Abstract:Wafer defect detection is a critical step in ensuring product quality within semiconductor manufacturing processes. Analyzing the distribution areas and presentation patterns of defects enables precise tracing of weak links in production processes, providing essential evidence for optimizing manufacturing technologies. To address the demands for high precision, real-time processing, and lightweight solutions in identifying hybrid wafer defects under complex scenarios, this paper proposes a lightweight wafer map defect detection algorithm based on an improved YOLOv8s-seg. The algorithm first introduces a context-guided mechanism within the backbone feature extraction network to optimize the C2f module. It then proposes a shift-enhanced multi-branch & scale fusion feature pyramid network (SEMF-FPN) to enhance the model′s feature fusion capabilities. Finally, it incorporates lightweight asymmetric detector head concepts to optimize the head network. Experimental results demonstrate that the improved model achieves mean average precision (mAP@0.5:0.95) of 90.9% for bounding boxes and 84.5% for masks, with a detection speed of 333.3 fps. It significantly reduces parameters and computational load by 54.84% and 37.18%, respectively, while maintaining accuracy loss below 0.5%. This lightweight, high-precision and fast-detection model meets industrial production requirements and holds considerable engineering application value.

Abstract:A new five level inverter circuit is designed to address the issues of a large number of components and modulation difficulty in MLIs. This circuit uses a single DC source, eight switching transistors, two capacitors, two diodes and one inductor, reducing the number of components and simplifying the modulation complexity of the system. Compared with existing multilevel inverters(MLIs) circuits, this circuit has the ability to autonomously boost voltage, eliminate leakage current, self balance capacitor voltage, and suppress capacitor peak current. This circuit adopts the SVPWM modulation strategy, which reduces the harmonic content of the output voltage compared to the SPWM modulation strategy. Firstly, the circuit structure and working principle are introduced in the article, and the expression for device selection is established. Then, the SPWM and SVPWM modulation strategies are analyzed. Finally, a simulation model of the circuit and an experimental platform are built for verification, and the results prove the feasibility of the inverter circuit and modulation strategy.

Abstract:Coal and gas outburst is one of the highly destructive geological hazards in coal mining operations, and efficient outburst early warning is crucial for ensuring the safe production of mining areas. To resolve traditional Stacking models′ issues of inefficient hyperparameter optimization, local optima traps, and manual base learner selection, this paper develops an improved Stacking early warning model with modified artificial rabbit optimization （ARO） and differential evolution （DE） algorithms. The AMSARO algorithm—incorporating an elite memory pool and a multi-strategy adaptive selection mechanism—is applied for parameter optimization. The DE algorithm dynamically optimizes weights to realize base learner selection while performing feature selection synchronously. Combining these two algorithms with the Stacking ensemble method, the DE-AMSARO-Stacking coal and gas outburst early warning model is constructed. Eight different benchmark functions are adopted for testing, and experimental results show that the AMSARO algorithm achieves faster convergence speed and higher optimization accuracy compared with the ARO algorithm and its improved variants. Experiments are conducted on a dataset of 500 samples expanded from 50 sets of original data collected from a coal mine in Shanxi Province. The results indicate that the DE-AMSARO-Stacking model outperforms single models and comparative models based on different optimization algorithms in prediction performance. This research provides a more efficient approach for coal and gas outburst early warning.

Abstract:To enhance the technological feel and interactivity of exhibition halls such as museums, and to improve the service level of these halls, this study designs an recognition system based on computer vision technology, featuring functions such as navigation and positioning, as well as intelligent audio guide for exhibits. The system is mainly controlled by Raspberry Pi, with external cameras, infrared remote sensors and other hardware modules. After the image is collected by the camera of the guide vehicle, the path information is identified by the image processing algorithm to control the movement of the vehicle body. The test results showed that when the channel retention rate was 40%, the model parameters were reduced by 18.09%, the floating-point operations were reduced by 12.35%, and the test set accuracy still reached 98.5%. The designed system can accurately identify guide lines, cross lines, and exhibit information, and is easy to deploy and has great promotional value.

