A visual method and apparatus for monitoring a homogeneous liquid phase reaction in real time

Abstract

The present disclosure belongs to the technical field of reaction monitoring, and provides a visual method and device for real-time monitoring of a homogeneous liquid-phase reaction, which comprises: collecting an image sequence of a reaction solution in real time by using a non-contact visual sensor; performing noise filtering and white balance correction on the collected image to obtain a preprocessed image; based on the preprocessed image, using a dual-channel fusion mechanism to obtain a comprehensive color index; performing time series analysis on the comprehensive color index, and determining that the reaction reaches an endpoint when the change rate of the comprehensive color index is maintained within a preset convergence threshold range within a continuous preset monitoring period.

Description

Technical Field

[0001] This disclosure belongs to the field of reaction monitoring technology, and particularly relates to a visual method and apparatus for real-time monitoring of homogeneous liquid phase reactions. Background Technology

[0002] In chemical processes such as peptide synthesis, homogeneous liquid-phase reactions are crucial steps that determine synthesis efficiency and the purity of the final product. Taking the coupling reaction of amino acids as an example, this reaction currently relies mainly on offline analysis techniques for progress monitoring, such as high-performance liquid chromatography (HPLC) and the ninhydrin method.

[0003] High-performance liquid chromatography (HPLC), as a widely used analytical technique, has limitations mainly in the following aspects: First, this method requires a complex sample pretreatment process, including sampling, dilution, and filtration. These operations are not only time-consuming and labor-intensive but may also introduce external contaminants. Second, HPLC analysis requires expensive instruments and professional operators, significantly increasing the detection cost. Most importantly, this method cannot achieve true real-time monitoring; each detection requires interrupting the reaction process, resulting in a significant lag in the monitoring data.

[0004] The ninhydrin method, another commonly used detection method, has relatively simple equipment requirements, but its technical shortcomings are more pronounced. This method exhibits significant differences in reactivity to different reactants; some amino acids may react incompletely or weakly, making it difficult to guarantee the accuracy of the detection results. Furthermore, this method has poor selectivity and is easily affected by interference from other components in the reaction system, further impacting the reliability of the detection.

[0005] These traditional methods share a common drawback: the lag in monitoring the process leads to prolonged synthesis cycles, raw material waste, and difficulty in achieving precise control and intelligent management of the synthesis process. This is particularly true when synthesizing longer-chain peptides, where multiple offline sampling and detection processes significantly increase the time and cost of process development. Summary of the Invention

[0006] This disclosure provides a visual method and apparatus for real-time monitoring of homogeneous liquid phase reactions, which can effectively solve the above-mentioned problems.

[0007] This disclosure is implemented as follows: In a first aspect, this disclosure provides a visual method for real-time monitoring of homogeneous liquid-phase reactions, the method comprising: Real-time acquisition of image sequences of the reaction solution using a non-contact vision sensor; The acquired images are subjected to noise filtering and white balance correction to obtain preprocessed images; Based on the preprocessed image, a comprehensive color index is obtained using a dual-channel fusion mechanism, including: Extracting RGB predicted values ​​based on convolutional neural networks; Extracting composite color indices based on traditional image processing methods; The RGB predicted values ​​and the composite color index are fused using a Kalman filter algorithm to obtain the comprehensive color index. A time-series analysis is performed on the comprehensive color index. When the rate of change of the comprehensive color index remains within a preset convergence threshold range for a continuous preset monitoring period, the reaction is determined to have reached its endpoint.

[0008] Secondly, this disclosure provides a visual device for real-time monitoring of homogeneous liquid phase reactions, the device comprising: The acquisition module is used to acquire image sequences of the reaction solution in real time using a non-contact vision sensor; The preprocessing module is used to perform noise filtering and white balance correction on the acquired image to obtain a preprocessed image; The index fusion module is used to obtain a comprehensive color index based on the preprocessed image using a dual-channel fusion mechanism, including: Extracting RGB predicted values ​​based on convolutional neural networks; Extracting composite color indices based on traditional image processing methods; The RGB predicted values ​​and the composite color index are fused using a Kalman filter algorithm to obtain the comprehensive color index. The monitoring module is used to perform time-series analysis on the comprehensive color index. When the rate of change of the comprehensive color index remains within a preset convergence threshold range for a continuous preset monitoring period, it is determined that the reaction has reached the endpoint.

[0009] Thirdly, this disclosure provides an electronic device, including: Memory, the memory storing execution instructions; and A processor that executes execution instructions stored in the memory, causing the processor to perform the method described in the first aspect.

[0010] Fourthly, this disclosure provides a readable storage medium storing executable instructions, which, when executed by a processor, are used to implement the method described in the first aspect.

[0011] Compared with the prior art, the beneficial effects of this disclosure are: This disclosure provides a visual method and apparatus for real-time monitoring of homogeneous liquid phase reactions, realizing real-time, in-situ, non-contact monitoring of homogeneous liquid phase reactions. It can provide feedback on the reaction progress in seconds without interrupting the reaction or sampling, avoiding the lag and contamination risks of traditional offline detection. At the same time, the dual-channel fusion mechanism significantly improves the anti-interference capability and endpoint judgment accuracy in complex reaction environments. Moreover, the hardware cost is low and it is easy to integrate into existing automated synthesis equipment, providing a key process perception means for intelligent closed-loop control. Attached Figure Description

[0012] Figure 1 This is a flowchart of the visual method S100 for real-time monitoring of homogeneous liquid phase reactions provided in the embodiments of this disclosure.

[0013] Figure 2 This is a schematic diagram of the structure of a vision system for real-time monitoring of homogeneous liquid phase reactions provided in an embodiment of this disclosure.

[0014] Figure 3-12 These are comprehensive color index curves of amino acid coupling reactions under different reaction conditions provided in the embodiments of this disclosure.

[0015] Figure 13 This is a comparison chart of the comprehensive color index obtained by real-time monitoring and the detection results of high-performance liquid chromatography by synchronous offline sampling in a typical experiment provided in this embodiment of the present disclosure.

[0016] Figure 14 This is a schematic diagram of the structure of a vision device 1000 for real-time monitoring of homogeneous liquid phase reactions provided in an embodiment of this disclosure. Detailed Implementation

[0017] The present disclosure will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the disclosure. Furthermore, it should be noted that, for ease of description, only the parts relevant to the present disclosure are shown in the accompanying drawings.

[0018] It should be noted that, where there is no conflict, the embodiments and features described in this disclosure can be combined with each other. The technical solutions of this disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0019] Unless otherwise stated, the exemplary implementations / embodiments shown are to be understood as providing exemplary features of various details that provide ways in which the technical concepts of this disclosure can be implemented in practice. Therefore, unless otherwise stated, the features of various implementations / embodiments may be additionally combined, separated, interchanged and / or rearranged without departing from the technical concepts of this disclosure.

