The working performance of an optical communication system is not only related to the light source, but als-o to its transmission medium. With the development of fiber optic technology and the pursuit of longdistance trans- mission in the field of optical communication, optical amplifiers have emerged. Optical amplifiers can directly amplify optical signals and have great application value in the field of communication. Summarized the basic p-rinciples and development of optical amplifiers. Firstly, the concept, classificaee types of optical amplifiers were introduced, inclu- ding their working princ-iples and performance. Then, the current situation of various types of optical amplifiers at home and abroad was analyzed. Finally, the future development of optiction, and principle of optical amplifiers are in- troduced. Then, the key technologies of the thral amplifiers is prospected.

According to the statistics of the National Cancer Center, there were nearly half a million deaths from upper gastrointestinal tumors in China in 2022, which seriously threatens the health and safety of the people. Photody- namic therapy for upper gastrointestinal tumors is the targeted treatment of lesions by introducing the therapeutic laser beam into the body through endoscopy and conduction fiber, which has become a medical method with the best effect and the least side effects in clinical treatment of tumors in recent years. One of the key factors affecting the efficacy of photodynamic therapy is the type and dose of photosensitizer. This paper studies the development of photosensitizers, their current application status and therapeutic effects in the treatment of upper gastrointestinal tumors. It compares the research progress at home and abroad, analyzes the characteristics and requirements of photosensitizers in photodynam- ic therapy, and discusses the future research directions.

When conducting autonomous navigation, mobile robots have poor ability to avoid obstacles, resulting in a longer time required to complete tasks. Therefore, a method for optimizing autonomous navigation of mobile robots using laser SLAM technology is proposed. Using graph optimization theory, obtain the corresponding sub nodes of all pose extension nodes, calculate the pose with the smallest error through closed-loop detection, update the pose accord- ing to the SPA algorithm, and complete the pose estimation of the mobile robot. By using laser SLAM technology, the optimized pose of the mobile robot is output, and the A? algorithm is used to obtain the optimal passage path. The DWA algorithm is introduced to calculate the trajectory and achieve autonomous navigation optimization of the mobile robot. The experimental results show that the proposed method can accurately estimate the pose of mobile robots. After autonomous navigation optimization, the length of the autonomous navigation path of the mobile robot is only 1 900 mm, the number of turns is only 2, the running cycle is only 190, and the average speed is as high as 280 mm / s. It has good autonomous navigation optimization effect for mobile robots.

Picosecond laser-driven high-energy X-ray sources with excellent characteristics of micro-focus，highbrightness and high resolution，have become one of the popular diagnostic methods for studying micro-crack and micro-jetting phenomena of metal materials． Tin has become the preferred material for ejection research on the melting effectbecause of its low melting point，and aluminum is relatively high due to its relatively high melting point，which is noteasy to impact or unload melting to form droplet injection，so the results of aluminum micro-jetting experimental research are relatively few． Based on the nanosecond loading source and picosecond beam diagnostic source of the XGIIIlaser device，the influence of surface geometric defect configuration on the metal micro-injection process was studied，and high-resolution physical images of two typical metal micro-injection processes were obtained through delay control，and important information such as morphological distribution，average velocity and areal density distribution of theejected substances was obtained． The SPH method was used to simulate the experimental results，and the formationmechanism of the ejecta in the experiment and the source of the high-speed ejected material were revealed．

For the measurement of objects with highly reflective surfaces，the traditional fringe projection method cannot measure the complete 3D information． In order to solve this problem，this paper adopts the region mapping a- daptive fringe projection method ( ＲMAFP) ，which distinguishes the saturated region by the grey value of the image， and then adopts the improved phase-shift algorithm to get the phase and saturation of the saturated region; by the phase value to achieve the projection region mapping，and then project the standard fringe map corrected according to the saturation，and then remeasure the original saturated region，which can eliminate the saturation phenomenon，and combined with the original non-saturated region Combined with the high-precision phase of the original non-saturated area，the accurate measurement of highly reflective objects can be completed． The experiments show that this method is practical and feasible，and can complete the measurement of complex high-reflectivity surfaces．

A CRS-YOLO algorithm based on flexible printed circuit board target detection is proposed to solve the problem that surface defects of flexible printed circuit board are small in size and not obvious in feature, and the exist- ing detection schemes are poor in real-time. On the basis of the single-stage network YOLOv5 model, CA attention mechanism module is first added to Backbone to strengthen the feature extraction capability. Secondly, the SPP pool pyramid is replaced by the Basic RFB pool pyramid with better performance to enlarge the receptive field and reduce the missing rate of the model. At last, SIoU loss function is used to replace CIoU loss function to accelerate the conver- gence of the training model and improve the detection ability of the model. The experimental results show that under the verification of the defect data set of the flexible printed circuit board, the mAP of the CRS-YOLO algorithm is 9% higher than that of the original network model, the detection speed is greatly improved, and the FPS is up to 48, which meets the accuracy and real-time detection of the surface defects of the flexible printed circuit board.