Abstract:The GNSS has been widely adopted in maritime ships, however, with the increasingly complex global situation, ports have assumed greater strategic importance during military exercises and other special periods, making them more susceptible to external interference. This can reduce the reliability of ships that rely solely on a single navigation system and may even result in GNSS receiver failure. Therefore, the problem of continuous localization and accurate target ranging for intelligent ships in ports under satellite-denied conditions has become a critical challenge restricting the development of autonomous navigation technologies. To address the practical requirements of intelligent ship navigation in port scenarios, this paper proposes an auxiliary localization approach based on prior port maps. On this basis, a target measurement algorithm is developed by constructing an adaptive ROI partitioning strategy for the shipborne LiDAR according to the ship′s navigation state. The proposed method enables rapid re-localization of intelligent ships and real-time, high-precision measurement of targets within navigable waters under satellite-denied conditions. Field experiments demonstrate that the proposed method achieves satisfactory performance in both auxiliary localization and ranging stability, with an APE of 0.058 m, STD of 0.044 m, and RMSE of 0.073 m. These results indicate a significant improvement in the robustness of intelligent ship localization in complex port environments.

Abstract:Aiming at the key issues of speed cooperative control of dual permanent magnet synchronous motors for transmission line robots, an improved control strategy is proposed. Firstly, in terms of rotational speed loop control, a non-singular terminal sliding mode controller based on the new approach law was designed, and combined with the expanded state observer to conduct real-time observation and compensation of uncertain disturbances in the system, thereby improving the speed tracking performance of the motor. Secondly, in terms of synchronous control, a cross-coupling control architecture was adopted, and a fuzzy control error compensator based on the improved particle swarm optimization （PSO） algorithm was designed. The fuzzy control parameters were tuned through the optimization algorithm, thereby reducing the synchronous error of the dual-motor rotational speed under external disturbances. The simulation experiment results show that the proposed control strategy exhibits significant advantages in terms of robustness, rotational speed tracking accuracy and synchronization performance, providing an effective solution for the high-precision control of live-line operation robots.

Abstract:The traditional implementation of power scheduling in wind farms mainly adopts yaw control methods. However, whether the yaw process of wind turbine generators can be actually carried out depends on the assessment effect of the safety status of the units (nacelle vibration measurement), which leads to inaccurate reliable power scheduling in actual wind farms. For this purpose, this paper proposes a reliable power scheduling method for wind farms based on the prediction of the vibration acceleration of the nacelle of wind turbine generators. Firstly, aiming at the difficulty in modeling the nacelle vibration mechanism during the yaw process of the unit, a deep neural network is utilized to construct the nacelle vibration acceleration model. Then, under the predictive control framework, combined with the prediction of nacelle vibration acceleration, a reliable power scheduling method for wind farms is proposed to improve the accuracy of power safety scheduling in wind farms by 3%. Finally, a wind farm with 12 wind turbine generators was constructed using Fast.Farm to verify the effectiveness of the method proposed in this paper.

Abstract:Aiming at the problem that traditional non-destructive testing methods are difficult to detect rail head damage under different radii, a phased array ultrasonic testing method based on the complex curved surface damage of rail heads is proposed. For the complex structural features of the rail head, a phased array ultrasonic static focusing detection scheme is formulated; based on Fermat′s principle, a calculation method for the delay time of the phased array ultrasonic synthetic beam under different radii of the rail head is proposed; a finite element simulation model is established to analyze the propagation characteristics of the phased array beam in the rail head; based on the discontinuous Galerkintime-domain(DGTD) method, a phased array ultrasonic dynamic response simulation model of the rail head is established and the A-type signal characteristics of different defect positions in the rail head are analyzed. Finally, phased array ultrasonic experiments were conducted on defects pre-embedded at different positions of the rail head and defects with a diameter of 1 mm can be accurately detected. The results show that the optimal average signal-to-noise ratio for defect detection in the central area is 22.97 dB and for the near-surface area with a larger rail head curvature, it is 17.16 dB. Compared with the traditional single-probe ultrasonic testing method, this method increases the testing speed by approximately 40%. Compared with the conventional phased array method that has not been optimized for the rail head′s curved surface, the average defect detection signal-to-noise ratio at the center of the rail head and in the large curvature area near the surface has been increased by approximately 10 and 12 dB respectively.

Abstract:To address the issues of environmental interference and high cost of force sensors in electro-mechanical brake (EMB) systems, this paper proposes a sensorless clamping force estimation method and a composite control strategy. An electromechanical model of the EMB system is established and an improved sliding mode observer using a saturation function is designed to reduce chattering. The finite-time convergence of the observer is proven via Lyapunov theory. An extended state observer is also designed to estimate and compensate for unmodeled disturbances via feedforward control. Furthermore, a novel reaching law is developed by modifying the dual-power reaching law with a variable-gain function to suppress chattering while maintaining convergence speed. A controller based on a nonsingular terminal sliding mode and the improved reaching law is constructed. Bench tests show that the proposed method outperforms dual-power, exponential reaching law and PID methods, offering faster force tracking, higher precision, and enhanced anti-interference capability under various braking conditions.