[0020] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of this disclosure. The singular forms “a,” “the,” and “the” used in the embodiments of this disclosure and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0021] It should be understood that the term "and / or" used in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0022] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0023] The terms "first" and "second" used herein are merely to distinguish similar objects and do not represent a specific ordering of the objects. Understandably, the specific order or sequence of "first" and "second" can be interchanged where permitted. It should be understood that the objects distinguished by "first" and "second" can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein.

[0024] Example 1 Please refer to Figure 1 This disclosure provides a visual method S100 for real-time monitoring of homogeneous liquid phase reactions.

[0025] Specifically, method S100 includes: S102 utilizes a non-contact vision sensor to acquire image sequences of the reaction solution in real time; S104, perform noise filtering and white balance correction on the acquired image to obtain a preprocessed image; S106, Based on the preprocessed image, a dual-channel fusion mechanism is used to obtain the comprehensive color index, including: Extracting RGB predicted values ​​based on convolutional neural networks; Extracting composite color indices based on traditional image processing methods; The RGB predicted values ​​and the composite color index are fused using a Kalman filter algorithm to obtain the comprehensive color index. S108, perform time-series analysis on the comprehensive color index, and determine that the reaction has reached the endpoint when the rate of change of the comprehensive color index remains within the preset convergence threshold range within a continuous preset monitoring period.

[0026] In some embodiments, the homogeneous liquid-phase reaction system is a stirred reaction system. The visual feature analysis area avoids the mechanical obstruction area of ​​the stirring paddle to ensure the stability of image acquisition and the accuracy of feature extraction.

[0027] In some embodiments, the homogeneous liquid-phase reaction is an amino acid coupling reaction. Amino acid coupling is a key step in the liquid-phase synthesis of peptides. Addressing the limitations of traditional reaction monitoring methods in peptide liquid-phase synthesis, such as poor timeliness, complex operation, and high cost, this method captures visual changes in the solution system during the homogeneous reaction, enabling accurate judgment of the reaction progress and thus overcoming the limitations of existing technologies.

[0028] In the liquid-phase synthesis process, a reaction system consisting of condensing agent TBTU, activator DIEA, and amino acids dissolved in a suitable solvent undergoes a coupling reaction under homogeneous conditions. During this liquid-phase reaction, the visual properties of the reaction system undergo systematic changes, primarily manifested in the solution's color depth, transparency, and hue. These changes in visual characteristics are closely related to the formation of reaction intermediates and the consumption of amino acids. As the coupling reaction proceeds, the visual characteristic parameters of the reaction system exhibit regular changes, and these parameters tend to stabilize when the reaction reaches its endpoint.

[0029] Step S102 employs machine vision technology for real-time image monitoring of the liquid-phase synthesis process, achieving in-situ, non-invasive monitoring. Compared to HPLC sampling, this eliminates the risk of cross-contamination and material loss caused by physical sampling; compared to offline detection, it provides extremely high temporal resolution, capable of capturing millisecond-level instantaneous chemical changes, and providing a complete data chain for subsequent kinetic analysis.

[0030] In step S102, specifically, a visual monitoring system suitable for the liquid-phase synthesis environment is constructed. This system uses a high-resolution industrial camera as the core acquisition device, coupled with a constant temperature control system, to ensure the consistency and stability of imaging conditions. The visual monitoring system includes two operating modes: a background calibration mode, used to establish a baseline image; and a real-time monitoring mode, used to continuously capture image sequences during the reaction process. In the background calibration mode, a baseline image of the reaction system is acquired before the reaction begins to establish a visual feature reference baseline, used to eliminate systematic errors such as ambient light and container color. The baseline image established through the background calibration mode allows the calculation of the comprehensive color index in subsequent steps to be based on a relative increment, significantly improving robustness across devices and batches.

[0031] In some implementations, a high-resolution industrial camera vertically captures a 1080P@1fps video stream of the reaction liquid.

[0032] Step S104 effectively suppresses high-frequency noise and ambient light interference introduced during image acquisition by performing noise filtering and white balance correction on the acquired image, improves the image signal-to-noise ratio and color consistency, ensures the consistency of the monitoring benchmark, solves the problem that visual detection is easily affected by ambient light fluctuations, provides high-quality input data for subsequent dual-channel feature extraction, and ensures the accuracy and stability of the comprehensive color index calculation.

[0033] In step S104, in some embodiments, noise filtering and white balance correction are performed on the acquired image to obtain a preprocessed image, including: Define the visual feature analysis area in the acquired images; The visual feature analysis area is subjected to noise filtering and white balance correction to obtain a preprocessed image, wherein the noise filtering adopts a Gaussian filtering algorithm. The white balance correction is based on the gray world assumption. It adjusts the gain of the R and B channels by calculating the mean values ​​of the R, G, and B channels within the visual feature analysis area, so that the mean values ​​of the three channels tend to be consistent.

[0034] The system defines a fixed Region of Interest (ROI) at the center of the acquired image. ROI locking isolates optical interference from the container edges and liquid surface, while also avoiding mechanical obstruction from the agitator, ensuring the monitoring area is focused on the main body of the reaction liquid. Specifically, the ROI size is 800×600 pixels. Noise filtering and white balance correction are performed on the images within the ROI to obtain a preprocessed RGB image sequence. This sequence is simultaneously input into both the convolutional neural network channel and the traditional image processing channel for parallel feature extraction.

[0035] A Gaussian filtering algorithm with a 5×5 convolution kernel and a standard deviation σ=1.5 is used to effectively smooth high-frequency noise in the image while preserving the edge information of the color change of the reaction liquid.

[0036] White balance correction is used to eliminate variations in the color temperature of the light source. The calculation steps are as follows: 1. Calculate the mean values ​​of the R, G, and B channels for the entire ROI region ( ); 2. Adjustment coefficient: , ; 3. Corrected pixel values: , .

[0037] By using white balance correction based on the gray world hypothesis, the adjustment coefficients of the R and B channels are dynamically calculated with the G channel as the reference, eliminating color drift caused by changes in the color temperature of the light source and fluctuations in ambient light, and making the average values ​​of the three channels tend to be consistent.

[0038] In some embodiments, the method further includes: By using morphological operations and connected component analysis, the interference of bubbles and suspended matter in the reaction system in the preprocessed image is identified and eliminated.

[0039] Specifically, morphological opening operations using circular structuring elements are used to eliminate small bubbles; area and roundness are calculated through connected component analysis to identify bubbles and suspended matter; variance calculation and temporal median filtering are performed on multiple consecutive frames of images to eliminate motion interference; and local binary pattern (LBP) features are calculated to distinguish between uniform textures of the reaction liquid and irregular textures of foreign matter.