In order to improve the feature extraction ability of the network in complex environments, a lane line recognition method for double feature extraction network is proposed. Firstly, a double feature extraction network is constructed to reduce the loss of detailed semantic information and enhance the recognition ability of the model in com- plex environments. Then, the improved atrous spatial pyramid pooling structure is used to increase the receptive field and extract more rich contextual information. In addition, depthwise separable convolutions are combined to reduce the computational complexity of the model. Finally, a channel attention module is constructed to focus on feature channels with more effective information. Experimental results show that the proposed method achieves an accuracy of 97. 7% and an mIoU of 76. 2% on the Tusimple dataset, with a single image recognition time of 26. 24 ms. When recognizing lane lines in complex environments, the proposed method demonstrates good robustness.

To address the problem of poor object detection due to low contrast and inconspicuous object contours in infrared images in security scenes, an improved YOLOv7-based infrared security object detection algorithm is pro- posed. The recursive gated convolution is used to improve the backbone network and enhance the ability to interact with higher-order information of the input image; the ELAN-S module is constructed using the SimAM attention mech- anism to reduce the information loss rate while reducing the network parameters; the anchor box size is optimized using the K-means++ clustering algorithm to improve the detection accuracy. Data enhancement and experiments are con- ducted on the InfiRay public dataset, and the results show that the algorithm proposed in this paper has an mAP value of 87. 15% while maintaining a high detection speed, which is a significant improvement compared with the original YOLOv7 network and other mainstream algorithms, proving that the improved method is advanced and effective.

Aiming at the problem of poor imaging quality of two-dimensional laser warning system in extreme envi- ronment, an adaptive control system based on Kalman algorithm is designed. Two InGaAs focal plane array detectors are used to measure the Angle of the incoming laser, respectively, and the Kalman algorithm is derived and optimized and verified by simulation. At the same time, based on the improved Kalman algorithm, the adaptive integral control of the two detectors is realized by FPGA to achieve the adaptive control of the system output. The results show that the average error of Kalman prediction for coarse measurement is 0. 78%, and that for fine measurement is 0. 56%. Under the illumination of 532 nm simulated light source, the adaptive control system reduces the interference of under-expo- sure and over-exposure under extreme light intensity, and the spot size is stable at about 55 pixels under different light intensity. The experiments show that the adaptive control system can improve the imaging stability of the laser warning system under different light intensity, which is of great significance for the laser warning system to work in harsh envi- ronment.

Accurate identification and positioning of aircraft targets is the key to the victory of aviation safety and information war. In view of the problems that traditional aircraft target identification has poor anti-interference perform- ance and is sensitive to occlusion, illumination and scale and difficult to cope with complex scene requirements, an aircraft target detection algorithm based on improved YOLOv5 is proposed. IOU-NWD Similarity Metric for Bounding Boxes solves the ambiguity of label assignment for aircraft small targets by IOU mechanism. Using GFPN based on NL- net module, the " cross-layer" and " cross-scale" adaptive fusion is completed, and more abundant and representative characteristic information is obtained. soft-NMS method is used to solve the problem of missing detection of small air- craft targets in crowded target areas. The experimental results show that compared with the original YOLOv5, the Pre- cision, Recall, mAP0. 5 and MAP0. 5:0. 95 of the improved model are increased by 1. 9%, 10. 4%, 3. 6% and 5. 8%, respectively. Through targeted network adjustment and module migration, the algorithm can improve the detec- tion effect of the model on small and blocked aircraft targets. The superiority of the algorithm is verified by experi- ments. The experimental results show that AIR-YOLO is superior to YOLOv5 in terms of detection accuracy and ro- bustness, which solves the problem of false detection of small aircraft targets in the original YOLOv5 algorithm.

