Design and Verification of Beam Diagnostics System for Pepper-Pot Method

Abstract

The pepper-pot method is a beam diagnostics technique used to measure the transverse beam profile, divergence angle, and envelope in particle accelerators. However, its practical application faces challenges, such as insufficient point recognition accuracy and signal quality degradation in complex environments. Based on the Boron Neutron Capture Therapy (BNCT) facility at the Hefei Comprehensive National Science Center—Energy Research Institute (Anhui Energy Laboratory), this study developed an improved pepper-pot beam diagnostics system to optimize the beam quality of the accelerator ion source. The key innovation is adaptive threshold segmentation for spot segmentation, and the experimental results indicate that the enhanced image segmentation method outperforms traditional methods in terms of segmentation accuracy and robustness.

1. Introduction

The pepper-pot method [1,2,3] is a classical beam diagnostics technique for low-energy beams in accelerators and is widely used to measure the spatial characteristics of proton beams, including the transverse distribution, divergence angle distribution, and beam envelope. Compared to other beam measurement methods (such as Faraday cups [4] or wire scanners [5]), this approach offers the distinct advantages of structural simplicity and the simultaneous acquisition of horizontal and vertical beam profiles in a single-shot measurement.

The primary objective is to significantly enhance the measurement accuracy of the beam transverse distribution and divergence angle distribution through an optimized pepper-plate design and image processing algorithms, thereby enabling efficient real-time ion beam diagnostics. Pikin et al. developed a comprehensive emittance measurement system using traditional threshold-based segmentation, achieving reasonable accuracy under controlled conditions [6]. Barabin et al. implemented multi-scale spot detection algorithms for high-intensity beam applications [7]. Pitters et al. proposed machine learning approaches for automated spot recognition [8]. However, these existing methods face limitations in handling non-uniform illumination conditions and achieving real-time processing capabilities required for modern accelerator facilities. Our approach addresses these challenges through adaptive dual-threshold dynamic segmentation and optimized matching algorithms.

The Hefei Comprehensive National Science Center Energy Research Institute (Anhui Provincial Energy Laboratory) is developing a Boron Neutron Capture Therapy (BNCT) [9,10,11,12,13] device based on a DC high-voltage accelerator, as shown in Figure 1. An ion source generates a proton beam, which is transported through an injection section into an accelerating tube. After acceleration by a 2.5 MV DC high voltage, the resulting 2.5 MeV proton beam strikes a lithium target to produce neutrons for the cancer treatment. The parameters of the proton beam injected into the accelerating tube, such as the beam envelope and divergence angle, directly affect both the beam transmission efficiency within the tube and the stability of the 2.5 MV high voltage. To optimize the beam quality generated by the BNCT ion source and enhance the operational efficiency of the BNCT device, this study systematically investigates the challenges in pepper-pot beam measurements and develops an improved pepper-pot beam diagnostics system.

Nevertheless, the practical application of the pepper-pot method faces critical challenges. First, spot identification accuracy (the percentage of correctly detected and localized beam spots compared to the theoretical hole pattern, accounting for both detection rate and centroid position precision) is compromised by factors such as fluorescent screen imaging blur, Charge-Coupled Device (CCD) resolution limitations, and background noise interference. Second, under high-intensity beam conditions, scattered particles and secondary radiation degrade the signal quality. Finally, traditional image processing algorithms exhibit low computational efficiency and insufficient accuracy in automated spot recognition and positioning. These limitations diminish the reliability and precision of pepper-pot measurements, and there are still limitations to enabling the adoption of automated pepper-pot analysis in modern accelerator facilities.

This study validates the performance of the improved image processing algorithms and comprehensively evaluates the effectiveness of the system for proton beam diagnostics. The experimental results demonstrate that the optimized pepper-pot measurement system achieves real-time high-precision monitoring of the beam parameters.

The practical significance of this study is twofold. First, the proposed technical enhancements provide novel solutions for applying the pepper-pot method in high-precision beam diagnostics, addressing the need for improved automation and precision in real-time pepper-pot measurements. Additionally, the developed image processing algorithms and diagnostics system exhibited broad compatibility. Beyond beam diagnosis for BNCT ion sources, they can be extended to medical accelerators [14], industrial ion implantation equipment [15], and plasma beam applications in fusion devices [16], enabling beam characterization and optimization with extensive applications.

Section 1 establishes the research background for the BNCT facility, identifies the challenges of traditional pepper-pot measurements, and outlines key innovations. Section 2 describes the experimental setup and measurement principles. Section 3 describes the software implementation, image processing algorithms, and algorithm performance comparisons. Section 4 summarizes this work and discusses future applications.