Abstract:AC electromagnetic field stress detection technology can achieve rapid screening of stress concentration areas in pipelines, but current AC electromagnetic field stress detection technology is limited to uniaxial stress measurement and cannot adapt to the identification of complex stress states in pipelines. This study integrates magneto-anisotropic stress detection technology to establish a theoretical pipeline plane stress model and a finite element simulation model. It designs an orthogonal stress detection structure with dual coil sets. Simulation results verify that under axial stress, the variation amplitude of the longitudinal magnetic flux density reaches 2.61 times that of the transverse direction. Verified through universal testing machine loading tests, within the tensile range of 200 MPa, the sensor′s output voltage exhibits a quantitative relationship with the angle between the stress gradient and stress direction, expressed as ΔU=K*εl-εhcos2θ. The conclusion confirms that this method breaks through the uniaxial detection limitation of traditional ACSM technology, achieves biaxial stress decoupling and principal stress direction determination, providing a new means for pipeline stress state detection.

Abstract:To effectively suppress the influence of time-varying loads and insufficient model accuracy on the control performance of vtol UAV power systems, an ISTSM-ADRC strategy is proposed. By leveraging the characteristic that ADRC is not completely dependent on the accurate mathematical model of the system, a speed-loop expansion state observer（ESO） is designed to online estimate and compensate for the comprehensive system disturbances. The sliding mode surface function with strong robustness is adopted to replace the nonlinear error feedback link in active disturbance rejection control(ADRC), thus enhancing the robustness of the speed-loop controller. A novel super-twisting sliding mode （STSM） reaching law is developed to achieve adaptive gain, reduce chattering and strengthen robustness against disturbances. Step response tests, load mutation tests and parameter perturbation experiments are conducted. The results indicate that compared with STSM-ADRC, the steady-state error of the proposed strategy is reduced by 50% and the steady-state fluctuation range after parameter perturbation is decreased by 46%. These findings verify that ISTSM-ADRC exhibits better dynamic performance and stronger robustness than STSM-ADRC and PI control strategies.

Abstract:To address the problems of missed detections and low detection accuracy of YOLOv11n in photovoltaic (PV) panel defect detection, this paper proposes an improved YOLOv11n-based PV panel defect detection algorithm. A novel SPPF-LDESKA module is designed to enhance the original spatial pyramid pooling fast (SPPF) module. The module integrates the concepts of lightweight detail-enhanced convolution, which combines model lightweighting and fine-grained feature enhancement—with a large separable kernel attention mechanism to expand the receptive field and effectively strengthen the extraction of small-object information.In the feature pyramid network （FPN） stage, inspired by the context-guided feature modulation (CGFM), an efficient shared convolutional module (ESCM) is designed. This module incorporates the efficient channel attention （ECA） mechanism, employing 1D convolution along the channel dimension to perform local inter-channel interaction. Furthermore, shared convolution is utilized to reduce redundant computation and prevent repetitive learning, thereby decreasing the parameter count and enhancing the adaptive adjustment of contextual information in multi-scale feature fusion.Additionally, a lightweight detection head is introduced, leveraging shared convolution to further reduce model parameters. Experimental results demonstrate that, compared with the original YOLOv11n model, the proposed method reduces computational cost by 6.25%, accuracy improved by 4.5%,while mAP@50 improves from 87.5% to 89.5%, achieving a 2.3% performance gain. These results validate that the improved algorithm achieves superior performance and better adaptability for PV panel defect detection tasks.

Abstract:To address the issues of low detection accuracy, high computational cost and deployment difficulties in existing tomato leaf disease detection algorithms, this paper proposes a lightweight improved algorithm, YOLO11-WCL, based on the YOLO11n model. The algorithm introduces lightweight anti-aliasing wavelet pooling in the backbone network to replace traditional downsampling operations, effectively reducing network complexity. In the Neck part, a lightweight CA-HSFPN module is incorporated to enhance multi-scale feature fusion, while the LADH detection head is adopted to improve inference speed. Based on the experimental dataset, comparative analyses of lightweight networks, different detection algorithms, and ablation experiments were conducted. The experimental results show that the YOLO11-WCL model achieves only 50.1% and 60.3% of the parameters and computational complexity of the original YOLO11n model, corresponding to 1.29 MB and 3.8 GFLOPs, respectively, while reaching an mAP of 98.5%. These findings demonstrate that the proposed algorithm significantly improves the detection accuracy of tomato leaf diseases while maintaining model compactness, achieving high detection efficiency and good generalization performance. It is suitable for deployment on UAVs and other mobile devices, with broad application prospects and practical market potential.