To further effectively suppress the noise of the detection system and improve the detection accuracy, an improved minimum mean squared deviation adaptive algorithm is studied for noise reduction in the methane concentra- tion detection system of tunable diode laser absorption spectroscopy. Through matlab simulation experiments based on TDLAS technology, the methane detection system is built, and the methane absorption peak position at 1 653. 72 nm is selected to analyze the relationship between the filtering order, step factor and sampling period on noise in the LMS adaptive algorithm, and improve the selection of parameters to optimize the noise reduction process with the best filte- ring effect. It is shown that the optimal filtering effect can be achieved by converging to the optimal filtering order and step factor at the high frequency sampling time. The results show that the signal-to-noise ratio is effectively improved by 94% and the goodness-of-fit R2 reaches 0. 997, which proves that the improved LMS adaptive filtering algorithm can effectively suppress the effect of noise on the second harmonic signal.

Extracting the center of the light stripe of the line structure is a key step in the three-dimensional meas- urement system. In view of the complex lighting environment and the time-consuming problem of the Steger algorithm, a fast and accurate method for extracting the center of the light stripe is proposed. Filtering algorithm is used to de- noise, top hat operation and edge detection to eliminate complex lighting in the background; then the improved Rosen- feld skeletal is used to obtain the initial center, and the initial center point and the upper and lower contour points of the stripes are fitted to calculate the normal direction of the initial center and the line width of the light stripes; finally, the sub-pixel center point is obtained based on the improved Hessian matrix. Experiments show that the algorithm is suitable for complex and changeable environments; the running time of the algorithm is 0. 151 s, which is nearly 8 times higher than that of the Steger algorithm; the extraction accuracy is 0. 23 pixel, which retains the high precision of the Steger algorithm and can achieve fast and accurate extraction of the center of the light stripe.

An Improved Whale Optimization Algorithm (IWOA) -BP neural network image restoration model was proposed to solve the problem of obvious lag in the process of restoring degraded images by traditional restoration algo- rithms. First, the uniformity and diversity of the initial population were enhanced by Tent chaos. Secondly, nonlinear weights and improved convergence factors are used to balance the global search and local optimization capabilities of the algorithm. Finally, the Levy flight strategy is combined to update the individual position to help the algorithm es- cape the local optimal. Then the IWOA-BP model is established by using the classical image data. PSNR, SSIM and NMSE were selected as the evaluation indexes of the network model, and compared with BP, GWO-BP and WOA- BP. The experimental results show that IWOA-BP model has better visual effect and improves the quality of image res- toration.

Due to factors such as imaging equipment and transmission environment, laser remote sensing images exhibit significant blurring. To restore the quality of laser remote sensing images, a fuzzy laser remote sensing image reconstruction method based on visual communication is proposed. Using visual communication technology to obtain fuzzy laser remote sensing images, image features are extracted from texture, statistical distribution, frequency, and other aspects as image reconstruction rules. Through image registration, deblurring, noise, and other steps, the recon- struction results of fuzzy laser remote sensing images are obtained and output in a visualized form. Through testing ex- periments, it was concluded that compared with traditional reconstruction methods, the optimization method achieved a peak signal-to-noise ratio of 38 dB and a structural similarity of 0. 94, indicating that the optimization method has sig- nificant advantages in reconstruction quality.

When there are multiple focused target subjects in the infrared image, it will lead to a decrease in the quality of the infrared image and blurring in some areas. Therefore, a deep neural network-based nonlinear enhance- ment method for multifocal infrared images is proposed. Under the guidance of guided filtering, detail layer images and background layer images are obtained from infrared images, and a local clarity evaluation function is established in the detail layer image to obtain clear detail layer and background layer, and the two are fused. Using the constructed deep neural network structure, establish a nonlinear gain function, and achieve nonlinear enhancement of multi focus infra- red images by setting thresholds and adjusting subband coefficients. The test results show that the proposed method did not change the brightness and texture fluctuations of the original image during the enhancement process; The informa- tion entropy of the enhanced image is about 57% higher than that of the original image. The image anti noise perform- ance value is higher, with an average of 7. 4 dB. The image is clearer, and the SSIM value is closer to 1, with an av- erage of 0. 98. The enhanced image has a higher similarity with the original image, which is closer to the real image.

The resolution of laser night vision images is low, and edge information is difficult to detect, resulting in low efficiency and poor accuracy in the image segmentation process. Therefore, a laser night vision image automatic segmentation method based on Roberts operator is designed. Use a camera to collect color information from nighttime scenes, reconstruct image data using the weighted average method, and output a complete laser night vision image. Fit adjacent pixels around the image, use cubic spline interpolation function to smooth the image, use Roberts operator to calculate image gradient, achieve edge detection of laser night vision images, obtain smooth information after image in- terpolation, and improve the resolution of night vision images. By using fuzzy C-means clustering to clarify the cluste- ring range, chaotic particle swarm optimization algorithm is introduced to achieve image segmentation. The experimen- tal results show that the segmentation difference rate of the proposed method is below 3%, the Dice similarity coeffi- cient of image segmentation is above 0. 97, and the average image segmentation time is 9 seconds.