[0040] The reason this method distinguishes bubbles from other suspended matter is that the two have fundamentally different interference effects on reaction monitoring, requiring different treatment strategies: Characteristics and interference of bubbles: Bubbles are gaseous phase particles mixed into the reaction system. Their formation is random and transient, mainly caused by physical factors such as stirring and temperature changes, and is unrelated to the chemical reaction process itself. In images, due to surface tension and optical reflection, bubbles typically appear as high brightness, high roundness, and sharp edges. They are pure noise that severely interferes with the analysis of the overall color and turbidity of the solution and must be identified and removed.

[0041] Characteristics of other suspended matter: Suspended particles such as trace precipitates, colloids, or insoluble byproducts that may be generated during the reaction are often related to the chemical reaction. In images, they often appear as low-contrast, irregularly shaped, and with blurred edges compared to the bulk solution. Although they also interfere with imaging uniformity, their presence or changes may contain information about the reaction process and should not be simply filtered out as noise.

[0042] Therefore, the core purpose of calculating roundness is to utilize the spherical physical property of bubbles as a key morphological indicator for initially screening candidate bubble regions from complex backgrounds. By setting a roundness threshold, most typical bubbles can be captured efficiently. However, roundness alone is insufficient for accurate judgment. Therefore, it is necessary to further combine its high brightness characteristics with the spatiotemporal instability (i.e., the rapid movement, deformation, or disappearance of bubbles between consecutive frames) evaluated in the subsequent "multi-frame fusion" and "variance calculation" steps for comprehensive discrimination.

[0043] To effectively eliminate common interferences from bubbles and suspended matter in liquid-phase reactions, the system employs a multi-stage algorithm. 1. Preliminary morphological filtering processing.

[0044] Opening operation to remove bubbles: A circular structuring element (radius 3 pixels) is used to perform a morphological opening operation (erosion followed by dilation) on the image to eliminate small bubbles with a diameter of less than 6 pixels.

[0045] Mathematical morphological operations: corrosion: ; Expansion: ; Opening operation: .

[0046] 2. Accurate identification of connected components through analysis.

[0047] Binarization segmentation: Apply an adaptive threshold (block size 11×11, constant offset 2) to the grayscale image to generate a binary mask.

[0048] Feature filtering: Calculate for each connected component: area Area (Pixel count): Remove areas with an area less than 10 or greater than 500 (atypical bubbles).

[0049] Circularity: ,in This represents the perimeter (number of boundary pixels) of the connected region.

[0050] Based on the characteristic that bubbles in images typically appear approximately circular, the system calculates the circularity of each connected region. Regions with higher circularity are more likely to be bubbles. In some implementations, regions with a circularity greater than a preset threshold (e.g., a range of [0.6, 0.8], preferably 0.7) can be initially screened as candidate bubble interference regions for subsequent steps (such as multi-frame fusion) to perform comprehensive verification and final determination.

[0051] Multi-frame fusion: Variance calculation: Calculate the variance of each pixel over 5 consecutive frames of images (window size 5×5):

[0052] Motion region detection: High variance regions ( This area is identified as a zone of air bubble or other suspended matter movement.

[0053] Median filtering repair: For pixels in motion-interference areas, use temporal median replacement (take the median value of 5 frames) to eliminate temporary interference.

[0054] 3. Texture Analysis: Calculate Local Binary Pattern (LBP) features to distinguish between uniform textures of the reaction liquid (low LBP variance) and irregular textures of foreign objects. For identified interference areas, fill with the mean of neighboring pixels.

[0055] Through the above multi-stage processing, the system can effectively identify and eliminate the interference of most bubbles, and suppress and repair the local optical disturbances caused by other suspended matter, thereby ensuring that the extracted comprehensive color index can stably reflect the overall change trend of the solution.

[0056] Step S106 employs a dual-channel fusion mechanism combining a deep learning channel and a traditional statistical channel, balancing complex feature recognition with algorithm robustness: the deep learning channel learns the complex nonlinear mapping from image features to RGB values, thereby capturing nonlinear features such as changes in solution turbidity and refractive index texture, and identifying intermediate evolution processes that traditional colorimetry cannot quantify; the statistical channel utilizes linear color trends to provide interpretive anchors for the system, effectively preventing prediction drift of the neural network under extreme conditions; the two channels are dynamically weighted and fused using Kalman filtering, significantly suppressing random noise in a single channel, making the comprehensive index curve smoother, and greatly improving the signal-to-noise ratio, laying a reliable mathematical foundation for high-precision first-order derivative calculation and endpoint determination.

[0057] By combining deep learning and statistical channels, the core challenges of weak homogeneous reaction signals and complex interference are effectively addressed: the deep learning channel excels at handling nonlinear interference such as bubbles and uneven stirring, and can capture subtle features of changes in turbidity even when the color remains unchanged; the statistical channel excels at capturing the linear trend of the overall color, with fast computation and strong interpretability; the two channels are dynamically fused through Kalman filtering to form a "double insurance" mechanism: when the CNN misjudges due to abnormal bubbles, the traditional channel pulls it back to the baseline; when the color is stable but the texture evolves, the CNN keenly captures the changes in intermediates, significantly improving the detection sensitivity and robustness of the system in homogeneous and low-contrast environments.

[0058] In some implementations, the convolutional neural network is MobileNetV3, and its structure includes: Input layer: Receives the RGB data of the preprocessed image; Feature extraction layer: consists of multiple Bottleneck modules, each containing depthwise separable convolutions and squeeze-activated attention mechanisms; Output layer: Outputs the RGB predicted values ​​through a fully connected layer.

[0059] In some implementations, specifically, the input layer receives an RGB image of size 224×224 pixels. For example, it may be cropped from the center of a preprocessed ROI region.

[0060] In some implementations, specifically, the feature extraction layer consists of 16 Bottleneck modules.

[0061] Each module adopts an Inverted Residual structure, which includes the following sub-layers in sequence: 1. Dilation Layer: A 1×1 pointwise convolutional layer expands the number of channels in the input feature map to six times its original size. This operation aims to map low-dimensional features to a higher-dimensional space, thereby enhancing their feature representation capabilities and providing a richer information foundation for subsequent deep feature extraction.

[0062] 2. Feature Extraction Layer: A 3×3 depthwise separable convolution is used for efficient spatial feature extraction. This layer is crucial for reducing computational complexity. Depthwise convolution: Each input channel uses an independent convolution kernel for spatial filtering to extract spatial features within a single channel.

[0063] Pointwise convolution: Using a 1×1 convolution kernel, the output of depthwise convolution is combined and transformed along the channel dimension.