Abstract:To address the issue of degraded matching accuracy in reflective regions, a stereo matching algorithm is proposed based on a multi-kernel attention decoder and a cost volume dynamic enhancement mechanism. First, the feature extraction network employs a multi-kernel attention decoder, which utilizes parallel depth-wise convolutions to capture spatial details at different resolutions. This enhances the algorithm′s ability to fuse local details and global information, thereby improving disparity prediction accuracy in reflective regions. Next, the algorithm adopts a cost volume dynamic enhancement mechanism to optimize the cost volume. By selectively fusing multi-scale cost vol-umes based on global contextual information, it avoids redundant information accumulation, thus strengthening the cost volume′s modeling capability in reflective regions and mitigating the accuracy degradation problem. Experi-mental results on the Scene Flow dataset show that the proposed method achieves 0.45 EPE and 2.40% D1, while on reflective regions of the KITTI dataset, it attains 4.59% 3-All error, representing an 8.2% reduction com-pared to baseline methods. These results demonstrate that the proposed approach effectively improves matching accuracy in reflective regions while also reducing model parameters.

Abstract:In order to solve the problem of low detection precision caused by complex background environment, large target scale difference, various categories and uneven distribution, an improved YOLOv11n defect detection algorithm for fan blades is proposed. Firstly, an efficient multi-scale convolution EMSConv was innovatively proposed, and C3k2 was redesigned to enable the model to efficiently capture the input feature map information and enhance the detection accuracy; secondly, introduce the large separable kernel attention mechanism and Residual-Conv design SPPF_LSKR module to replace the original pyramid pooling module, enrich the context information and improve the multi-scale feature extraction and fusion capabilities of the model; thirdly, the re-parameter shared convolution detection head (RSCD) is adopted to reduce the number of parameters and computational load of the head by sharing parameters, thereby enhancing the speed and accuracy of the defect detection task; finally, a loss function Inner-Wise-MPDIoU is proposed based on the ideas of MPDIoU, Inner _IoU and Wise_IoU, which balances the detection of defects of different scales and accelerate the convergence speed of the model. The results of the experiments indicate that the modified YOLOv11n model attains an mAP value of 89.2%, which is 2.9% higher than that of the original model. The results show that the improved model can meet the needs of efficient and accurate detection of fan blade defects.

Abstract:With the extensive application of multi-task learning in complex business scenarios such as intelligent recommendation and autonomous driving, the handling mechanism for heterogeneous task priorities has emerged as a critical bottleneck constraining model performance. Existing approaches predominantly rely on empirical strategies of linear loss function combination, which suffer from dual deficiencies: First, The manual parameter-tuning paradigm for task weighting leads to combinatorial explosion in hyperparameter search space as task quantity increases; second, gradient competition among tasks induces negative transfer effects that significantly erode the performance boundaries of high-priority tasks. To address these challenges, this study proposes an innovative constrained optimization-based progressive multi-task learning method. By encoding task priority structures into inequality constraints, we formulate an optimization paradigm with strict priority guarantees. A constrained optimization framework is established through Lagrangian duality theory to ensure that the performance lower-bound constraints of high-priority tasks remain unaffected by secondary task optimization processes. Meanwhile, a progressive gradient projection algorithm enables dynamic adjustment of constraint spaces. Theoretically, we provide convergence guarantees through non-convex optimization theory. Experimental results on public datasets demonstrate that our method enhances the performance of secondary tasks while ensuring the stability of high-priority tasks, establishing a novel theoretical framework and technical pathway for multitask learning.

Abstract:Addressing the limitations of the dung beetle optimization algorithm, such as weak global search capability, slow convergence rate and susceptibility to local optima, this paper innovatively proposes an enhanced dung beetle optimization algorithm based on the golden sine algorithm, named GSDBO algorithm. Firstly, the population initialization is optimized using Tent chaotic mapping and lens imaging reverse learning strategy to generate high-quality initial solutions; secondly, an improved golden sine algorithm is adopted to replace the original position update mechanism of the rolling ball beetle, in order to improve the global search accuracy and convergence speed of the algorithm; finally, by combining the t-distribution mutation strategy and adaptive adjustment mechanism, a dynamic balance is achieved between global exploration and local development capabilities. The experiment used CEC2005 and CEC2020 test functions, combined with Wilcoxon rank sum test, to verify the effectiveness and feasibility of the proposed algorithm. The results showed that the GSDBO algorithm exhibited significant improvements in convergence speed and solution accuracy. In the three engineering optimization problems of cantilever beam design, welding beam design, and robot path planning, this algorithm has obtained the optimal solution, further verifying its effectiveness in solving complex practical problems.