In complex and changing lighting environments, images are prone to distortion, leading to blurring of video monitoring details and directly affecting the accuracy and effectiveness of image data. In order to improve the quality of blurred images, a method for enhancing blurry details in video monitoring images based on grayscale com- pensation is proposed. The guided filtering method is used to extract the brightness component values of exposed and under lit images, and the two-dimensional linear relationship between the brightness component and image resolution is solved. Dynamic stretching or compression adjustments are given based on the maximum and minimum values of the imbalance in the linear relationship of the image. On this basis, image regions are divided, and a fuzzy set domain is established based on the membership relationship between the brightness components of each region and the fuzzy set. The grayscale values of pixels belonging to the fuzzy set domain are calculated, and the enhancement of fuzzy details is achieved by adjusting the grayscale values. Experimental results show that the proposed method can reduce the inter- ference of complex lighting conditions on the image, and complete the fuzzy detail enhancement with high efficiency and high quality. The peak signal-to-noise ratio of image enhancement is as high as 36 dB, and the structural similar- ity is closest to 1, indicating that the image enhancement effect and applicability of the proposed method are better.

The quality of image fusion in complex visible light images is affected by factors such as occlusion and overlap. It is necessary to optimize the design of image fusion quality evaluation. A visible light image fusion quality e- valuation model based on big data analysis is proposed, and a deep stereo matching model for visible light images is es- tablished using the visual feature extraction method between corresponding image blocks, Map the pixel values of ima- ges collected under different lighting intensities to the embedded feature space, finish preprocessing, construct a dy- namic pixel big data matching model for visible light images, achieve dynamic fusion of visible light images through end-to -end disparity fusion estimation, and the superresolution reconstruction method was used to obtain the real paired images, and the similar contents of SR results and LR images were analyzed. The feature level image distribu- tion domain was used to reflect the visible image fusion quality evaluation, and the visible image fusion quality evalua- tion was realized. The simulation results show that the matching performance of visible image fusion using this method is better, the image contrast and saturation are high, and the imaging quality of visible light is improved. The time consuming is 0. 012 s, the average number of iterations is 1. 569, and the mean square error is only 1. 071, and the to- tal error is only 4. 646. The method effectively improves the image fusion quality. Improve the evaluation effect.

current infrared and visible image registration does not consider the gray level and spatial position, and the average normalized correction retrieval rank of the matching results is high. Therefore, a fusion feature based infra- red and visible light heterogenous image matching algorithm is proposed. Analyze the gray difference characteristics of the middle rectangular area of the image to be matched, draw the Haar feature density distribution map, and implement image filtering based on this. Extracting filtered image grayscale features, pixel density features, and texture features through discrete transformation (K-L), completing the compression and fusion of multiple image features. Based on fusion features, calculate the lattice closeness between the infrared image to be matched and the visible light image, and obtain the matching results of heterogeneous images. The experimental results show that after the application of the proposed method, the average Ar value of the heterogenous image matching results is only 0. 35, which better meets the image matching requirements.

Currently, there are some difficulties in the fusion of infrared and visible light images, resulting in low accuracy, large errors, and low fusion efficiency. In order to solve the problems in the current process of infrared and visible light images, a fusion method for infrared and visible light images based on feature extraction using convolution- al neural networks was designed. Firstly, the infrared and visible light images of the object were collected separately, and the original image was preprocessed for denoising to improve the quality of the image. Then, convolutional neural networks were used to extract the fusion features of the infrared and visible light images, and the fusion results of the infrared and visible light images were obtained based on the features. Finally, simulation experiments were conducted, and the results showed that the fusion ratio of the fusion results of the infrared and visible light images using the pro- posed method increased by 0. 24, The average gradient value increased by 0. 22, resulting in higher image fusion qual- ity.

Aiming at the problems of small target building omission, target building misclassification and boundary bonding in remote sensing image building segmentation by DeepLabV3+, this paper proposes an improved remote sens- ing image building segmentation method for DeepLabV3 +. Firstly, an improved dense cavity pyramidal pooling DenseASPP module is used in the encoder stage to obtain larger sensory fields and denser feature pyramids, and a bar pooling module is added in parallel to enable the backbone network to make effective use of the long-range dependen- cies. Secondly, the SE channel attention module is introduced in the decoder stage to enhance the correlation between channels to obtain richer edge features. Finally, the optimised features from the SE module are fused with the original features to enhance the segmentation performance of the network. The experimental results on the WHU Building data- set show that the building segmentation results of this paper’s method achieve 92. 33% and 95. 54% in the intersection and merge ratio (Iou) and F1 index respectively.