[0064] The computational cost of standard convolution is proportional to the product of the number of input and output channels, while depthwise separable convolution decouples this process, significantly reducing the number of parameters and computational cost (typically by an order of magnitude) while maintaining similar feature extraction capabilities. This enables the model to run efficiently on resource-constrained embedded devices.

[0065] 3. Attention Layer: A lightweight squeeze-and-excitation (SE) attention mechanism is introduced, enabling the network to adaptively calibrate channel feature responses. This mechanism consists of two steps: Squeeze: By using global average pooling, the feature map is compressed into a 1×1×C channel descriptor in the spatial dimension (H×W), and global spatial information is aggregated to generate global statistics for each channel.

[0066] Excitation: The descriptor is then passed through a gating mechanism consisting of two fully connected layers. The first fully connected layer performs dimensionality reduction (e.g., to C / r dimensions, where r is the compression ratio), followed by a ReLU activation function; the second fully connected layer restores the dimension to C, followed by a Sigmoid activation function, thereby generating a weight coefficient between 0 and 1 for each channel.

[0067] Re-weighting: Finally, the learned channel weights are multiplied channel by channel with the output feature map of the feature extraction layer to complete the feature re-weighting.

[0068] By employing the MobileNetV3 convolutional neural network, a balance between high accuracy and lightweight design is achieved: its input layer receives RGB data from preprocessed images, and the feature extraction layer, through depthwise separable convolution and squeeze-excited attention mechanism, enables the model to automatically lock onto color channels sensitive to homogeneous liquid-phase reactions (such as amino acid coupling reactions), significantly improving detection sensitivity in homogeneous, low-contrast environments; the output layer outputs RGB prediction values ​​through fully connected layers, and the overall lightweight structure supports real-time simulation on embedded devices next to the reactor, without the need for expensive server support, reducing hardware costs and deployment barriers.

[0069] The steps for calculating the composite color index are as follows: 1. RGB mean calculation: Within the preprocessed ROI region, calculate the arithmetic mean of the R, G, and B channels of all pixels.

[0070] 2. Generation of Composite Color Index: Generate composite color index based on weighted fusion formula. The weighting coefficients were determined by optimizing a large amount of experimental data through linear regression.

[0071] Specifically, , , The values ​​are 0.45, 0.3, and 0.25 respectively.

[0072] In some implementations, the RGB predicted values ​​and the composite color index are fused using a Kalman filter algorithm to obtain a comprehensive color index, including: Using the Kalman filter's prediction-update mechanism, the RGB predicted values ​​are used as observations, and the composite color index is used as a state estimate, and the combined color index is output.

[0073] By employing a Kalman filter algorithm to achieve complementary heterogeneous data and optimized accuracy, the system dynamically fuses a composite color index based on traditional image processing as a physically interpretable state estimate with RGB predictions based on convolutional neural networks as highly sensitive observations. The system automatically adjusts the weights based on the variance of the two, obtaining an optimal state estimate that is closer to the actual reaction process than a single channel, significantly improving the accuracy of capturing subtle chemical changes in homogeneous systems. Simultaneously, it significantly reduces random errors caused by stirring, bubbles, and light and shadow fluctuations. Through prediction-update logic, it identifies and suppresses transient observation noise, maintaining signal smoothness while producing almost no phase lag. The output composite color index eliminates the sawtooth fluctuations of the original data, providing a high-quality mathematical basis for the rate of change criterion, making endpoint determination more reliable and effectively preventing false triggering. Furthermore, this mechanism establishes a self-correcting closed loop, using physical laws to constrain AI predictions. When CNN predictions deviate significantly from physical common sense, their impact is limited, avoiding the risk of AI model drift and ensuring the safety and prediction stability of the monitoring system under extreme conditions.

[0074] Specifically, the following Kalman filter mathematical model is constructed: 1. Discrete-time state-space equations: State transition equation: ; Observation equation: .

[0075] in, express k The composite color index at any given moment; Representing the state transition matrix, in the application scenario of this embodiment, due to the short sampling interval (e.g., 1 second), it is assumed that the reaction process changes smoothly between adjacent sampling points, therefore, we take... =1; This represents the RGB predicted values ​​(i.e., observed values) extracted by the convolutional neural network. This represents the observation matrix, which takes a value of 1 since the state and observation are scalars of the same scale. Let the process noise be represented, and assume it follows a zero-mean Gaussian distribution with a covariance of . Q ; This represents measurement noise, also assumed to follow a zero-mean Gaussian distribution, with a covariance of... R . Q and R The initial values ​​are obtained based on historical data or experimental calibration.

[0076] 2. Recursive fusion algorithm logic: The algorithm executes the following "prediction-update" loop every frame: (1) Prediction step: using k- Posterior state projection at time 1k Predicted prior state values ​​at time: Prior covariance prediction: .

[0077] (2) Calculate the Kalman gain: In this step, the system dynamically adjusts the weights based on the current reaction stage: during periods of intense reaction, the process noise covariance is increased. Q The weights enable the system to track more sensitively. The variation; during the stationary period or when transient bubble interference is detected, increase the measurement noise covariance. R This makes the system more dependent on historical states. This mechanism ensures that the comprehensive color index curve is both sensitive and smooth, providing a reliable data basis for endpoint determination.

[0078] (3) Update step (fusion output): The prior prediction is corrected using the observations extracted by the CNN to obtain the posterior estimate of the comprehensive color index: Posterior covariance update: .

[0079] Step S108 realizes the intelligent endpoint judgment from "subjective experience observation" to "objective quantitative judgment". Through the change rate convergence logic, it can accurately lock the reaction kinetic equilibrium point, avoiding the problems of over-reaction or incomplete reaction caused by traditional fixed reaction time, significantly improving the purity and yield of peptide synthesis products. At the same time, it supports the dynamic adjustment of monitoring cycle and convergence threshold according to real-time reaction kinetics, adapting to different reaction rates and process conditions, and providing a reliable decision basis for intelligent closed-loop control.

[0080] In some implementations, time-series analysis is performed on the composite color index, including: The composite color index sequence is smoothed by applying a moving average filter. Instantaneous rate of change is calculated based on time difference.

[0081] Specifically, a moving average filter with a sliding window size of 5 is applied to the composite color index sequence to smooth out random fluctuations.

[0082] Instantaneous rate of change calculated based on time difference: .

[0083] In some embodiments, the method further includes: After determining that the homogeneous liquid phase reaction has reached its endpoint, the monitoring time window is extended further; If a significant reverse rebound or drastic fluctuation is detected in the comprehensive color index during this period, a process abnormality warning signal will be output.

