concentration in the range of 0.01~10μg/L, and the detection limit was 0.53μg/L. The recoveries of the spiked standard ranged from 95.97% to 113.13%, and the specificity and immunity of the sensor were good, thus meeting the needs of a refined detection of AFB1.

Gold-wire bonding is widely used in the packaging of infrared detectors. In this study, a 25μm gold wire was selected to determine the best process parameters for bonding with respect to the tensile value of bonding strength based on the orthogonal experimental method. The combination of the ultrasonic process parameters of pressure, power, and time, with contact force was optimized, to improve electrical connection performance and connection strength of the bonded leads, thereby enhancing the signal transmission quality of the chip system. The combination of the wire bonding process parameters proposed in this study is applicable to the bonding of infrared detection chips.

This study proposes a road crack image preprocessing method based on the fusion of a line segment detector (LSD) and fast line detector (FLD) to address the issue of linear interference, such as road signs and road edges, which affect road crack recognition. First, based on the LSD and FLD algorithms, line detection is performed on the crack image to obtain the line segment coordinate values of linear interferences. Second, by reconnecting the broken lines based on the line segment coordinate values returned by the line detection algorithm, the problem of extracting discontinuous line segments is solved using the line detection algorithm. Finally, the fast marching method (FMM) inpainting algorithm is used to eliminate linear interferences using the mask image of the linear interferences and the original image of the crack image obtained after reconnecting the line segments. Based on extensive experimental analysis, this method can effectively eliminate linear interference in crack images, improving the accuracy of crack detection by 7.1%.

The error in placement angle of the fiber Bragg grating (FBG) in fiber optic shape sensing can affect the accuracy of curvature calculation, thereby increasing the error of shape inversion. To achieve self-correction of the angle of the sensing rubber rod, a self-correction shape sensing structure was designed. Nine FBGs were deployed on the sensing cross-section of the sensing rubber rod, with equal angle intervals of 120° as the reference positions, and strains were extracted at ±10° positions, thereby completing the functional mapping of the FBG response and angle deviation. A self-correction algorithm based on angle deviation is proposed, which optimizes the threshold of sensing parameters α and k through the fitness function to achieve self-correction of any angle deviation. A simulation analysis of the response relationship under different α and k conditions revealed that α had good linear variation characteristics, and k only fluctuated with the main sensitive FBG. In the single section experiment, the average response of the nine FBGs was between \[-1.012με/N,0.987με/N\] after loading 0~100N stress changes. The positive or negative response characterized the bending direction, and the responses of adjacent FBGs had good linearity. In the composite cross-section experiment, the three-dimensional structure of the sensing rubber rod was reconstructed based on the inversion results, and the coordinates and stress values of each point were output.

A sub-pixel edge detection method based on a tanh function model is proposed to address the problem of the difficult balance among real-time performance, accuracy, and robustness of traditional sub-pixel edge detection algorithms. First, an edge function model is established, and the approximate value and error of the constant term of the function equation corresponding to each pixel edge are calculated. Combined with the least-squares method, the overall functional model error is then minimized to obtain more accurate edge parameters. Experimental results based on both standard and actual images show that the proposed algorithm has advantages such as fast speed, strong robustness, and high detection accuracy and can meet the requirements of real-time detection using better evaluation indicators than other traditional algorithms.

To achieve real-time monitoring and evaluation of the structural health status of solid rocket motors (SRMs), a high-strength femtosecond grating sensing network was designed and developed. The sensors were coated and packaged, and SRM interface debonding simulation experiments were conducted. A simulation demonstration system for SRM strain, temperature, and load was designed. A digital twin model of the engine based on the data collected by the optical fiber sensor was established, and an accurate mapping between the real SRM and the digital twin was formed. Finally, the optical fiber sensor array was implanted in the SRM, and water pressure monitoring experiments, high-and low-temperature storage experiments, long-term monitoring experiments, and ignition tests were conducted on the SRM. Non-destructive testing conducted verification experiments on the monitoring results. The research results indicate that the sensor network can accurately measure the strain state of the SRM, providing effective data support for its structural health management.

This study proposes a spatio-temporal dynamic multiscale convolutional neural network (DMS-CNN) classification model based on a dimensional attention mechanism to improve classification accuracy and applicability to practical scenarios, in order to address the problems of non-stationarity, time-varying complexity, and low classification accuracy of electroencephalogram (EEG) signals, as well as the shortcomings of traditional machine learning methods in extracting complex features. First, the data are bandpass-filtered to eliminate artifacts and pre-processed using downsampling and channel selection. The processed data are then input into the constructed spatiotemporal convolution model, to further enhance the feature extraction capability of the network and multidimensional attention mechanisms of timing. This is followed by the incorporation of channel and regularization technology. To address the problem of insufficient data, a frequency-band exchange method is used to enhance the data, thereby improving the generalization performance of the model. Average classification accuracies of 90.97% and 90.21% were obtained for the HGD and self-collected laboratory datasets, respectively. Compared with other algorithms, the classification accuracy of this method was significantly improved.

In order to reduce the influence of noise in the fiber optic perimeter security signal on the classification results and improve the accuracy and operating efficiency of signal classification, in this paper a classification method combine Correlation Variational Mode Decomposition (CVMD), Dung Beetle Optimizer (DBO), and Support Vector Machine (SVM) was proposed. CVMD was used to remove the noise component in the original signal. The energy, energy entropy and kurtosis of the denoised signal were extracted as feature vectors. The DBO algorithm was adopted to optimize the SVM to obtain the best penalty factor and kernel function parameters. The DBO-SVM classification model was constructed. A perimeter security system based on phase-sensitive optical time-domain reflectometry (Φ-OTDR) technology was built to collect four types of signals: climbing, knocking, stepping and non-intrusion. The experimental results show that the classification accuracy of CVMD-DBO-SVM is higher than that of CVMD-PSO-SVM and CVMD-GA-SVM, reaching 98.75%. At the same time, the running time is shorter and the overall performance is the best.