Abstract:Maliciously tampered images pose serious threats to both daily life and society. Although numerous detection models have been developed for image tampering detection, they still suffer from issues such as the loss of fine details and insufficient capability in detecting small targets. To address these challenges, this study proposes a multi-scale cross-layer fusion-based for image tampering localization. Built upon the RRU-Net framework, the model changes the skip connection scheme and enhances the encoder features in skip connections using the MSC-SC structure with atrous spatial pyramid pooling. A cross-layer fusion module is then designed to adaptively fuse the encoder and decoder feature maps. In addition, a triple attention mechanism is introduced to enhance feature perception during the down sampling process. Finally, a joint loss function is utilized to alleviate the imbalance between positive and negative samples. Experiments conducted on the CASIA v2 and COLUMB datasets demonstrate that the proposed method achieves F1 score improvements of 11.33% and 6.84%, respectively, compared with the original RRU-Net, indicating that significant effectiveness is achieved.

Abstract:Addressing the issues of insufficient accuracy and low computational efficiency in quantitative precipitation estimation for dual-polarization radar, this paper proposes a lightweight multi-scale fusion algorithm, DPCR-Net, based on an improved DeepLabV3+. This algorithm employs a lightweight backbone network combined with inverted residual and dilated convolution to reduce parameters while maintaining feature extraction capabilities; enhances feature reuse and sensitivity to small targets through a densely cascaded adaptive feature pyramid; and utilizes a hybrid attention decoder to fuse multi-scale features, suppress noise and preserve details. Experimental results on the NJU-CPOL dataset show that the method achieves a mean absolute error (MAE), root mean squared error (RMSE) and correlation coefficient (CC) of 0.519 7, 4.174 4 and 0.661 0, respectively, with a hit rate of 0.897 4, a parameter count of only 4.11 M and a computational cost of 8.29 G. Compared to mainstream models, the algorithm proposed in this paper improves both estimation accuracy and computational efficiency, enabling high-precision real-time precipitation estimation and deployment on edge devices.

Abstract:Denoising diffusion probabilistic models（DDPM）have demonstrated powerful image generation capabilities and have been successfully applied to image inpainting. While many recent approaches introduce structural priors to assist the inpainting process and have achieved promising results, they still suffer from limitations in structural integrity and the naturalness of texture details, often resulting in fractured or discontinuous regions. To address these issues, this paper proposes a gradient-guided feature reconstruction algorithm based on diffusion models. Specifically, the gradient map of the corrupted image is first restored to preliminarily reconstruct its structure and texture. The restored gradient map is then used as generation guidance to assist the inpainting of the original image. Furthermore, an efficient channel attention (ECA) module is integrated into the noise prediction network to enhance feature interaction and improve the consistency of the reconstructed images. Experimental results show that compared with state-of-the-art methods, the proposed approach achieves improvements of 3.19% in PSNR and 2.74% in SSIM, and a reduction of 8.82% in LPIPS on the CelebA-HQ dataset. On the Places2 dataset, it achieves gains of 0.42% in PSNR and 2.81% in SSIM, with an 2.75% decrease in LPIPS, demonstrating superior capability in structural restoration and detail preservation for image inpainting tasks.

Abstract:Aiming at the problem of defects occurring on the surface of 3D metal printed parts due to technology and materials, a surface defect detection method for 3D metal printed parts based on an improved Canny algorithm is proposed, which processes the defect images of 3D metal printed parts by using the improved algorithm. First, an improved bilateral filter is used to replace Gaussian filtering to handle the surface defect images of metal printed parts under salt- and-pepper noise. Then, a dynamically weighted four-direction Scharr operator is adopted to calculate the gradient amplitude and determine the gradient direction, so as to better highlight the edge information of 3D metal printed parts. Next, an improved non-maximum suppression algorithm is employed to further process the image. Finally, the dynamic threshold and hysteresis edge tracking algorithm are utilized to realize the selection of high and low thresholds and enhance the continuity of edges. By comparing the defect detection results of different algorithms on the surface of 3D metal printed parts, the experimental results show that the improved Canny algorithm performs better in noise smoothing and edge continuity than the traditional algorithm. Specifically, the peak signal-to-noise ratio (PSNR) value is increased by 46.70% compared with the traditional Gaussian filter, the structural similarity index (SSIM) value is improved by 39-93%, and the probability of figure of merit (PFOM) value of the processed image is enhanced by 36.46% compared with that of the traditional Canny algorithm. This method can effectively detect the defects on the surface of 3D metal printed parts and has strong practicability.