[0084] The system determines the reaction endpoint and issues an endpoint signal based on intelligent decision-making through real-time data analysis, but this does not mean the monitoring process immediately terminates. Issuing the endpoint signal and executing the stop-reaction command are designed as two separable stages in the system. After issuing the endpoint signal, a buffer time between issuing the endpoint signal and executing the stop-reaction command can be flexibly set according to specific process requirements. During this period, image acquisition and data monitoring continue. This design, based on three rigorous engineering considerations, achieves an optimal balance between real-time endpoint warning and control reliability, and lays a solid data foundation for continuous process optimization. (1) Dynamic endpoint confirmation and anti-false judgment mechanism. The judgment logic of "maintaining within the preset convergence threshold range within the continuous preset monitoring period" is itself a dynamic confirmation process. Subsequent continuous monitoring constitutes an important confirmation window to ensure that the stable state is continuous, rather than an illusion caused by instantaneous disturbances (such as microbubbles, light intensity flicker), thereby fundamentally avoiding product impurity or yield loss caused by premature termination of the reaction due to false judgment.

[0085] (2) Complete recording of process data. Continuous sampling fully recorded the entire process data from reaction initiation, progress, endpoint determination to the plateau period after the endpoint. This complete dataset, which includes information after the endpoint, has irreplaceable value for process review, in-depth research on reaction mechanisms, and subsequent optimization of intelligent algorithm models, reflecting the completeness and foresight of the system design.

[0086] (3) Empirical verification of system robustness. Continuous monitoring data after the endpoint provides empirical evidence of the system's stability in complex real-world environments. Figure 3 For example, after determining the endpoint (approximately 30 minutes later), the index showed slight fluctuations (rising from ~182 to ~184). Experimental analysis suggested this might be due to complex factors such as the formation of by-products, slight solvent evaporation, or temperature fluctuations. However, crucially, such disturbances did not fundamentally reverse the index trend or trigger the endpoint signal again; the monitored value quickly returned to the plateau after a brief fluctuation. This demonstrates that the system's endpoint determination logic is not a fragile, single-point trigger, but rather a robust design with strong anti-interference capabilities, effectively distinguishing between background noise after the reaction is complete and the actual changes in the reaction process.

[0087] In some implementations, a vision system for real-time monitoring of homogeneous liquid-phase reactions is constructed to achieve efficient, in-situ real-time monitoring and intelligent closed-loop control of amino acid coupling reactions during liquid-phase peptide synthesis. The system achieves direct monitoring of chemical changes in the homogeneous reaction solution through non-contact image acquisition and intelligent analysis.

[0088] Please refer to Figure 2 Specifically, the system includes: Reagent storage unit: Used to store amino acids, coupling reagents and solvents.

[0089] Solution preparation unit: connected to the reagent storage unit, used to prepare reaction solution according to a preset ratio and deliver it to the R1-coupling unit.

[0090] R1 - Coupling Unit: This refers to the container for a homogeneous liquid-phase reaction, which is an amino acid coupling reaction carried out in a homogeneous solution.

[0091] C1 - High-resolution industrial camera: Fixed above the coupling unit, used to acquire video stream data of the reaction solution in real time.

[0092] C2 - Edge Intelligent Processing Unit: Communicates with C1 - High-Resolution Industrial Camera, uses MobileNetV3 convolutional neural network to convert video stream data into RGB predicted values ​​in real time, and fuses them with composite color indices extracted based on traditional image processing methods through Kalman filtering algorithm. Finally, it outputs a more robust and interference-resistant comprehensive color index for endpoint determination.

[0093] C3 - Decision Optimization Center: Communicates with C2 - Edge Intelligent Processing Unit, uses a cloud-deployed intelligent decision model to learn and analyze in-situ characterization data, accurately identifies the time point when amino acids are completely consumed, and outputs optimization and screening parameters to the solution preparation unit to form a feedback control closed loop.

[0094] The system is applied to the closed-loop control of fully automated peptide synthesis. When the C3-decision optimization center determines that the amino acid coupling reaction has reached its endpoint, it automatically sends a control command to the solution preparation unit, triggering the discharge of the reaction solution and the addition of amino acids and coupling reagents in the next cycle, thus realizing fully closed-loop automated synthesis of "characterization-analysis-decision-optimization".

[0095] Compared with traditional monitoring methods, the visual monitoring technology provided in this disclosure exhibits unique advantages in homogeneous liquid-phase synthesis processes: it not only achieves true real-time, in-situ monitoring, avoiding interference with the homogeneous liquid-phase reaction system, but also significantly reduces equipment costs and operational complexity. This method has good adaptability and can be widely applied to various liquid-phase synthesis systems such as batch reactors and continuous flow reactors, providing an effective technical means for the intelligent control of homogeneous liquid-phase reaction processes. For example, in peptide liquid-phase synthesis processes.

[0096] The main advantages of this disclosure include: Real-time and in-situ: It enables true in-situ, online monitoring of homogeneous liquid phase reaction processes without interrupting the reaction or sampling, providing feedback at the second or even millisecond level.

[0097] Non-invasive: The monitoring process does not come into contact with the reaction system and will not introduce contamination or interfere with the reaction process.

[0098] Low cost and easy integration: The hardware used is low cost and the device structure is simple, making it easy to modify and integrate into most existing automated synthesis equipment.

[0099] It lays the foundation for intelligent synthesis: It provides key process awareness means in this field, and the generated real-time data stream can be directly used to drive intelligent decision-making systems. It is the core enabling technology for realizing the full closed-loop automation of "representation-analysis-decision-optimization".

[0100] Example 2 I. Experimental System and Data Sources Please refer to Figure 3-12 The 10 sets of data used in this embodiment are all from the vision system for real-time monitoring of homogeneous liquid phase reaction described in Example 1. They were continuously collected in amino acid coupling reactions under different reaction conditions (including but not limited to temperature, reactant ratio, and mixing intensity parameters (characterized by material flow rate for continuous flow reactors and by stirring speed for batch reactors) to fully verify the universality and reliability of the method provided in this disclosure.

[0101] II. Data Regularity Analysis and Method Validation Analyzing the data from each group reveals the following regularities: Appendix Figure 3-8 During the main reaction period, the composite color index typically starts from a higher initial value (as shown in the attached figure). Figure 3 The value of ~205 in the figure shows a stable decreasing trend, and its rate of change can intuitively reflect the reaction rate. When the reaction approaches the endpoint (as shown in Appendix), the reaction rate decreases. Figure 3 (Approximately 25 minutes later), the rate of change of the comprehensive color index decreased significantly and tended to stabilize, and the index curve entered a plateau period.

[0102] It should be noted that atypical trajectories appeared in some experiments. For example... Figure 9 As shown, the comprehensive color index exhibits a significant decrease followed by a rebound during the middle of the reaction, and then decreases again. This atypical data suggests the possible presence of reversible reactions, side reactions, or significant physicochemical disturbances in the reaction system.

