Nickel-based alloys have many advantages, such as high strength and corrosion resistance, and are widely used in aerospace, energy power and other fields. Laser cladding utilizes a high-energy laser beam to rapidly melt the cladding material on the surface of the base material, forming a good metallurgical bond. Laser cladding of nickel-based alloys has the advantages of high production efficiency and low cost, and has a broad application prospect. In this paper, the current research status is sorted out from the types of nickel-based alloys for laser cladding, the research on laser cladding process parameters and the performance of cladding layer, the current research results are summarized, and the application of nickel-based alloys laser cladding in reactors is prospected, with a view to providing references to the in-depth research and practical application of nickel-based alloys laser cladding.
Fruit consumption has many benefits for human health, but in the course growth, transportation and storage, fruit may contain harmful substances, these substances will pose a potential threat to human health. Therefore, it is necessary to detect harmful substances in fruits quickly and accurately. Surface-enhanced Raman scattering (SERS) technique, as a sensitive and non-destructive detection method, has been widely studied in the detection of fruit contaminants in recent years. SERS technology can be used to detect pesticide residues, heavy metals, mold and additives in fruits. This paper introduced the principle of Raman scattering and SERS technology, as well as the application and research progress of SERS technology in the detection of common harmful objects in fruits, summarized the advantages and development trend of SERS technology in the detection of fruit safety, and discussed its limitations and challenges. It provides a new thought for the future study of SERS technology in the field of fruit harmful substances detection. Therefore, it has a strong guiding significance and practical value
Magnesium alloy has good biocompatibility, biodegradability, density and elastic modulus similar to human bone, so it has broad application prospects in the biomedical field. Additive manufacturing technology provides an effective way to produce magnesium alloy porous structure implants with complex shapes. However, the characteristics of magnesium alloy, such as low melting point and easy gasification, bring great difficulties to additive manufacturing. The forming quality of three - period minimal curved surface structure (TPMS) of WE43 magnesium alloy made by Laser powder bed fusion (LPBF) additive was studied from two aspects of design method and process. The influences of different TPMS structure and LPBF process parameters on porosity and forming accuracy were analyzed. The compressive mechanical properties of TPMS structure were further tested. The in vitro degradation performance of TPMS porous structure was studied. The relationship between TPMS structure type, forming quality and compressive mechanical properties was obtained, and the effect of TPMS structure type on the degradation rate in vitro was analyzed. The results show that the TPMS structure has better forming quality and compressive mechanical properties than the sheet TPMS structure with the same design porosity, and the sheet TPMS structure has stronger ability to withstand large deformation than the network TPMS structure. The degradation rate of TPMS structures increases first and then decreased. The structural integrity of porous structures of TPMS was lost at 12 h, and the degradation rate is positively correlated with the actual porosity.
The severe safety hazards in buildings pose significant threats to personal and property safety. To effectively protect personal and property safety, the article proposes an optical fiber FP cavity displacement sensor with a dual-ring concave mirror structure for safety inspection. First, hydrofluoric acid is used for chemical etching to obtain a groove structure. Then, photoresist is used to create waveguides at the core position of the fiber. The fabricated fiber structure is then coated with a silver film using magnetron sputtering. Finally, the waveguide structure is washed away to obtain a double-ring concave mirror structure. Two fibers with the same structure form a Fabry-Perot cavity structure. The research results indicate that the dual-ring concave mirror structure can effectively enhance the sensitivity and linearity of the sensor. The designed sensor achieves a sensitivity of 105.535/m within a range of 0-30 micrometers, with a linearity exceeding 0.999.
A high-wave semi-transparent optical interference pigment with a main wavelength of 665nm and a color effect of purplish red to yellowish green was prepared by using 5 layers of full dielectric film system with alternating stacking of high and low refractive index films, titanium dioxide for high refractive index layer and silicon dioxide for low refractive index layer. The transmission spectrum and reflection spectrum of the sample in the three states of film, film powder and pigment were detected by the UV-3600i Plus UV-visible near-infrared spectrophotometer of Shimadzu. The practical application effect of the ink made by mixing the pigment sheet with light curing ink was tested by scraping sample. The results show that the prepared pigment can clearly display the text or pattern on the bottom plate under the premise of retaining bright colors and obvious anti-counterfeiting effect with the corner. The coating of the pigment can provide anti-counterfeiting effect with the corner and improve the anti-counterfeiting ability of the logo under the premise of retaining the anti-counterfeiting design effect of the logo.