In order to enhance the image of laser imaging radar and improve the image matching effect of laser im- aging radar, a laser imaging radar image matching system based on visual communication technology is designed. Ob- tain the laser imaging radar image. After the image enhancement processing is completed by the optimized Retinex en- hancement algorithm in the visual communication technology, SIFT and rotation invariant LBP methods are used to ob- tain the rotation invariant LBP features of the key points and the area around the key points, and the laser imaging ra- dar image matching is realized according to the similarity judgment of the key points. Experimental results show that the proposed system has a maximum signal-to-noise ratio of 47 dB and a maximum structural similarity of 0. 92. The high- est matching accuracy of image feature points is 97%, and the fastest matching speed is 3. 6 s. The system enhances the image quality of laser imaging radar, and can realize the image matching of laser imaging radar with high precision, which provides a reliable basis and reference for practical work.

In the Optical interconnect system of data center, a digital signal processing scheme for joint equaliza- tion and timing recovery feedback loops is proposed to address the issue of conflicting prerequisites between the clock recovery module and equalization module due to their interdependence. This scheme adopts an improved Gardner feed- back all digital clock synchronization algorithm based on the characteristics of PAM4 signals to reduce clock recovery errors and improve convergence performance; In the equalization module, an improved cascaded multimodal blind e- qualization algorithm based on T / 2 fractional interval was proposed and adopted to reduce steady-state error and im- prove signal equalization performance. The simulation results show that this joint scheme can reduce the system error rate. Under the condition of meeting the hard decision forward error correction threshold, the receiver sensitivity after 40 km transmission is -16dBm, which is improved by at least 3dBm compared to the cascaded scheme. Meanwhile, the joint approach has a stronger ability to resist sampling clock offset (SCO), with a maximum tolerable increase of approximately 200 clock offsets, indicating that this scheme can effectively compensate for linear damage and clock er- rors.

The virtual indoor scene is reconstructed by building a virtual indoor scene, and a virtual indoor scene construction technology based on lidar point cloud data is proposed. The lidar point cloud data is collected by using a three-dimensional optical surface scanning device, and the lidar point cloud data of the virtual indoor scene is filtered by using a fusion Kalman filtering method to reduce the interference of the point cloud data, extract the spatial distribu- tion domain features of the virtual indoor scene, and obtain the global mark points of the measured objects in the virtual indoor scene. Through reverse design and reconstruction of 3D boundary contour feature points, the scattered point cloud of virtual indoor scene is reconstructed and sparse feature points are fitted, and the virtual indoor scene is recon- structed by texture rendering on the fitting surface. The simulation results show that the lidar point cloud data recon- structed by this method is better, and the feature information of the original point cloud is more abundant, which can meet the needs of better reconstruction geometric accuracy, the reconstruction accuracy reached 0. 98, and the recon- struction efficiency reached 4. 5 h.

In order to ensure the security and stability of communication and the integrity of data transmission, an intelligent detection of link defects in multi-core optical fiber network based on incoherent optical frequency domain re- flection is proposed. Firstly, the mathematical model of multi-core optical fiber network is constructed, and the optical fiber node mapping and spectrum allocation are constrained, so that the minimum frequency spectrum occupies the maximum number of frequency gaps. Then the data transmission link model is established by cluster detection, and the communication transmission function is obtained. The link is adjusted according to the correlation matching filter, and the link data is collected by baud interval balancing. Finally, Rayleigh scattered light waves generated by incoherent optical frequency domain reflection step frequency modulation are used to obtain the spatial distribution position of net- work health by combining with inverse Fourier transform to realize intelligent detection of multi-core optical fiber net- work link defects. The experimental results show that the proposed method can accurately detect link defects, with de- tection errors of less than 0. 4%, 0. 2%, and 0. 5 for link blocking defect detection, link average fault risk detection rate, and link spectrum utilization defect detection, respectively. The detection time is controlled within 1. 5 seconds, which can help the network recover operation in a timely manner.