[0103] Initial stage of the reaction (as shown in the attached image) Figure 3 , 5 The first 3-5 minutes of the 6th minute: The overall color index usually starts from a relatively high initial value (as shown in the attached image). Figure 3 ~205 in the middle, attached Figure 5The concentration of reactants (~200) begins to decrease rapidly. This phenomenon is directly related to the highest reactant concentration and fastest consumption rate during the reaction initiation stage, intuitively reflecting the kinetic characteristics of the initial reaction and demonstrating the high sensitivity of this method to reaction initiation signals.

[0104] The main reaction period: The index shows a stable downward trend (as shown in the attached figure). Figure 3 , 5 The rate of change of 7, 8, and 10 can intuitively reflect the reaction rate.

[0105] Reaction endpoint: When the reaction approaches completion, the reactants are almost completely consumed, the system composition stabilizes, and the rate of change of the overall color index decreases significantly and approaches zero. At this point, the exponential curve enters a clear plateau phase (as shown in the attached figure). Figure 3 Approximately 25 minutes later, attached Figure 7 , 10 (Approximately 15 minutes later). The core logic of this method is to capture this plateau characteristic where "the rate of change converges to below the threshold and persists for a period of time".

[0106] Explanation and value of atypical data: Under certain experimental conditions (such as...) Figure 11 , Figure 12 As shown in the figure, the comprehensive color index trajectory exhibits atypical fluctuations or reverse changes. This is not a monitoring defect, but may be a true reflection of reversible reactions, competing side reactions, or complex physicochemical perturbations in the system. The value of this method lies precisely in its ability to record such complex trajectories completely and at high resolution. These atypical data not only demonstrate the robustness of the method in complex scenarios, but their complete spectra also provide crucial information that traditional offline methods cannot obtain for subsequent process diagnostics and reaction mechanism studies. For a detailed analysis of such atypical trajectories and the intelligent response mechanism of this method, please refer to the dedicated explanation in "Part Four" below.

[0107] III. Logic Implementation and Data Support for Intelligent Endpoint Judgment Based on the above data patterns, an endpoint judgment logic is constructed: when the system detects that the rate of change of the comprehensive color index is lower than the preset convergence threshold within a continuous preset monitoring period, the reaction is judged to have reached the endpoint.

[0108] For example, combined with appendix Figure 5 Data shows that after approximately 25 minutes of reaction time, the rate of exponential change has significantly decreased, allowing the system to issue an endpoint signal. (See attached data.) Figure 10 Once the index stabilizes after about 15 minutes of reaction time, the endpoint determination can also be triggered.

[0109] Specifically, the preset convergence threshold, determined based on numerous experiments, is 0.03 units / minute.

[0110] Specifically, continuous monitoring is performed for 3 minutes. This preset monitoring period is based on statistical analysis of a large amount of experimental data, aiming to balance the timeliness and robustness of endpoint determination: a period that is too short is susceptible to random fluctuations leading to misjudgments, while a period that is too long will unnecessarily delay the endpoint signal. Experiments show that a 3-minute window is sufficient to cover most transient disturbances caused by tiny bubbles or light flickering, while also accurately capturing the substantial end of reaction kinetics. This parameter can be configured according to the specific reaction rate and process stability requirements.

[0111] In the above 10 sets of experiments, HPLC sampling was used for verification. When the system determined the endpoint, HPLC showed amino acid residue <1% and product purity >95%, confirming a high degree of consistency between the detected endpoint and the actual endpoint. To visually verify the accuracy of the endpoint determination method, the comprehensive color index (represented as ΔK in the figure) obtained from real-time monitoring in a typical experiment was compared with the purity analysis results of simultaneous offline sampling using high-performance liquid chromatography (HPLC). Figure 13 As shown.

[0112] Depend on Figure 13 It can be seen that in the initial stage of the reaction (approximately 0-4 minutes), the ΔK value decreases rapidly on a logarithmic scale, indicating a large rate of change and corresponding to the rapid phase of the reaction. From approximately 6 minutes onwards, the ΔK curve becomes extremely flat on a logarithmic scale, entering a plateau phase, indicating that its value has stabilized and the instantaneous rate of change remains below the preset threshold. Simultaneously, the HPLC purity reaches approximately 90% at 6 minutes and remains at a high plateau of 90%-95% at all subsequent sampling points (10, 12… minutes).

[0113] The results clearly demonstrate that when this method determines the reaction is complete by monitoring the convergence of the rate of change of the comprehensive color index (ΔK) (manifested as a plateau in logarithmic coordinates), HPLC analysis also confirms that the reaction system has reached and maintained a stable state of high-purity product. This verifies the high consistency between the visual monitoring endpoint judgment logic and the offline chromatographic analysis results.

[0114] The above 10 sets of data were obtained under different reaction conditions. Although the absolute values ​​and curve shapes differed under each condition, they all effectively identified the reaction endpoint, verifying the effectiveness and accuracy of the method and endpoint determination logic provided in this disclosure. Moreover, this method and endpoint determination logic have broad applicability to homogeneous liquid-phase reactions under different synthesis systems and process conditions, rather than being limited to a specific synthesis system.

[0115] IV. Explanation of Atypical Response Trajectories and System Intelligent Response Under certain complex reaction conditions, the overall color index of the system will exhibit the following characteristics: Figure 9 , Figure 10 , Figure 11 , Figure 12 The atypical, drastically fluctuating trajectories shown are a true reflection of the potential for significant reversible reactions, competing side reactions, or strong physicochemical disturbances in the reaction system, demonstrating first and foremost the monitoring system's ability to record complex processes with high fidelity.

[0116] Faced with such trajectories, the "continuous convergence of the rate of change" judgment logic upon which this method is based demonstrates its intelligent decision-making hierarchy and safety boundaries: For cases where there are local trend platforms (such as...) Figure 9 At a specific stage (e.g., approximately 5.0 to 7.5 minutes), although fluctuations exist throughout the process, the statistical characteristics of the rate of change may have met the convergence criteria. Based on this, the system can provide a signal indicating "the main reaction trend has slowed down" or "a plateau has formed," providing crucial information for process judgment.

[0117] For extreme cases with drastic fluctuations throughout the entire process and no clear platform (such as...) Figure 11 , Figure 12 Since the data consistently fails to meet the core criterion of "the rate of change remaining below a threshold within a continuous preset period," the system will not output a definite endpoint time. This is precisely the core manifestation of the reliability and security of this method—it will not force potentially erroneous conclusions when the data is unsupported. Instead, the system can trigger warning signals for "abnormal reaction process" or "endpoint cannot be reliably determined" according to preset rules (see claim 6), prompting operators to intervene in the analysis.