An optical millimeter-wave system based on optical heterodyne technology is proposed. The system realizes the parallel transmission of multi-frequency millimeter wave signal. In this scheme, CW Laser Array is used as the system light source. CW produces multiple emission laser frequencies and one local oscillator laser frequency (LO). Single-mode fiber (SMF) transmits multiple optical frequencies from a central station (CS) to a base station (BS). The photodetector (PD) beats to obtain millimeter-wave signals in multiple frequency bands. In this paper, simulation experiments are carried out based on four-frequency RoF channel. The minimum bit error rates of 39 GHz, 60 GHz and 80 GHz mmwave signals are 5.56×10-157, 0 and 8.12×10-28, respectively. The feasibility and good performance of the system are verified by the simulation results of spectral map, eye map and bit error rate. The influence of system phase noise and laser linewidth is discussed. The experimental results show that the system has a good effect of photogenerated millimeter wave and parallel transmission of multi-band signals. The results show that the desired baseband signal can be effectively restored under BS. This research provides support for the future development of fiber optic wireless communication.
Through the two-dimensional particle-in-cell simulations method, we investigate the generation of rays in the interaction between ultraintensity lasers and gas-solid composite targets. The results show that, the laser wakefield are derived by the laser, the electrons are trapped and accelerated to high energy. The laser reflected by the solid target and interacts with the high-energy electrons. The electrons continue to propagate forward, pass through the solid target, then interacts with another incident laser propagating to the left. Consequently, two groups of rays with different lengths, energies and delays are generated. The effects of driving laser polarization and pre-plasma density on rays are analyzed, and the physical mechanism is discussed. The results show that more photons can be generated under the higher density pre-plasma and circularly polarized laser, and the resulting rays have the advantages of small divergence angle and high flux.
Atmospheric turbulence is one of the important factors affecting the performance of atmospheric laser communication. Traditional measurement methods have the problems of limited measurement range and influence of environment. Based on the optical phase-locked loop (OPLL) technology in homodyne coherent optical communication system, the atmospheric turbulence intensity is demodulated through the additional phase fluctuation information introduced by the atmospheric turbulence between channels. It can realize the real-time measurement of the atmospheric turbulence intensity in the communication channel with high precision, wide range and high response. Firstly, a theoretical model of the measurement system is established, simulating random phase noise model of atmospheric turbulence under different intensities. Secondly, the homodyne coherent optical communication system is built through software. Finally, the turbulence random phase information is introduced into the simulation system. The simulation results verify the feasibility of the theory and shows that the measurement range of the atmospheric turbulence measurement system is 10-18 m-2/3 ~ 10-12 m-2/3, and the measurement accuracy is better than 14.8%. The results show that this method has advantages of higher measuring accuracy, wider measuring range and more flexible measurement distance than the traditional measurement methods of atmospheric turbulence, and the results are of great significance for studying the intensity of atmospheric turbulence between channels in atmospheric optical communication system.
In order to improve the accuracy of methane gas concentration detection by tunable semicondu- ctor laser absorption spectroscopy (TDLAS) technology, a noise reduction method of complementary ensemble empirical modal decomposition (CEEMD) combined with permutation entropy and S-G filtering has been proposed for the problem of noise interference in the directly absorbed signals during the detection process, and permutation entropy and S-G filtering are used to solve the problem of CEEMD decomposition with the occurrence of spurious components and noise residual noise in the CEEMD decomposition. By simulating different noise contents of methane gas absorption signals and comparing with the traditional CEEMD method and wavelet transform method, the effectiveness of the proposed method is verified, which provides the basis for the subsequent experiments. The experiments show that the proposed method can effectively reduce the noise interference in the gas absorption signal and improve the measurement accuracy of the system, compared with the comparative methods, with the highest goodness-of-fit of 0.991 1 for the Lorentz line fit of the absorption spectrum curve, and the goodness-of-fit of the absorption spectrum amplitude to the methane gas concentration of 0.998 46.