2. Experimental Setup and Principle

This section details the structural design and operational principles of the proton beam measurement experimental apparatus based on the pepper-pot method, covering the parameters of the pepper-pot plate and the optical detection system as well as the theoretical foundations of the measurement principles.

2.1. Pepper-Pot Plate and Its Parameters

The pepper-pot device primarily consists of a pepper-pot plate, fluorescent screen, sealing flange, and an externally mounted camera. The beam measurement principle is shown in Figure 2, and the physical diagram is shown in Figure 3. The high-precision pepper-pot plate used in this experiment is made of molybdenum material and processed using laser drilling technology. The hole array adopted a square lattice pattern with an aperture of 1 mm, a pitch of 4 mm, a total plate diameter of 56 mm, and a thickness of 2 mm. A design incorporating a light shield above the plate significantly reduced stray light interference (e.g., plasma luminescence from the ion source and reflections within the vacuum transport tube), thereby enhancing the signal-to-noise ratio of the valid spots on the fluorescent screen.

The scientific-grade CCD camera had a resolution of 1920 × 1200 pixels, a pixel size of 5.86 μm, and a high-sensitivity sensor capable of capturing faint spot signals. To ensure imaging quality across varying beam parameters, it is equipped with a low-distortion wide-angle lens (focal length: 15 mm) that provides a sufficient field of view to cover the entire fluorescent screen. Additionally, the distance between the fluorescent screen and the plate can be adjusted within the 100–260 mm range using spacers of varying thicknesses, thereby extending the measurement range for beam divergence angles. Adjusting the plate–screen distance alters the camera–screen separation. Reliable focusing was achieved through lens adjustments at approximately 20× magnification, delivering clear images. During the experiments, the camera captured images at a fixed frame rate and transferred them via USB to a computer for real-time processing using the self-developed measurement software detailed in Section 3. The original image and color level plot of the pepper-pot beam measurement are shown in Figure 4. All subsequent processing is based on this image. The two highlighted spots in the image are a reference hole (plugged) and a calibration marker.

Furthermore, the light shield between the fluorescent screen and sealing flange featured a vent hole positioned away from the light source (ion source plasma). This design prevents continuous gas emission from the pepper-pot device into the vacuum chamber during beam extraction measurements, thereby avoiding interference with the beam transmission.

2.2. Fundamental Principles of Beam Measurement Using the Pepper-Pot Method

Figure 2 schematically illustrates the core principle: a precisely designed pepper-pot plate (featuring multiple microapertures) and a fluorescent screen collectively capture the spatial distribution of the particle beams. This enables the calculation of parameters, including the beam spot size, transverse distribution, divergence angle, and emittance. When the particles pass through the microapertures, they strike the fluorescent screen, generating luminescent spots that are captured by the camera. The analysis of the spatial information of these spots facilitates beam parameter measurements. Experimentally, proton beams passing through the apertures create an array of spots on the screen. The position and intensity (gray value) of each spot reflect the local divergence angles and beam intensity.

Figure 5 shows the divergence angle measurement. Assuming L denotes the plate–screen distance, when a beamlet passes through an aperture centered at (x_i, y_i), its divergence angles θ_x and θ_y (along x/y directions) cause spot displacement. By measuring the spot centroid position (X_i, Y_i) and combining it with the known aperture location (x_i, y_i) and distance L, each beamlet’s divergence angle can be calculated as follows:

$$\begin{gathered} {\theta }_x={\displaystyle \frac{X_i-x_i}{L}} \\ {\theta }_y={\displaystyle \frac{Y_i-y_i}{L}} \end{gathered}$$

(1)

Further analysis of the angular distribution across all spots revealed beam divergence characteristics, such as the RMS divergence angle or 90% inclusive divergence angle. Additionally, the spot brightness is linearly correlated with the beam intensity transmitted through the corresponding aperture. The normalization of the intensities allows the reconstruction of the beam’s transverse intensity profile at the plate plane. This distribution critically evaluates the beam uniformity and focusing performance.

During practical measurements, spot images may suffer from background noise, spot overlap, and non-uniform screen response. Robust algorithms must be applied for image denoising and other processing methods to extract accurate divergence angle data.