In order to improve the transmission capability of ultra-high definition video, a millimeter wave commu- nication based ultra-high definition video transmission method is proposed. Construct a channel model for ultra high definition video transmission using multi frequency modulation communication networking technology, and establish an adaptive image information exchange and acquisition model for ultra high definition video millimeter wave communica- tion transmission; Combined with Lyapunov exponent spectral density feature analysis, spectral feature decomposition and channel equalization model construction are realized; Establish a model for scene conversion and high-order spec- tral feature extraction for video transmission; Adopting millimeter wave communication technology to achieve filtering compensation and feedback gain adjustment; By using the statistical feature analysis method of joint autocorrelation, a dynamic equalization iterative function is obtained. Through anti-interference suppression and equalization scheduling in complex channel environments, lossless transmission, encoding and decoding of ultra-high definition videos are a- chieved. The simulation results show that the bit error rate of UHD video transmission with this method is low, with an average of 0. 058 2, the Transmission delay is low, with an average of 248. 4 ms, and the peak signal-to-noise ratio is 49 dB.

Near infrared facial expression recognition mainly relies on local features of lazy images. When the ex- tracted features are interfered with, the accuracy of facial expression recognition is low. Therefore, a new near-infra- red facial expression recognition method based on deep learning networks is designed. Relying on the local optimization and preservation method of the image to reconstruct the image structure information, the reduced dimensionality near- infrared facial image is obtained. The application point distribution model detects all key points on the face, extracts regions of interest for facial expression recognition, and constructs a facial expression classification and recognition model using a deep learning network architecture. By adjusting the parameters of the recognition model, the recogni- tion results of facial expressions are obtained. The experimental results show that the average Acc value of the proposed method’s recognition results reaches 0. 95, greatly improving the accuracy of near-infrared facial expression recogni- tion.

In order to solve the distortion problem caused by nonlinear effects on fiber optic communication signals during transmission and improve the stability of fiber optic communication systems, an improved k-means multi domain fiber optic communication nonlinear distortion compensation method is proposed. Build a multi domain optical fiber communication transmission model, use a wavelength converter at the transmission end to transmit the input signal to the fiber, and linearize the pulse recovery optical signal based on interference principle. Describe the dispersion char- acteristics of fiber optic communication using signal-to-noise ratio, and clarify the occurrence of nonlinear distortion changes in signal interaction. By improving the k-means method through the Dijkstra method, the distortion constella- tion is demodulated to avoid clustering falling into local optima, making all cluster signals as close as possible to the o- riginal modulation center, and achieving distortion compensation. The experimental results show that after compensa- ting for nonlinear distortion in fiber optic communication using the proposed method, the clustering effect is better, and the information error rate can be reduced to 10-7 , effectively reducing network transmission consumption and improving the quality of fiber optic communication signals.

The node failure wave of the near infrared Internet of Things has an important impact on the network connection reliability. In order to improve the security protection of the near infrared Internet of Things, a node failure wave impact analysis model based on link survivability traffic is constructed. Construct the behavior trust profile of the near-infrared Internet of Things nodes, weighted sum the subjective trust value, the objective recommended trust value and the past trust value, get the comprehensive trust value of the node behavior, and evaluate whether the node is ab- normal; Use node importance and load capacity to redistribute the failure load, build a node failure propagation model, and calculate the additional service load of random adjacent nodes. If it is higher than its own load capacity, the node has failure risk; Derive the link survivability according to the link duration and residual capacity, calculate the incom- ing traffic and failure wave impact probability of adjacent nodes, divide the node failure wave impact level, and obtain the node failure wave impact of the near infrared Internet of Things in different periods. The experimental results show that the constructed model can accurately describe the impact of node failure waves in different external environments of the near infrared Internet of Things, and provide reference for the stable operation of the near infrared Internet of Things.

One encryption of fiber optic communication data cannot cause all data to appear in a scrambled state, and data security is still relatively poor. Therefore, a fiber optic communication data scrambling encryption method based on discrete chaotic mapping is studied. Wavelet decomposition is performed on the original fiber optic communi- cation data, and all wavelet coefficients are scrambled based on the Knight’s Patrol Theory and the characteristics of wavelet subbands to complete the scrambling of fiber optic communication data. By constructing a chaotic sequence u- sing discrete chaotic mapping and Sine mapping, the data is encrypted once using the chaotic sequence, and the fiber optic communication data is encrypted twice using a position matrix to achieve scrambled encryption of fiber optic com- munication data. The experimental results show that the proposed method can effectively improve the security of fiber optic communication data, with good data encryption and confidentiality performance.