Aiming at the problems of low target detection accuracy and difficulty in detecting small targets in current UAV infrared aerial photography images, this paper proposes an improved UAV infrared aerial photography target detection method based on YOLOv8. Dual channel feature fusion structure is introduced to increase the feature fusion capability and reduce the loss of feature information. The lightweight small target detection layer is added to improve the detection ability of the model to the infrared small target. The lightweight convolutional module GSConv is used to replace the traditional convolution in the neck network C2f, reducing the size of the model and improving the detection speed of the model. Finally, the convolutional attention module is added to the SPPF module of the backbone network to further increase the model’s attention to infrared target information and improve the accuracy of model detection. The feasibility of the improved network is verified through experiments. Compared with the benchmark model YOLOv8n, the accuracy rate is increased by 4.1%, and the average accuracy of mAP50 is increased by 3.7. Compared with eight current mainstream models, the model proposed in this paper has the highest accuracy, reaching 83.3%, and the FPS reaching 153. The effectiveness of the method is proved.
In light of the intricate texture, subtle imperfections, and diverse issues present in PCB boards, a novel edge detection method tailored for the internal circuits of PCB boards is proposed. Initially, adaptive local noise reduction filtering is utilized to preprocess the image and diminish noise interference. Subsequently, edge information is amplified by computing the absolute value of the Laplace transform output. Following this, a straight line detection kernel is introduced for a secondary convolution operation, and the 'Euclidean norm' of the outcome is computed to further accentuate the edge characteristics. Ultimately, circuit edges are precisely extracted through the application of local threshold segmentation technology. Experimental results demonstrate that this method enhances anti-interference performance by 1.5 times compared to traditional edge detection algorithms, resulting in improved edge detection effects on PCB board circuits.
When the laser beam passes through optical components, diffraction and defocusing phenomena occur, which affect the optical signals of different parts to a certain extent, resulting in spatial nonlinear distortion and errors in long-distance laser ranging sensors. Therefore, a method for correcting errors in long-distance laser ranging sensors based on equidistant lines is proposed. Calculate distortion parameters using equidistant lines in a two-dimensional coordinate system to correct nonlinear distortion in long-distance laser ranging sensors. Based on the principle of laser scanning interference, the phase change is used to compensate for the phase of the ranging signal and correct measurement errors caused by target motion. Based on the linear relationship between inclination angle and error, construct a support vector machine regression model to predict and correct the inclination angle error of long-distance laser ranging sensors. The experimental results show that the PLCC and SROCC of the studied method reach 0.935 and 0.876, respectively, indicating good nonlinear distortion correction; After phase compensation correction, the spectral resolution is 12.56 mm, with a difference of only 0.06 mm; Compared to other comparison methods, the distance measurement results under multiple measurement points always approach the actual distance the closest.
Aiming at the problem of low detection accuracy in traditional semiconductor laser chip defect detection, a semiconductor laser chip defect detection and classification method based on improved Faster R-CNN is proposed. Firstly, a semiconductor laser chip defect acquisition device was built to collect the state of the chip during operation and establish a dataset; then the ResNet50 feature extraction network was optimised to reduce the residual block structure while using multiple 3*3 convolutional layers to improve its ability to detect important information; finally, the CA (Coordinate Attention) attention mechanism was introduced into different layers of the improved Res-Net50 network in different layers to adaptively learn the importance weights of each channel and further improve the feature representation ability. The experimental results show that compared with the original network, the proposed method improves the detection precision and classification accuracy, has better recall and accuracy, and is able to quickly and accurately carry out defect detection to further improve the process.
Aiming at the problem that the diameter measurement system of cylinder parts is complicated and the error is large, a diameter measurement method based on double line structured light vision system is proposed in this paper. In this method, two non-cross laser lines are irradiated on the surface of the workpiece, the curvature information of the laser fringe is effectively used to extract the subpixel center point accurately, and the normal deviation threshold is set to eliminate the abnormal point. Through biplane calibration, the equation of the vertical plane of the center axis is derived to correct the error of non-vertical irradiation of the laser line. Finally, the effective points are fitted on the vertical plane, and the average value of the multi-angle fitting data is taken as the diameter measurement value. The experimental results show that compared with the local laser scanning method, the measurement accuracy of the proposed method is improved by about 30 m, and the average measurement error of the diameter of the cylindrical workpiece within 200mm diameter is less than 17 m, which improves the detection accuracy and loosens the measurement system requirements.