3. Software Implementation and Image Processing Method

This chapter elaborates on the design of beam measurement software and core image processing methods based on the pepper-pot method. The software aims to efficiently extract key proton beam parameters, including the transverse distribution, divergence angle distribution, and beam envelope, through the automated analysis of captured light spot images from experiments. The discussion covers four aspects: system architecture, image preprocessing, feature extraction, and parameter calculation, demonstrating the innovation and application value of the proposed method in beam diagnostics.

3.1. System Architecture Design

3.1.1. Overall Software Framework

The beam measurement software developed in this study employed a modular architecture, as illustrated in Figure 6. It consists of four components: a data acquisition module, an image preprocessing module, a light spot feature extraction module, and a beam parameter calculation module. These modules interact via standardized data interfaces to ensure system scalability and maintainability. The software was built using Python (3.8.10), with image processing functions implemented via open-source libraries such as OpenCV (4.5.3) and NumPy (1.21.2), and a user interface constructed with PyQt (5.15.4) to provide real-time visualization and interactive analysis capabilities.

3.1.2. Data Flow Design

The data processing flow of the software is shown in Figure 7, which includes data preprocessing and image processing. The system first acquires fluorescent screen images in real time via a CCD camera and optimizes the camera parameters by analyzing the image data. The preprocessed images then enter the spot detection phase, in which the centroid coordinates and intensity values of all light spots are extracted. Subsequently, a matching algorithm establishes mapping relationships between the detected spots and theoretical hole positions. Finally, the beam parameter calculation and visualization are completed based on these mappings.

3.2. Image Preprocessing Optimization

3.2.1. Noise Analysis and Denoising Methods

The noise in fluorescent screen images primarily originates from three sources: background, random, and salt-and-pepper noises. Background noise includes fixed-pattern noise caused by uneven Light-Emitting Diode (LED) illumination and CCD dark currents. Random noise mainly stems from thermal and electronic noise, whereas salt-and-pepper noise is often caused by CCD sensor defects or signal transmission interference. To address these noises, this study systematically compared the effectiveness of various denoising methods after further improving the signal-to-noise ratio (SNR) by subtracting background noise and applying image averaging techniques. The evaluation was based on two key metrics: SNR improvement and edge retention rate (This metric specifically evaluates whether denoising algorithms blur important structural features while removing noise). The comparison results are shown in

Table 1. Denoising method comparison.

Method	Final SNR (Unit: dB)	SNR Improvement (Unit: dB)	Edge Preservation Ratio
Median filtering [17]	24.64	0.54	80.56%
Gaussian filtering [18]	24.83	0.73	84.72%
Bilateral filtering [19]	24.43	0.33	81.94%
Morphological operation [20]	25.41	1.31	87.50%
Morphological + median filtering	25.34	1.24	59.72%
Morphological + Gaussian filtering	25.16	1.06	81.94%
Morphological + bilateral filtering	24.67	0.57	80.56%

. The experimental results indicate that morphological operations achieved an effective balance between SNR and edge retention. Specifically, median filtering and bilateral filtering improved SNR by only 0.54 and 0.33, respectively, reflecting their limited effectiveness in noise reduction. In terms of edge retention, morphological operations maintained a rate of 87.5%.

All comparison methods were implemented with optimized parameters to ensure fair evaluation. Median filtering kernel size was selected through cross-validation in the range of 3 × 3 to 9 × 9 pixels, with 5 × 5 providing the best balance between noise reduction and edge preservation. Gaussian filtering standard deviation σ was optimized within the 0.5–2.0 range using grid search, with σ = 1.0 yielding optimal results. Bilateral filtering parameters (spatial σ_s = 75, intensity σ_r = 75) were determined through a systematic parameter sweep. Morphological operations employed a 3 × 3 circular structuring element, validated as optimal for the typical spot sizes (3–7 pixels) in our experimental conditions.

3.2.2. Improved Adaptive Threshold Segmentation Algorithm

Based on the traditional OTSU algorithm [21], this study proposes an adaptive dual-threshold dynamic segmentation method using global statistical features, aiming to address the issues of weak light spot omission and overexposed region misdetection caused by uneven illumination in fluorescent screen images. This method achieves precise segmentation of light spot regions by establishing a dynamic threshold mechanism, with the specific process shown in Figure 8.

The algorithm first calculates the global statistical features of the input image, including:

$$\left\{\begin{array}{c}{\mu }_{global}={\displaystyle \frac{1}{MN}}{\sum }_{i=0}^{M-1}{\sum }_{j=0}^{N-1}I(i,j)\\ {\sigma }_{global}=\sqrt{{\displaystyle \frac{1}{MN}}{\sum }_{i=0}^{M-1}{\sum }_{j=0}^{N-1}{(I\left(i,j\right)-{\mu }_{global})}^2}\end{array}\right.$$

(2)

where M × N represents the image resolution, I(I, j) denotes the grayscale value at pixel (I, j), μ global represents the average grayscale value of the entire image, and σ global represents the standard deviation of the whole grayscale image.