[0118] Therefore, preserving and displaying these atypical data has dual value: first, it empirically demonstrates the working boundary and safety design of this method under complex conditions; second, these high-resolution trajectories provide key information that offline analysis cannot obtain for reaction mechanism research and process fault diagnosis, realizing the sublimation from "endpoint judgment" to "process diagnosis".

[0119] This method is not only an endpoint determination tool, but also a process analysis system with intelligent boundary and deep diagnostic potential.

[0120] V. Appendix Figure 11 Analysis and interpretation of specific data 1. The Deeper Reaction Processes Revealed by the Data: (Appendix) Figure 11 The data trajectory indicates that this specific reaction is not a simple unidirectional process of "reactant consumption → product formation → system stability". The phenomenon that its exponent declined in the middle stage and then rebounded to near the starting point strongly suggests that at least one of the following complex situations may have occurred in the reaction system: ① Reversible reaction or equilibrium shift: After the initial reactants are consumed, the system may reach a certain equilibrium. Subsequently, due to local changes in temperature or concentration or the properties of intermediate products, the reaction direction may be temporarily reversed or new coloring substances may be produced.

[0121] ② Secondary or parallel reactions are triggered: After the main coupling reaction is completed, the remaining reactants, catalysts or intermediates may trigger new side reactions with different optical properties, causing the solution color characteristics to revert to the initial state.

[0122] ③ Significant physicochemical disturbances: such as local overconcentration, uneven temperature, or the introduction of trace impurities, leading to atypical fluctuations in the optical properties of the system.

[0123] 2. Value Demonstration: This atypical trajectory illustrates that the system in Example 1 can not only determine the endpoint of the main reaction, but the complete data curve shape it records can itself serve as a diagnostic marker for abnormal reactions (such as significant side effects or reversible reactions). When the system detects the endpoint, the index exhibits a significant reverse fluctuation (as shown in the attached figure). Figure 11 This can trigger a level 2 warning, prompting operators to check process parameters, thereby upgrading from "monitoring" to "diagnosis".

[0124] For complex and atypical data, the system demonstrates its forward-thinking design and engineering practicality: ① Triggering of the endpoint warning: as shown in the appendix Figure 11 During the process, although the index eventually rebounded, in the middle of the reaction (approximately 5.0 to 7.5 minutes, when the index first showed signs of stabilization or slow change), the system algorithm most likely issued a warning signal that the "main reaction driving period has ended" based on the logic of "the rate of change being continuously below the threshold." This indicates to the operator that the main part of the reaction has been completed.

[0125] ② Comprehensive Monitoring Beyond Endpoint Determination: The system's value lies not only in endpoint determination but also in continuously providing complete, high-temporal-resolution in-situ reaction process maps. (Appendix) Figure 11 This is a perfect demonstration of its value: it didn't stop recording after the suspected plateau (7.5 minutes), but instead captured the entire subsequent unusual rebound. This provides chemists with crucial clues about reaction reversibility or side reactions that are unavailable through traditional methods. It should be noted that, as... Figure 11 , 12The extreme atypical situation illustrated here means that the reaction process itself no longer has a clear, single plateau period. In such cases, relying solely on the "rate of change convergence" judgment logic may not provide a definitive endpoint signal. This precisely reflects the conservatism and safety of this system design: when the reaction process becomes too complex and exceeds the clear boundaries of the preset judgment model, the system will avoid providing a potentially unreliable endpoint determination, and instead trigger an anomaly warning (as described in claim 6) to prompt the operator to intervene and analyze the situation.

[0126] ③ Basis for anomaly diagnosis: This complete trajectory chart itself is a powerful diagnostic tool. It shows that if the index does not stabilize at the plateau period but fluctuates significantly in the opposite direction after the system issues an endpoint warning, it directly indicates that the batch reaction may have experienced an abnormal or complex secondary process, requiring intervention and analysis by process engineers. This is precisely the upgrade from "automatic control" to "intelligent insight".

[0127] Therefore, appendix Figure 11 This not only does not diminish the feasibility of the method provided in this disclosure, but also strongly demonstrates its unique value in revealing the truth of complex reactions and providing deep process insights beyond simple endpoint judgments. It shows that the method provided in this disclosure can: ① Provide timely early warning when the main reaction is completed.

[0128] ② A complete record may reveal the dynamic process of reversible reactions, side reactions, or process anomalies.

[0129] ③ Upgrade the monitoring data from "control signals" to "process diagnostic charts".

[0130] V. Explanation Regarding Fluctuations in Monitoring Data and System Robustness It is important to emphasize that the data fluctuations shown in the attached figures are an inevitable result and a manifestation of the advantages of operating in a real industrial reaction environment. Intrinsic disturbances such as minute temperature gradients, solution convection, and microbubbles present in homogeneous liquid reaction systems can cause high-frequency, low-amplitude random fluctuations in the optical signal. The intelligent endpoint determination logic provided in this disclosure is designed to handle this type of real-world noise. Its core lies in extracting signals from long-term trends, rather than tracking instantaneous fluctuations. For example, in the attached figures… Figure 11 Although fluctuations exist throughout the process, the rate of change systematically enters and remains below the threshold during the middle of the reaction (approximately 5.0 to 7.5 minutes), thus enabling the system to reliably determine the endpoint. This ability to maintain stable judgment amidst noise demonstrates the system's exceptional robustness, providing not just an ideal laboratory curve, but an engineering solution capable of handling real-world complexities.

[0131] Example 3 This disclosure provides a vision device 1000 for real-time monitoring of homogeneous liquid phase reactions.

[0132] The vision device 1000 may include corresponding modules for performing one or more steps in the flowchart of the vision method for real-time monitoring of homogeneous liquid phase reactions described above. Therefore, each or more steps in the flowchart can be performed by a corresponding module, and the vision device 1000 may include one or more of these modules. A module may be one or more hardware modules specifically configured to perform a corresponding step, or implemented by a processor configured to perform a corresponding step, or stored in a readable storage medium for implementation by a processor, or implemented through some combination thereof.

[0133] Specifically, such as Figure 13 As shown, the vision device 1000 includes: The acquisition module 1002 is used to acquire image sequences of the reaction solution in real time using a non-contact vision sensor; The preprocessing module 1004 is used to perform noise filtering and white balance correction on the acquired image to obtain a preprocessed image; The index fusion module 1006 is used to obtain a comprehensive color index based on the preprocessed image using a dual-channel fusion mechanism, including: Extracting RGB predicted values ​​based on convolutional neural networks; Extracting composite color indices based on traditional image processing methods; The RGB predicted values ​​and the composite color index are fused using a Kalman filter algorithm to obtain the comprehensive color index. The monitoring module 1008 is used to perform time-series analysis on the comprehensive color index. When the rate of change of the comprehensive color index remains within a preset convergence threshold range within a continuous preset monitoring period, it is determined that the reaction has reached the endpoint.