The safety and stability of autonomous vehicle driving are inseparable from accurate lane recognition. However, daily driving faces challenges such as complex and changing weather and lighting conditions, blurred or blocked road markings. Research and design lane line recognition algorithms based on deep neural networks to improve the robustness of recognition technology in complex environments and the accuracy of detection results. By constructing a fully convolutional neural network model with VGG-16 as the main chain and embedding channel attention and spatial attention mechanisms, end-to-end pixel-level lane lines semantic segmentation is achieved. The new model embedding the attention module is verified on the CULane general data set. Compared with the VGG-decoding semantic segmentation method, its average pixel accuracy and Mean Intersection over Union (MIoU) increased by 2.2% and 1.3% respectively. And in the scenario where lanes do not exist, the pixel accuracy of the prediction results reaches 70%. Research on image segmentation algorithms embedding attention mechanisms provides an effective solution to the problem of lane line recognition, and strongly supports the application of lane line detection technology in driverless driving scenarios.
In response to the issue that remote sensing images have lower resolution than traditional images and are affected by complex degradation processes, traditional generative adversarial networks can generate unrealistic features, leading to problems such as artifacts and a large number of false, sharp edges. This paper proposes an edge extraction and enhancement-based remote sensing image super-resolution network called EEEGAN. The network first employs the edge extraction algorithm TEED to extract image edges. It then designs a dual attention mechanism, TAM, to capture rich spatial and channel information of the image. Additionally, a basic block RRDJB is introduced to expand the model’s processing capabilities, and a downsampling network SPD is incorporated to further reduce detail loss. Based on the RSOD dataset, different data degradation treatments were applied according to degradation models. The results show that the proposed model in this paper has improved metrics under various degradation conditions compared to current mainstream image super-resolution models. The method presented in the paper shows a 0.034 increase in SSIM and a 1.329 8 dB increase in PSNR on samples with degradation condition I compared to the real enhanced image super-resolution generative adversarial network. After reconstruction, the visual effect of edge details in the images is better. Furthermore, good generalization effects were achieved on both the DIOR and HRSC2016 datasets.
The thesis constructs a fast image fusion method based on pixel-wise voting is explored and constructed. First, the relevant visualization and quantitative analysis experiments were carried out under different step sizes of sliding window. It is confirmed from this that the fused-image performance with a 2-pixel step of sliding window is very close to that with 1-pixel step of sliding window. Therefore, the 2-pixel step of sliding-window is used in the new fast image fusion method to reduce the overall fusion computation time. Secondly, the method of reducing the source-image resolution is used to further reduce the computational complexity of the fast fusion algorithm. The experimental results show that, without reducing the fusion effect, choosing the appropriate scaling ratio can effectively increase the operation speed of the fusion algorithm. In addition, relevant experiments were also carried out on the data set with image shift caused by the jitter of the photographic equipment. The experimental results tell that the proposed fast fusion algorithm has strong robustness and can effectively fuse the original multi-focus images with slight jitter.
In this paper, a 3D high-resolution imaging method based on compressive sensing reconstruction theory is proposed for Frequency Modulated Continuous Wave (FMCW) Colocated Multiple Input Multiple Output (MIMO) radar. For the array signals obtained by Time Division Multiple access MIMO radars, the block compressed sensing (BCS) method based on antenna spatial layout was adopted. In addition, a three-dimensional Constant false alarm Rate (3D-CFAR) filter was designed to reduce the influence of clutter signals on the quality of reconstructed signals, and a correlation smoothing method was designed to eliminate the reconstruction errors between different blocks. The proposed algorithm can achieve high-resolution imaging of complex targets at a low sampling ratio and avoid the problems of algorithm complexity and hardware resource occupation in traditional compressed sensing reconstruction. Simulation and experiment prove its feasibility and practicability. Compared with other imaging methods, the high-resolution imaging effect of the proposed algorithm is comparable to that of Synthetic Aperture Radar (SAR). The simulation results show that when the compressed sampling ratio is 0.6, the signal-to-clutter ratio of the target imaging has reached the level of the original target, and the target structural similarity index has reached 80%, which realizes the ideal imaging effect. The algorithm has practicability in the engineering field and is expected to be widely used in target detection and recognition of millimeter wave radar.