Based on the statistical parameters, dynamic dual thresholds are constructed:

$$\left\{\begin{array}{c}{T}_{high}={\mu }_{global}+2{\sigma }_{global}\\ T_{low}={\mu }_{global}+{0.5\sigma }_{global}\end{array}\right.$$

(3)

The high threshold T_high is used to suppress the halo effect in overexposed regions, while the low threshold T_low preserves the characteristics of weak spots. For overexposed regions satisfying I(x, y) ≥ T_high, morphological closing is applied for spot core reconstruction:

$$I_{core}=(I○B)·B$$

(4)

where B is a 3 × 3 circular structuring element, ○ denotes closing, and ∙ denotes opening. This operation effectively eliminates diffuse halos in overexposed regions while maintaining the geometric center stability of the spots.

For transition regions where T_low ≤ I(x, y) < T_high, an improved local contrast enhancement algorithm is employed:

$$I_{enhance}\left(x,y\right)={\displaystyle \frac{I\left(x,y\right)-{\mu }_{local}}{{\sigma }_{local}+\epsilon }}\times 50+127$$

(5)

In the equation, μ_local and σ_local represent the local mean and standard deviation of the grayscale values within a window centered at (x, y), where ϵ = 0.01 is an extremely small constant to prevent division by zero, and 50 and 127 are empirically determined scaling factors for the best contrast in an image. This processing can enhance the contrast of weak light spot regions by approximately 40%. The window size for computing local statistical parameters (μ_local and σ_local) was set to 5 × 5 pixels based on the following rationale: this size is comparable to the typical spot diameter (3–7 pixels) in our experimental setup, ensuring adequate local statistics while avoiding over-smoothing of spot boundaries. Experimental validation confirmed that this window size provides the optimal balance between local feature preservation and noise suppression. The 40% contrast enhancement was quantified by measuring the average gray-level difference in weak spot regions before (15–20 gray levels) and after processing (25–30 gray levels), representing a conservative estimate of the actual improvement.

After comparing other traditional segmentation algorithms in the next section, this research method was ultimately selected. The segmentation results were obtained by logically combining the processing effects of different regions:

$$BW\left(x,y\right)=\left\{\begin{array}{cc}1\hfill & \hfill I_{core}(x,y)\ge {0.9T}_{high}\\ 1\hfill & \hfill I_{enhance}(x,y)\ge 125\\ 0\hfill & \hfill otherwise\end{array}\right.$$

(6)

In the equation, BW is the binary segmentation mask, 0.9 T_high provides robustness against noise near the threshold, and 125 is chosen as the midpoint for enhanced contrast images.

3.2.3. Experimental Setup

To verify the effectiveness of the dual-threshold dynamic segmentation method, experiments were conducted on the acquired beam images with a resolution of 1920 × 1200 pixels. Experimental environment: ① Hardware: Intel Core i7 processor, 16 GB RAM, NVIDIA GTX 1080 Ti graphics card and ② Software: Python programming language, OpenCV library for image processing, and Matplotlib (3.4.3) library for result visualization. The segmentation algorithm and software were implemented using custom code. To comprehensively evaluate the performance of the proposed method, four segmentation approaches were introduced for comparative analysis: global threshold-based segmentation, adaptive threshold-based segmentation, and classical edge detection algorithms (Canny and Sobel). Four metrics were used to quantitatively assess the performance of these methods: ① Segmentation accuracy: calculates the proportion of overlap between segmentation results and manually annotated boundaries to measure accuracy; ② The F1-score is the blended average of precision and recall: F1-score = 2 (precision × recall)/(precision + recall). Considering the accuracy and recall of the segmentation results, the F1-score was calculated as the overall performance evaluation index of the segmentation method; ③ Robustness: Evaluates performance under noisy conditions by introducing varying levels of Gaussian noise and salt-and-pepper noise; and ④ Computation time: records processing time per image to assess computational efficiency.

3.2.4. Experimental Results

Segmentation experiments were conducted on the acquired pepper-pot beam images using the aforementioned four methods and the proposed method. The results are illustrated in Figure 9 and

Table 2. Comparison of different evaluation methods.