The optimal channel selection model of the optical communication system is constructed to improve the data transmission rate of the optical communication system and reduce the Transmission delay. First, the sensing time and transmission time between the sender and receiver in the optical communication system, taking the maximum trans- mission rate as the objective function, build the optimal channel selection model, use the improved discrete bat algo- rithm to solve the model objective function, obtain the results corresponding to the minimum fitness value as the current optimal solution, and output the optimal channel selection results. The test results show that this method has good ap- plication performance, with the optimal channel selection ratio results above 0. 944, which can greatly ensure the se- lection effect of the optimal channel; The maximum delay result of data transmission is 1. 08 seconds, which meets the application standards; After selecting the channel, the transmission rate of the optical communication system is around 10 Mbps, with a maximum value of 10. 57 Mbps.

In a number of areas, such as medical imaging, military target recognition, network security and image processing, due to the serious noise interference and large signal mutation, impulse noise problems exist widely. Ai- ming at the problem of image denoising affected by impulse noise, this paper studies the automatic selection of regulari- zation parameters in the method of removing pulse noise based on L1-TV model. The primal dual method is used to solve the problem of constraint model. In view of the difficulty in determining regularization parameters in the model, a method of automatically solving regularization parameters is proposed to reduce the number of repeated experiments. The experimental results show that the regularization parameter method proposed in this paper is robust, which can not only remove the impulse noise in the image, but also better retain the edge and detail information of the image.

Identifying early tomato leaf diseases and pests is one of the key steps in preventing tomato diseases and pests and increasing yield. This paper is based on the improved ResNet50 to identify tomato leaf pests and diseases. Five different tomato pest datasets were created according to different pest and disease categories, and the data were preprocessed by data augmentation. Based on the original model ResNet50, the SE attention mechanism module is added to the network model structure to enable the model to identify the target to be detected more accurately. In addi- tion, in order to reduce the number of parameters of the model and realize a lighter model, the traditional convolution is replaced by deep separable convolution. In order to illustrate the effectiveness of the improved model, the perform- ance of the improved model on the tomato leaf pest dataset was analyzed, and it was compared with the traditional con- volutional neural networks ResNet50, AlexNet, VGG16, and GoogLeNet. The experimental results show that the im- proved model reduces the number of parameters by 37. 5% compared with the original model, and the accuracy reaches 97. 4%, and the accuracy rate is increased by 4. 4% compared with the original model. In summary, this model a- chieves a good balance between performance and parameter quantity, which provides a possibility for the subsequent deployment of tomato leaf pest identification system in the actual environment.

By identifying defects in textile laser printing images, the ability to detect the quality of textile printing is improved. A sparse optimization based method for identifying defects in textile laser printing images is proposed. Different pixel block size feature matching methods are used to achieve defect detection and saliency parameter analysis in textile laser printing. Using texture feature matching method to extract and match feature points of textile printing lace, a sparse feature matching feature detection model for textile laser printing is established based on the difference in texture distribution between cotton threads. Based on the visual feature expression ability of printing lace itself and production, combined with position, scale The feature matching result of rotation invariant realizes defect recognition and detection of textile laser printing image. The test results show that the feature matching ability of using this method for defect recognition in textile laser printing images is good, and the dynamic detection ability of defect parts is strong. It has good ability to screen out false feature points in images and detect features.

Morphology detection has always been a key technology in the machining of mechanical parts. Current- ly, there are many shortcomings in the methods for detecting the morphology of mechanical parts, such as error and time-consuming. In order to obtain more ideal results for detecting the morphology of mechanical parts, a method for detecting the morphology of mechanical parts based on laser vision sensors has been designed. Firstly, the research status of mechanical part morphology detection was studied to find the reasons for the poor performance of mechanical part morphology detection. Then, a laser vision sensor was introduced to collect mechanical part images, and the origi- nal images were denoised and equalized to extract mechanical part morphology detection features. Finally, the mechan- ical part morphology detection was performed based on the features, and a simulation test of mechanical part morpholo- gy detection was conducted. The results show that the proposed method has a detection error of 3 μm for part 1 and 5 μm for part 2. The detection time of part 1 is 7 ms, and that of part 2 is 6. 5 ms. The error of mechanical parts topog- raphy detection is reduced effectively, the accuracy of mechanical parts topography detection is improved, and it has higher practical application value.

A low resolution visual recognition method for mobile robots based on lidar echo signals is proposed to achieve visual tracking guidance and real-time control during the robot operation process through low resolution visual recognition. Using the emitted lidar echo signal as the visual guidance signal of the mobile robot, the ranging parame- ters of the lidar echo signal are extracted. Through radar echo detection and distance parameter estimation, the positio- ning detection and process guidance of the robot visual tracking process are achieved. Combined with low resolution visual enhancement methods, a signal filtering and visual guidance information enhancement model is established to optimize the design of visual recognition algorithms. Design a binocular camera acquisition module, visual recognition module, and visual positioning module to optimize the design of the robot visual recognition system in an integrated programmable logic environment. Tests have shown that the designed system can effectively achieve echo localization detection and visual guidance tracking recognition for mobile robots, with small parallax and high tracking accuracy.