To address the issues of spectral shift and spectral redundancy in cross-domain classification of hyperspectral images, this paper proposes a domain adaptation classification method for hyperspectral images based on the Transformer network. This method introduces a novel Pixel-wise Hyperspectral Long-wave Block Partitioning strategy and Neighborhood Correlation-based Central Pixel Feature Extraction strategy. It effectively extracts local-long range spectral correlation features and central pixel information from hyperspectral images. Finally, knowledge transfer is realized through a dual classifier architecture. The experimental results on the Houston and YRD datasets confirm the effectiveness of the proposed method. The introduction of this method provides a new perspective and technical path for the research of domain adaptation classification in hyperspectral imaging.
Large scale laser images refer to a large amount of laser scanning image data with rich information. The application of such images in fields such as geographic information acquisition, urban planning, and autonomous driving is becoming increasingly widespread, usually containing rich spatial information and detailed features. Therefore, there is a problem of high processing complexity. Cloud computing, as an emerging computing mode, has strong computing power and flexible resource allocation advantages. Research on cloud computing based rapid recognition methods for multi type large-scale laser images. Process laser images through nonlinear matching to determine the pose change matrix of target points in the laser image; Filter the processed laser images using machine algorithms to identify key features in different types of laser images; By using cloud computing correlation methods, a fast recognition model is constructed to achieve rapid recognition of multiple types of large-scale laser images through the objective function within the model. The experimental results show that using two sets of different types of continuous frame laser images as test samples, the studied method can achieve rapid recognition under the design scheme and has practical value.
In the process of laser 3D image restoration, the image details are not enough and the efficiency is low, so a laser 3D image restoration method based on virtual reality technology is proposed. The three-dimensional laser image is decomposed by stationary wavelet transform technology, and the low frequency and high frequency components of the original image are obtained. The laser 3D image denoising model is designed, multiple networks of the model are trained, network parameters are adjusted, and nearly clean images are output. The denoising process of the laser 3D image is realized, and the restoration and output of the laser 3D image are realized. The experimental results show that the maximum mean square error is 0.008, the minimum structural similarity is 0.920, and the peak SNR is 0.22, which is obviously better than the comparison method, and the average time is less. It shows that the method in this paper can ensure the high level of image restoration and improve the efficiency of restoration processing.
For the traditional hyperspectral image classification algorithm’s problems of insufficient utilization of feature information and inability to reduce the spatial redundancy of the feature map effectively, an improved hybrid convolution-based multiscale model, MH-CNN, is proposed, which uses a multiscale 3DCNN module for the initial extraction of spatial and spectral features of hyperspectral images, and then adopts a multiscale 2DCNN network embedded with a spatial reconstruction module to the deep spatial features of the feature map is further extracted and optimized. Finally, the fully connected layer accurately calculates the hyperspectral remote sensing images. In this paper, the experiments are carried out on three open source datasets, including Indian Pines, Pavia Centre, and Pavia University, and seven classical classification methods are selected as comparisons and the overall accuracies of this paper’s MH-CNN algorithm on the three datasets reach 97.7%, 99.2%, and 98.5%, respectively. The experimental results show that the MH-CNN algorithm makes full use of both spatial and spectral features in hyperspectral images and, at the same time, effectively reduces the spatial redundancy of the feature maps, improves the classification accuracy compared with other models, and has better comprehensive performance.
in order to achieve rapid signal transmission and improve the stability of optical communication systems, an intelligent processing system for optical communication signals based on embedded technology is designed. Build an intelligent processing system framework for optical communication signals, embed DSP processing modules into the optical communication system, use LED arrays as light sources to obtain the original optical communication signals, complete RS encoding and truncated PPM modulation in DSP, and apply them to LED to achieve the transmission of optical communication signals. After channel transmission, the beam is concentrated and transmitted through convex lenses, and the detector array receives the optical signal. The signal interpretation is completed using the Berlekamp Massey algorithm, and the original input signal is restored using a bipolar signal processing circuit composed of a high bandwidth gain product operational amplifier, a fully differential operational amplifier, and a limiting amplifier. The experimental results show that the system can ensure stable transmission of optical communication signals at a communication distance of 5 km. As the communication distance increases, the bit error rate increases slightly. Within a communication distance of 2.5 km, the bit error rate is 0, which can achieve accurate recovery of input signals. The signal-to-noise ratio at the transmitting and receiving ends is not less than 74 dB, and the signal processing performance is outstanding.