Segmentation Method	Segmentation Accuracy/%	F1-Score/%	Robustness/%	Computational Time/s
Global threshold	90.2	72	70.2	0.12
Adaptive threshold	91.3	82	75.3	0.20
Canny	12.4	14.3	43.2	0.25
Sobel	80.6	66.8	61.9	0.23
Proposed method	94.5	86.4	78.2	0.22

.

As shown in Figure 9 global threshold segmentation can preliminarily distinguish the foreground from the background under simple conditions of illumination. The adaptive threshold segmentation method adapts well to image non-uniformity by considering local grayscale variations, yielding improved segmentation results compared to global thresholding, with clearer boundary contours than global thresholding. However, noticeable artifacts remain in the images. The Canny and Sobel edge detection methods exhibit low accuracy in edge localization. Both methods are sensitive to noise and are prone to generating false edges in noisy regions, leading to unstable segmentation results.

Table 2 demonstrates that the proposed method achieves a segmentation accuracy of 94.5%, significantly outperforming traditional methods and highlighting its suitability for pepper-pot beam image segmentation. The adaptive (91.3%) and global (90.2%) threshold methods showed competent results despite slightly lower accuracy. The proposed method attained an F1-score of 86.4%, indicating an effective balance between precision and recall. In contrast, other methods yielded substantially lower F1-scores (Sobel: 66.8%; Canny: 14.3%), revealing their limitations in handling complex boundaries.

Under noisy conditions, the proposed method maintained a segmentation accuracy of 78.2%, demonstrating robust anti-noise performance. Traditional methods suffer from significant accuracy degradation in noisy environments, particularly the Canny method (43.2%). Regarding computational time, the proposed method processes each image in 0.22 s, which is slightly higher than the global threshold (0.12 s) and adaptive threshold (0.20 s) methods. Nevertheless, given its substantial improvements in segmentation accuracy, F1-score, and robustness, the method maintains a high computational efficiency. This further validates the optimal balance between processing efficiency and effectiveness.

While the quantitative improvements (94.5% vs. 91.3% accuracy) may appear modest, they translate to meaningful practical benefits in automated processing scenarios where consistent spot identification is critical for reliable beam parameter extraction. The enhanced robustness under noisy conditions (78.2% vs. 75.3%) is particularly valuable for real-time monitoring applications where environmental factors can degrade image quality.

3.3. Spot Feature Extraction and Matching Algorithm

3.3.1. Single-Scale LoG Spot Detection Principle

The single-scale Laplacian of Gaussian (LoG) algorithm [24] was used for spot detection, and the LoG operator was detected by Gaussian smoothing and the Laplace edge:

$$LoG\left(x,y,\sigma \right)=-\frac{1}{{\pi \sigma }^4}(1-\frac{x^2+y^2}{{2\sigma }^2})e^{-\frac{x^2+y^2}{{2\sigma }^2}}$$

(7)

The LoG operator combines the advantages of Gaussian smoothing and Laplacian edge detection. The Gaussian kernel parameter σ controls the scale of the filter, with smaller σ values detecting fine features and larger σ values detecting larger structures.

Single-scale LoG spot detection involves the following steps: First, apply the Laplacian of Gaussian (LoG) operator to convolve the preprocessed image. Next, the sub-pixel accuracy positions of the feature points are determined using quadratic interpolation. The spot radius is calculated using the formula r = σ × √2. Finally, false detections are eliminated using an area threshold. Among them, the 40-pixel² area threshold is used to filter the light spots for secondary screening, aiming to eliminate light spots that are not on the same array while retaining the real light spots. Figure 10 shows the results of single-scale LoG spot detection, where green represents the centroids of the spots and white represents the extracted contours of the spots.

3.3.2. Application of Hungarian Algorithm in Spot Matching

After spot detection, a one-to-one correspondence between the detected spots and theoretical hole positions must be established. This study employs the Hungarian algorithm [25] to solve the optimal assignment problem. First, a cost matrix C is constructed, where element C_ij represents the Euclidean distance between the i-th detected spot and the j-th theoretical hole position, as follows:

$$C_{ij}=\sqrt{{(x_i-x_j)}^2+{(y_i-y_j)}^2}$$

(8)

The Hungarian algorithm performs row and column reduction operations to identify zero elements in the cost matrix, followed by minimum line coverage tests, ultimately obtaining an optimal assignment solution with a time complexity of O(n³).