[0134] This disclosure also provides an electronic device, including: a memory storing execution instructions; and a processor or other hardware module executing the execution instructions stored in the memory, causing the processor or other hardware module to execute the above-described visual method for real-time monitoring of homogeneous liquid phase reactions.

[0135] This disclosure also provides a readable storage medium storing execution instructions, which, when executed by a processor, are used to implement the above-described visual method for real-time monitoring of homogeneous liquid phase reactions.

[0136] The hardware structure of the vision device 1000, implemented using a processor-based hardware approach, can be implemented using a bus architecture. The bus architecture can include any number of interconnect buses and bridges, depending on the specific application and overall design constraints of the hardware. Bus 1100 connects various circuits including one or more processors 1200, memory 1300, and / or hardware modules. Bus 1100 can also connect various other circuits 1400 such as peripherals, voltage regulators, power management circuits, external antennas, etc.

[0137] Bus 1100 can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Component Architecture (EISA) bus, etc. Bus 1100 can be divided into address bus, data bus, control bus, etc. For ease of representation, only one connection line is used in this diagram, but this does not indicate that there is only one bus or one type of bus.

[0138] Any process or method description in the flowcharts or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of preferred embodiments of this disclosure includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which embodiments of this disclosure pertain. The processor performs the various methods and processes described above. For example, the method embodiments of this disclosure may be implemented as software programs tangibly contained in a machine-readable medium, such as memory. In some embodiments, part or all of the software program may be loaded and / or installed via memory and / or a communication interface. When the software program is loaded into memory and executed by the processor, one or more steps of the methods described above may be performed. Alternatively, in other embodiments, the processor may be configured to perform one of the methods described above by any other suitable means (e.g., by means of firmware).

[0139] The logic and / or steps represented in the flowchart or otherwise described herein may be specifically implemented in any readable storage medium for use by, or in conjunction with, an instruction execution system, apparatus or device (such as a computer-based system, a processor-based system or other system that can fetch and execute instructions from, an instruction execution system, apparatus or device).

[0140] For the purposes of this specification, a "readable storage medium" can be any means capable of containing, storing, communicating, propagating, or transmitting a program for use in or in conjunction with an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of readable storage media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable read-only memory (CDROM). Furthermore, a readable storage medium can even be paper or other suitable media on which a program can be printed, since a program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in memory.

[0141] It should be understood that various parts of this disclosure can be implemented in hardware, software, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0142] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0143] Furthermore, the functional units in the various embodiments of this disclosure can be integrated into a single processing module, or each unit can exist physically separately, or two or more units can be integrated into a single module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a readable storage medium. The storage medium can be a read-only memory, a disk, or an optical disk, etc.

[0144] Those skilled in the art should understand that the above embodiments are merely for illustrating the present disclosure and are not intended to limit the scope of the disclosure. Those skilled in the art can make other changes or modifications based on the above disclosure, and these changes or modifications still fall within the scope of the present disclosure.

Claims

1. A visual method for real-time monitoring of homogeneous liquid-phase reactions, characterized in that, The method includes: Real-time acquisition of image sequences of the reaction solution using a non-contact vision sensor; The acquired images are subjected to noise filtering and white balance correction to obtain preprocessed images; Based on the preprocessed image, a comprehensive color index is obtained using a dual-channel fusion mechanism, including: Extracting RGB predicted values ​​based on convolutional neural networks; Extracting composite color indices based on traditional image processing methods; The RGB predicted values ​​and the composite color index are fused using a Kalman filter algorithm to obtain the comprehensive color index. A time-series analysis is performed on the comprehensive color index. When the rate of change of the comprehensive color index remains within a preset convergence threshold range for a continuous preset monitoring period, the reaction is determined to have reached its endpoint.

2. The method as described in claim 1, characterized in that, The acquired image is subjected to noise filtering and white balance correction to obtain a preprocessed image, including: Define the visual feature analysis area in the acquired images; The visual feature analysis area is subjected to noise filtering and white balance correction to obtain a preprocessed image, wherein the noise filtering adopts a Gaussian filtering algorithm. The white balance correction is based on the gray world assumption. It adjusts the gain of the R and B channels by calculating the mean values ​​of the R, G, and B channels within the visual feature analysis area, so that the mean values ​​of the three channels tend to be consistent.

3. The method as described in claim 1, characterized in that, The convolutional neural network is MobileNetV3, and its structure includes: Input layer: Receives the RGB data of the preprocessed image; Feature extraction layer: consists of multiple Bottleneck modules, each containing depthwise separable convolutions and squeeze-activated attention mechanisms; Output layer: Outputs the RGB predicted values ​​through a fully connected layer.

4. The method as described in claim 1, characterized in that, The RGB predicted values ​​and the composite color index are fused using a Kalman filter algorithm to obtain the comprehensive color index, which includes: Using the Kalman filter's prediction-update mechanism, the RGB predicted values ​​are used as observations, and the composite color index is used as a state estimate, and the combined color index is output.

5. The method as described in claim 1, characterized in that, The method further includes: By using morphological operations and connected component analysis, the interference of bubbles and suspended matter in the reaction system in the preprocessed image is identified and eliminated.

6. The method as described in claim 1, characterized in that, The method further includes: After determining that the homogeneous liquid phase reaction has reached its endpoint, the monitoring time window is extended further; If a significant reverse rebound or drastic fluctuation is detected in the comprehensive color index during this period, a process abnormality warning signal will be output.

7. The method as described in claim 1, characterized in that, The homogeneous liquid phase reaction is an amino acid coupling reaction.

8. A visual device for real-time monitoring of homogeneous liquid phase reactions, characterized in that, The device includes: The acquisition module is used to acquire image sequences of the reaction solution in real time using a non-contact vision sensor; The preprocessing module is used to perform noise filtering and white balance correction on the acquired image to obtain a preprocessed image; The index fusion module is used to obtain a comprehensive color index based on the preprocessed image using a dual-channel fusion mechanism, including: Extracting RGB predicted values ​​based on convolutional neural networks; Extracting composite color indices based on traditional image processing methods; The RGB predicted values ​​and the composite color index are fused using a Kalman filter algorithm to obtain the comprehensive color index. The monitoring module is used to perform time-series analysis on the comprehensive color index. When the rate of change of the comprehensive color index remains within a preset convergence threshold range for a continuous preset monitoring period, it is determined that the reaction has reached the endpoint.

9. An electronic device, characterized in that, include: The memory stores execution instructions; as well as A processor that executes execution instructions stored in the memory, causing the processor to perform the method according to any one of claims 1-7.

10. A readable storage medium, characterized in that, The readable storage medium stores execution instructions, which, when executed by a processor, are used to implement the method described in any one of claims 1-7.

Images