In order to obtain target information from massive data in optical communication networks, optimize network performance and service quality, a deep mining method for optical communication network data based on improved fuzzy clustering is proposed. Using probability and neighborhood classification methods to separate real-time and historical data streams, and obtain a set of real-time and effective data streams. Using point density function to improve fuzzy clustering algorithm, determining the optimal initial clustering center, and then merging clustering points through inter class distance to accelerate iteration speed. Based on the effectiveness function, the number of clustering centers is determined. Calculate the high-order density spectrum of real-time effective data streams accumulated in two-dimensional space within two discrete sampling periods, correct the update trajectory of the data stream, use differential evolution to optimize fuzzy clustering iteration, and achieve deep data mining in optical communication networks. Experimental results have shown that the improved fuzzy clustering algorithm has good data mining performance and can accurately obtain valuable target information from the network.
To accelerate the construction of the Internet of Things and improve its efficiency, a real-time perception method for the status of key nodes in the Internet of Things based on fiber optic communication is designed. Regarding the perception layer, network layer, and application layer structure of the Internet of Things, fiber optic sensors are applied in the perception layer to collect real-time status data of key nodes in the Internet of Things, achieving preliminary real-time perception of key node status; The perception layer provides this data to the network layer, where it is transmitted through a wavelength division multiplexing (WDM) optical wireless network via fiber optic communication. The perception data is securely transmitted to the IoT application layer, where it is stored and processed. Experimental verification shows that this method achieves better time efficiency and relatively small errors in perceiving the state of key nodes; It can provide high communication efficiency and throughput during the transmission process. Therefore, this method has good real-time perception ability of the state of key nodes in the Internet of Things.
The number of nodes in ultra dense visible light communication networks is huge, and the connection relationships between nodes are complex. Key communication nodes carry a large amount of data streams and communication tasks, which can provide important reference for resource allocation. Therefore, a resource allocation method for ultra dense visible light communication networks based on improved differential evolution method is proposed. Calculate the information carrying capacity of ultra dense visible light communication links based on channel gain, signal-to-noise ratio, and communication interruption probability. Determine key communication nodes based on node blocking probability. Build a resource allocation model based on link information carrying capacity and key communication nodes. Use an improved differential evolution algorithm to solve the model and achieve optimal resource allocation for ultra dense visible light communication networks. The experimental results show that the proposed method can effectively improve the data transmission rate and throughput of ultra dense visible light communication networks, and has good practical application effects.
In order to solve the contradiction between the increasing data transmission demand and limited network resources faced by the current optical transmission network, a wavelength optimization method for optical transmission network routing based on data mining is designed. Build a hierarchical graph model to map wavelength resources in physical links to connection lines in a graphical structure, and solve routing and wavelength allocation problems within the same framework. Use genetic algorithms to design network load dynamic adaptation algorithms to achieve dynamic load adaptation optimization in optical transport networks, thereby achieving joint optimization of routing and wavelength. The experimental test results show that the method maintains a low blocking index throughout the entire testing process. As the number of services increases, the blocking index rises slowly, while the resource utilization index increases rapidly in the early stages of service, gradually stabilizes, and finally reaches a resource utilization index close to 0.7.
When transmitting laser communication network services, there is usually a communication load imbalance due to multi priority services. Therefore, to address this issue, dynamic routing optimization for multi priority service load balancing in laser communication networks is proposed. This method first requires the development of a priority queuing strategy for services based on the laser communication network, and the hierarchical division of services transmitted within the network to ensure priority processing of critical tasks. Utilizing graph convolutional networks to construct a link bandwidth occupancy prediction model, the prediction results will be fused with innovative triangular modal operators to generate a comprehensive path evaluation index - path selection degree. This indicator serves as the core basis to guide communication networks in selecting the optimal path during business transmission, achieving an efficient and short path communication routing strategy, thereby optimizing network performance; Finally, after completing the routing selection in the laser communication network, the bandwidth allocation method is used to allocate network bandwidth based on business priority order, achieving network load balancing and dynamic routing optimization in the communication network. The experimental results show that using this method for dynamic routing optimization of communication network load balancing has good effects and high performance.