To improve the matching efficiency, a distance constraint mechanism was introduced: based on the geometric relationship of the experimental setup, the maximum possible spot displacement d_max was estimated. When constructing the cost matrix, distances exceeding d_max were set to infinity, thereby effectively reducing the computational load. Additionally, sparse matrix storage technology was adopted to further optimize the computational efficiency for large-scale matching scenarios. Figure 11 shows the matching results of the Hungarian algorithm, where green lines connect the corresponding detected spots (red points) and theoretical hole positions (blue points). The experimental results demonstrate that this method maintains a correct matching rate of over 95%, even with up to approximately 30% spot loss.

3.4. Beam Parameter Calculation Mode

Transverse Distribution Calculation Method

After completing spot feature extraction and matching, the core task is to calculate the transverse beam distribution, which represents the density distribution of the beam in the cross-section and serves as a key indicator of beam quality. Based on the spot matching results, the spot intensity I_i at each corresponding hole position can be directly obtained, which is proportional to the local beam density, as follows: To obtain a continuous transverse distribution of the beam, this study employed the radial basis function (RBF) [26] interpolation method:

$$\rho \left(x,y\right)=\sum _{i=1}^n{\lambda }_i\varphi \left(\right||p-p_i|\left|\right)$$

(9)

where p = (x, y) represents any point on the plane, pi denotes the coordinates of the i-th spot center, φ is the selected radial basis function, and λ_i is the weighting coefficient determined by solving a system of linear equations.

The systematic error mainly arises from the geometric distortion of the optical system, the non-uniform response of the phosphor screen, and the limitations in spot positioning accuracy. Radial basis function interpolation, due to its radial symmetry properties, can effectively compensate for position deviations caused by geometric distortion, while its global interpolation properties ensure reliable reconstruction in sparse data regions. Compared to local interpolation methods, the RBF method can reduce systematic errors and enhance measurement accuracy and consistency. Figure 12 shows the lateral distribution of the proton beam reconstruction, clearly reflecting the beam’s transverse profile. Based on this transverse distribution and spot position information, the beam’s divergence angle distribution can be further calculated using Equation (1). Additionally, by computing the root-mean-square (RMS) size of the transverse distribution, the beam envelope at the pepper-pot plate position can be obtained.

3.5. Software Functionality and Interface Design

The beam measurement software developed in this study adopts a modern interface design, providing comprehensive functional modules.

Image Acquisition and Management: Supports real-time acquisition and historical image loading with image sequence browsing.

Parameter Setting and Control: Users can adjust the processing parameters, such as the denoising methods, threshold settings, and detection sensitivity.

Real-time Processing and Display: Visually presents intermediate and final results during processing.

Data Analysis and Statistics: Provides statistical analysis of parameters, including transverse distribution, divergence angle distribution, and emittance.

Result Export and Reporting: Supports data and image export with automatic generation of beam parameter reports.

The software employs a multithreaded design, in which image acquisition and processing are executed in independent threads to ensure smooth interface interaction. The user operation response time was controlled within 2 s, meeting the real-time monitoring requirements of the experiments. The software also provides extension interfaces for integration with experimental control systems, enabling closed-loop feedback control of the beam parameters and offering robust support for beam tuning.

4. Conclusions

This study successfully developed and validated an improved pepper-pot beam diagnostics system for BNCT applications, achieving measurable enhancements in automated spot recognition and parameter extraction accuracy. The proposed adaptive dual-threshold dynamic segmentation algorithm, combined with Hungarian algorithm-based spot matching and RBF interpolation for transverse distribution reconstruction, demonstrated quantifiable improvements over traditional methods: 94.5% segmentation accuracy (compared to 91.3% for adaptive threshold), 86.4% F1-score, and maintained 78.2% robustness under noisy conditions with 0.22 s processing time suitable for real-time monitoring. While these improvements are incremental rather than transformative, representing a 3.2 percentage point enhancement in accuracy, they provide consistent and statistically significant benefits for automated pepper-pot analysis in accelerator facilities. The system successfully addresses practical challenges, including non-uniform illumination, background noise interference, and the need for reliable automated processing, thereby reducing manual intervention requirements and improving measurement consistency. The developed diagnostics system demonstrates broad compatibility beyond BNCT applications and can be extended to medical accelerators, industrial ion implantation equipment, and other particle beam systems. Although the visual differences between methods appear subtle, the quantitative metrics confirm the practical value of the proposed approach for enhancing beam parameter monitoring reliability in modern accelerator facilities while acknowledging that more substantial improvements may require advanced machine learning techniques or hardware upgrades in future developments.