Abstract:Aiming at the challenges of small industrial defect targets, blurred boundaries and complex backgrounds, this paper proposes a lightweight industrial defect detection network named MSCENet based on multi-scale context enhancement. First, a multi-scale attentionenhanced feature extraction module （MAFE） is introduced, which employs parallel multi-scale dilated convolutions to capture defect features under varying receptive fields. Next, a residual enhancement fusion （REF） module is designed to adaptively integrate multi-level features using a dual-attention feature enhancement mechanism alongside residual connections, thereby improving the reconstruction quality of defect boundaries and details in the decoder. Furthermore, a global attention aggregation （GAA） module is proposed to focus on defect regions while suppressing background interference, further enhancing detection accuracy and robustness. Experimental results on three industrial defect datasets demonstrate that, compared to the backbone network FasterNet-T1, the proposed method achieves significant improvements in mean intersection over union （IoU）, mean pixel accuracy, and overall accuracy, with only 9.418 million parameters, while improving industrial inspection efficiency.

Abstract:To address the issues of long median search time, complex processing, and poor denoising performance of adaptive median filtering under high-density salt-and-pepper noise, a neighborhood-priority multi-level mean filtering algorithm is proposed. The algorithm uses switch filtering to classify pixels, constructs nine priority regions within a 7×7 neighborhood of noisy pixels, and progressively selects valid pixels for mean filtering. After searching the entire neighborhood, a misjudgment prevention mechanism is applied to restore normal pixels. Experimental results show that the proposed algorithm exhibits good denoising performance and robustness within the noise density range of 0% to 90%. Compared to AFMF, IMAF, NAMD and hybrid filtering algorithms, the proposed method achieves a peak PSNR improvement of over 3 dB at low noise densities, with a 20% improvement in SSIM at 90% noise density. The overall processing speed improves by more than 8 times, and the filtered SSIM remains above 87% even under extreme noise conditions. This algorithm requires no parameter tuning, is simple to implement, incurs low computational overhead, and has strong engineering applicability.

Abstract:Chipping defects on wafer dicing lanes severely impair wafer dicing quality, creating an urgent need for high-precision detection technology to support quality assessment. This paper proposes a sub-pixel edge detection method based on Gaussian fitting. First, image quality is optimized, and dicing lane features are isolated through region of interest (ROI) extraction. Second, accurate initial localization of pixel-level edges is achieved. Finally, edge points are selected with the pixel-level edge point as the center; a sub-pixel edge detection algorithm based on Gaussian peak position estimation is used to calculate sub-pixel coordinates via Gaussian integral curve fitting. Combined with random sample consensus（RANSAC） line fitting and density-based spatialclustering of applications with noise（DBSCAN） clustering, the identification and quantification of abnormal regions are completed. Experiments were conducted using white light interferometer measurements as the reference. Tests on 20 dicing lane samples show that the error range is -1.92~3.73 μm. Compared with the Zernike moment sub-pixel algorithm, interpolation method, and traditional Gaussian fitting algorithm, this method exhibits lower error and better stability, and can provide a reliable technical solution for industrial batch detection.

Abstract:In view of the high requirements for temperature control stability of multi-channel surface plasmon resonance (SPR) sensing systems in the detection process of biomolecular interactions, this study designs and implements a separable local constant temperature system. Taking the thermoelectric cooler (TEC) as the core executive component, this system combines oxygen-free copper clamps, digital temperature control modules, thermistor temperature sensors and air-cooled heat dissipation structures to realize local temperature regulation of the sensing area of SPR microfluidic chips. The experimental results show that the system can achieve stable temperature control in a wide temperature range from 15℃ to 45℃, and the temperature fluctuation range is less than ±0.05℃. After the optimization of external constant temperature compensation, the difference between the actual temperature and the set value in the chip sensing area can be controlled within 0.47℃. After being integrated with the SPR imaging system, the average baseline noise is reduced by 13.3%. The temperature control system designed in this study has the advantages of compact structure, separable form and high temperature control accuracy, which is suitable for multi-channel SPR biosensing systems.