In order to deeply understand the performance of optical signal transmission, modulation, demodulation, and improve the transmission rate, distance, and stability of optical communication systems, a spectral feature extraction based optical signal characteristic analysis method is proposed. This method establishes an optical signal detection channel consisting of a coupler, reflector, and fiber Bragg grating sensor array. After obtaining the optical field intensity and optical communication signal during optical signal transmission through this channel, Fourier transform is performed on it to obtain the frequency spectrum of the optical signal. Then, linear frequency modulation transform and window setting are used to detect single frequency components from the frequency spectrum of the optical signal, extract the frequency spectrum characteristics of the optical signal. Based on the frequency spectrum characteristics of the optical signal, a three wave coupling equation of the optical signal is established. After solving this equation, the pump light frequency, scattered light power, and other characteristics in the optical signal characteristics are obtained. The experimental results show that this method can effectively obtain the frequency spectrum of optical signals, extract spectral features from this frequency spectrum, and analyze the scattered light intensity frequency of optical signals under different pulse widths, with strong applicability.
In view of the large quantity and variety of waste plastic recycling, it is difficult to quickly and nondestructive identification, a plastic material identification method based on near infrared spectroscopy was proposed. The near-infrared spectral data of eight kinds of plastics, namely polyethylene terephthalate (PET), polyethylene (PE), nylon (PA), polycarbonate (PC), polypropylene (PP), polystyrene (PS), acrylonitrile-butadiene-styrene (ABS) and polyformaldehyde (POM), were collected by infrared spectrometer. Savitzky-Golay convolution smoothing and standard normal variable transformation were used for spectral data preprocessing. Unsupervised learning principal component analysis and supervised learning linear discriminant analysis (LDA) were used respectively to reduce the dimension of spectral data. The dimensions of the spectral data are reduced from 334 to 10 and 7. Finally, the identification model of plastic material is established based on Mahalanobis distance discrimination. The experimental results show that the combination of S-G smoothing and SNV preprocessing effectively improves the recognition accuracy. After dimensionality reduction of the validation set of the preprocessed data, the recognition accuracies of the two dimensionality reduction methods reached 95.24% and 100% respectively. These two methods can provide reference for the identification of various waste plastic materials.
This paper proposes a high-power laser bevel cutting roughness prediction method based on particle swarm optimization radial basis neural network. The 40 kW laser bevel cutting system is used to carry out 30°V- bevel cutting test on Q235 carbon steel with 50 mm thickness; based on the orthogonal test results, the regression prediction model between the laser bevel cutting process parameters and the roughness of the bevel cut surface is established by the radial basis neural network; the particle swarm algorithm is used to achieve the optimization of the center position and width of the function of the hidden layer of the radial basis neural network, as well as the optimization of the weights between the hidden layer and the output layer, and the optimized model is used for the prediction of bevel cut surface roughness. The optimized model is used to predict the roughness of the bevel cut surface. The experimental results show that compared with the multilayer feed-forward neural network and the standard radial basis neural network model, the model is more accurate in predicting the roughness of the bevel cut, and the coefficient of determination of the prediction model is 0.957 6, the root-mean-square error is 0.032 6, and the average error of deviation is 0.040 9. In this study, we can obtain the prediction model of the roughness of the bevel cut with a high degree of accuracy, and achieve the effective prediction of the roughness of the bevel cut of the high-power laser. This study can obtain a high accuracy prediction model of bevel cutting roughness and achieve the effective prediction of high-power laser bevel cutting roughness.
Deep learning-based point cloud semantic segmentation models often adopt complex attention mechanisms for improvement but show deficiencies in extracting local deep semantic features and neighbor point feature expressions. Therefore, this paper proposes a point cloud semantic segmentation model that combines the weighted Knearest neighbors algorithm with convolutional block attention. On the architecture of the Dynamic Graph Convolutional Neural Network, a weighted K-nearest neighbors algorithm is designed to obtain more effective local neighborhoods; then, convolutional attention is introduced to process features within the local neighborhoods. In the convolutional attention, channel attention is used to enhance the correlation among point cloud channels, and spatial attention is applied to perceive the three-dimensional spatial structure, acquiring contextual information and deep semantic features. Experimental results show that the model achieves an average Intersection over Union (IoU) of 89.92% on the ShapeNet Part dataset and 61.2% on the S3DIS indoor semantic segmentation dataset, demonstrating higher segmentation accuracy compared to other methods.