To improve the optimization planning and design capabilities of indoor spaces, a small area indoor en- closed space segmentation method based on point cloud data semantic segmentation under combined light perspective is proposed. Construct a three-dimensional environmental information perception model for small indoor enclosed spaces, extract coordinate information of indoor enclosed space images using indoor space point clouds, and map the fused spa- tial information to the high-resolution spatial heterogeneous unit structure using semantic combination feature segmenta- tion method. Introduce subspace projection feature information with constraints, and combine the high-resolution seg- mentation image model parameter fusion method with combined light perspective, Extract small indoor enclosed end el- ements and use point cloud data semantic segmentation method to achieve spatial segmentation. The simulation results show that this method can effectively realize the Iterative reconstruction of complex indoor scenes. The Root-mean- square deviation of spatial segmentation is low, the maximum is 0. 808%, the peak signal to noise ratio is high, the maximum is 42. 156 dB, and the spatial segmentation speed is fast, the average is 12. 83 ms.

There are many obstacles in the working environment of logistics sorting robots, so a research method for autonomous navigation of logistics sorting robots based on laser SLAM is proposed. Using laser SLAM to determine the distribution of obstacles in the robot’ s working environment. Add the target bias strategy to the RRT algorithm, and by setting fixed and variable probability target bias strategies, make the path nodes more directional when traver- sing the target points, better avoiding collisions between robots and obstacles, and then use the search random tree to search for the shortest path among them. Comparative experiments have shown that compared with conventional meth- ods, this method can guide the robot to follow the shortest path between the starting point and the target point, and the navigation process takes less time, resulting in an error rate and energy consumption of about 5% for robot autonomous navigation.

Traditional methods are unable to obtain ideal infrared weak vehicle target detection results, resulting in large detection errors that cannot meet practical application requirements. In order to address the limitations of tradi- tional infrared weak vehicle target detection methods, timely detect weak vehicles in infrared images, and improve ve- hicle detection accuracy, a convolutional neural network-based infrared weak vehicle target detection method was de- signed. Firstly, the infrared images required for weak and small vehicle target detection are collected, and the noise in the infrared images is processed to eliminate the interference of noise on weak and small vehicle target detection. Then, convolutional neural network is used to establish a weak and small vehicle target detection model. Finally, the performance of the weak and small vehicle target detection method in this paper is tested through specific simulation ex- periments. The results show that the detection accuracy of weak and small vehicle targets using this method exceeds 90%, significantly reducing the false detection rate of weak and small vehicle targets. At the same time, the detection time of weak and small vehicle targets is controlled within 5 seconds, which can meet the real-time requirements of weak and small vehicle target detection and has high practical application value.

A multi-modal fusion technology based infrared face recognition method is proposed to address the cur- rent issue of low accuracy in infrared human recognition. Perform denoising and enhancement processing on facial ima- ges to obtain high grayscale images. Use two-dimensional wavelet transform to extract concatenated feature vectors, and use multimodal fusion algorithm to calculate the contextual attention coefficient of feature channels to quantify fea- ture differences between classes. Combined with modal fusion decision functions, fuse facial image features and calcu- late the matching degree between images based on the optimal classification criteria of the classifier, achieving infrared facial recognition. In the experimental demonstration, the recognition performance of the proposed method was veri- fied, and the results showed that in the recognition of large and small samples, the recognition accuracy obtained by the proposed method was about 95%, which has a high recognition accuracy for infrared facial images.

When measuring the surface roughness of optical components, the measurement results are often one- sided. Therefore, a surface roughness measurement system for optical components based on image preprocessing tech- nology is designed. Obtain the interference image of the original component surface roughness, apply image denoising and tilt correction to the surface morphology image to obtain the filtering result, and implement image denoising through the adaptive median filtering algorithm. In the roughness parameter evaluation module, the evaluation length is defined based on the sampling length, and a comprehensive evaluation of the surface roughness of optical components is imple- mented. Measure the surface roughness of three optical components using a design system. The experimental results show that by designing the system, high imaging quality surface roughness interference images can be obtained, and comprehensive measurement of surface roughness of various components can be achieved, which has certain application value.