In order to reduce the negative impact of environmental factors on laser interferometric displacement measurement and meet the accuracy requirements of displacement measurement, a laser interferometric displacement automatic measurement system based on artificial intelligence technology is optimized and designed. Modify the laser, laser interferometer, photoelectric converter, and laser interference signal processor to complete the hardware design of the laser interference displacement automatic measurement system. Taking into account the principles of laser emission, superposition, and interference, automatically generate laser interference signals. By utilizing the artificial neural network algorithm in artificial intelligence technology, the characteristics of light intensity changes in laser interference signals are extracted. Through two steps of displacement calculation and measurement error compensation, the automatic measurement function of laser interference displacement in the system is completed. The experiment concludes that compared with the traditional measuring system, the displacement measurement error of the optimized design system is smaller, the maximum error is only 12.3 m, the measurement accuracy reaches 0.11 m, and the displacement measurement time is only 1.10 min, which has higher measurement accuracy and faster displacement measurement efficiency.
Laser printed documents, as key materials in daily office work, academic research, and legal notarization, have a direct impact on the accuracy and availability of digital information due to the quality of their scanned images. In the actual document scanning process, various factors can easily lead to skewed scanning images. Therefore, this paper studies the adaptive correction technology for skewness in laser printed document scanning images. Using the imaging theory of laser printing scanning technology, describe the pixel conversion relationship of document images during the scanning process; Divide tilt types based on pixel conversion relationships, use radial distortion vectors as correction targets, and establish a correction model under the influence of nonlinear relationships; After correcting the model to obtain the radial distortion parameters of the image, the Newton algorithm is used to solve the parameters and achieve adaptive correction of laser printed document scanning images. The results showed that using 6 types of oblique deformation scanning images as the test benchmark set, after correction by the studied method, the images had a high similarity with the real document, and local deformation was significantly improved.
The detection of signal characteristics plays a key role in target identification, distance measurement and velocity estimation, so the detection model of laser pulse echo signal characteristics is designed to improve the detection effect of echo signal characteristics. The transient radiation transmission model of laser pulse echo signal is constructed by Monte Carlo method. The transient radiation transmission model is solved by Duhame superposition theorem, and the peak power of laser pulse is obtained. According to the convolution transfer relationship between laser pulse emission and target reflection, the laser pulse echo power is obtained. By using echo power and peak power, the target echo signal ratio and scattered noise ratio are obtained, which can be used to describe the change of laser pulse echo signal characteristics and complete the detection of laser pulse echo signal characteristics. The experimental results show that the model can effectively calculate the laser pulse peak power and echo power. In different bands, the model can effectively detect the characteristics of laser pulse echo signal. The larger the band, the smaller the target echo signal ratio, indicating that the band has an effect on the energy attenuation characteristics of laser pulse echo signal. The time domain distribution of target echo signal ratio is exactly the same in different bands, indicating that the band has no influence on the time broadening characteristics of laser pulse echo signal.
The product nondestructive classification testing method based on deep convolutional neural network was studied to realize automatic production, optimize product classification, improve production efficiency and product quality control level. The principle of laser absorption spectrum technology is analyzed, and a near infrared laser absorption spectrum acquisition device is designed to collect the near infrared laser absorption spectrum of the product to be tested. Savitzky-Golay method was used to pretreat the absorption spectra, reduce the interference between spectra, and enhance the purity and sensitivity of spectra. A deep convolutional neural network model with four hidden layers is constructed, and cross-entropy is used as a cost function to implement backpropagation training on the network model. The near-infrared laser absorption spectrum of the pre-processed product to be tested is input into the trained deep convolutional neural network model, and the output result is the nondestructive classification test result of the product to be tested. Experiments show that this method can effectively realize the nondestructive classification testing of products, and the classification recognition rate of different types of products can reach more than 97%, the maximum detection time is 1.11 s, and the detection efficiency is higher.