Physical experiment data rapid analysis method and system based on image processing

Abstract

The invention discloses a physical experiment data rapid analysis method and system based on image processing, and relates to the technical field of image data analysis. Comprising the following steps: processing a continuously collected original experiment image sequence, and separating a foreground region representing an experiment object; automatically identifying a physical experiment type based on the image features of the region, and synchronously obtaining a unified configuration set bound with a feature extraction strategy and a physical constraint rule; performing feature extraction in parallel according to the configuration set to generate initial time sequence data, and verifying the data based on a physical constraint rule to generate a physical consistency residual error; querying a predefined mapping relation according to a residual error mode, dynamically adjusting a feature extraction parameter or foreground region positioning, enabling the residual error to meet a preset convergence condition through iteration, and generating corrected time series data; and finally, outputting a physical quantity analysis result. According to the invention, the physical experiment data processing efficiency and the physical reliability of the result are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image data analysis, in particular to a physical experiment data rapid analysis method and system based on image processing. BACKGROUND

[0002] The physical experiment data analysis technology based on image processing is an important tool for modern teaching and scientific research, aiming to automatically obtain physical quantity change data through non-contact measurement, and to improve experimental efficiency and data reliability. In recent years, with the progress of computer vision algorithms and computing hardware performance, this technology field is developing towards automation and intelligence.

[0003] The prior art solutions are mostly focused on improving image processing algorithms (such as optimizing optical flow method to track motion trajectory, or using more accurate edge detection operator to locate fringe spacing) to improve the measurement accuracy of specific physical experiments. These methods can effectively extract key features and generate data curves when targeting a certain type of fixed experimental scene, and have certain speciality and accuracy. However, the types of experiments covered in physical teaching and scientific research are diverse, from kinematics, mechanics to optics, electromagnetism, their observation targets, motion characteristics and measured physical quantities are quite different. When facing new types of experiments, technical personnel often need to manually reselect algorithms, configure parameters, and even adjust the entire processing pipeline. This high dependence and cumbersome configuration process makes it difficult for the existing technology to quickly adapt to diversified experimental scenarios, and the degree of automation and universality is insufficient, which seriously restricts the efficiency of its large-scale application in open experiment platforms.

[0004] Therefore, there is an urgent need for a technical means that can solve the problem of the existing technology being fixed in speciality and lacking cross-experimental universality and rapid self-adaptation capability. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides a physical experiment data rapid analysis method and system based on image processing.

[0006] In order to achieve the above-mentioned purpose, the technical scheme of the present application is as follows:

[0007] In a first aspect, the present application discloses a physical experiment data rapid analysis method based on image processing, comprising the following steps:

[0008] Processing the continuously collected original experiment image sequence to separate at least one foreground region representing the experimental object from the dynamic background;

[0009] Based on the image features of the foreground region, automatically identifying the experimental type of the physical experiment;

[0010] According to the identified experiment type, a unified configuration set is synchronously acquired; the unified configuration set is associated with a feature extraction strategy, a physical constraint rule and initialization parameters;

[0011] According to the unified configuration set, the following is performed in parallel:

[0012] According to the feature extraction strategy, feature extraction is performed on a foreground image sequence of the foreground region to generate initial time series data;

[0013] According to the physical constraint rule, the initial time series data is verified to generate a physical consistency residual;

[0014] Based on a mode of the physical consistency residual, a predefined residual parameter mapping relationship is queried, algorithm parameters of the feature extraction or positioning of the foreground region are dynamically adjusted, and iterative execution is performed until the physical consistency residual meets a preset convergence condition, to generate corrected time series data;

[0015] Based on the corrected time series data, a final physical quantity analysis result is output.

[0016] In a second aspect, the present application discloses a physical experiment data fast analysis system based on image processing, which uses the above-mentioned physical experiment data fast analysis method based on image processing, and comprises:

[0017] A preprocessing module is configured to process a continuously collected original experiment image sequence, and separate at least one foreground region representing an experiment object from a dynamic background of the original experiment image sequence;

[0018] An identification module is configured to automatically identify an experiment type of the physical experiment based on image features of the foreground region;

[0019] A configuration module is configured to acquire a unified configuration set synchronously according to the identified experiment type; the unified configuration set is associated with a feature extraction strategy, a physical constraint rule and initialization parameters;

[0020] A parallel processing module is configured to perform the following in parallel according to the unified configuration set:

[0021] According to the feature extraction strategy, feature extraction is performed on a foreground image sequence of the foreground region to generate initial time series data;

[0022] According to the physical constraint rule, the initial time series data is verified to generate a physical consistency residual;

[0023] The check module is configured to query a predefined residual parameter mapping relationship based on a mode of the physical consistency residual, dynamically adjust algorithm parameters of the feature extraction or positioning of the foreground region, and iteratively execute until the physical consistency residual meets a preset convergence condition, thereby generating corrected time series data.

[0024] The output module is configured to output a final physical quantity analysis result based on the corrected time series data.

[0025] Compared with the prior art, the present application has the following advantages:

[0026] 1. The present application overcomes the dependence of the prior art on specific experimental scenarios, enabling the same analysis system to be flexibly and accurately applied to various types of physical experiments, from simple linear motion to complex vibration and collision processes, thereby improving the universality and automation level of the technology.

[0027] 2. The present application can effectively identify and correct data anomalies caused by image noise, target occlusion, or temporary tracking failures, so that the final generated time series data not only comes from image measurement, but also strictly follows the basic physical laws, thereby improving the physical authenticity and reliability of the final analysis results at the root.

[0028] 3. The present application not only completes the measurement of physical quantities, but also verifies and enhances the physical reasonableness of the measurement results, thereby reducing the workload of subsequent manual data cleaning and correction, and providing solid technical support for the digital and intelligent transformation of scientific experiments and teaching practice. BRIEF DESCRIPTION OF DRAWINGS

[0029] The disclosure of the present application will be described with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of the present application. In the drawings, the same reference numerals are used to refer to the same components. Among them:

[0030] Figure 1 is a step flowchart of the present application;

[0031] Figure 2 is a working principle diagram of the present application;

[0032] Figure 3 is a flowchart of automatic identification of experimental types of the present application;

[0033] Figure 4 is a system module connection diagram of the present application;

[0034] Figure 5 is a system module flowchart of the present application. DETAILED DESCRIPTION

[0035] It is easy to understand that, according to the technical solutions of the present application, those skilled in the art can propose a plurality of structure modes and implementation modes which can be replaced with each other without changing the essential spirit of the present application. Therefore, the following specific embodiments and drawings are only exemplary descriptions of the technical solutions of the present application, and should not be regarded as the whole or as the limitation or restriction of the technical solutions of the present application.

[0036] SUMMARY

[0037] In the prior art, physical experiment data analysis based on image processing relies on preset fixed algorithm process, and it is difficult to balance generality and accuracy. When facing different types of physical experiments, the traditional method needs to manually configure parameters and algorithms, and the measured data lacks a physical reasonable verification link. The existing scheme cannot adaptively adjust the analysis strategy according to the experiment content, especially when the measured data is abnormal, it lacks a correction mechanism based on physical laws, and it is difficult to meet the strict requirements of scientific experiments on data reliability.

[0038] In order to solve the above problems, it is found that there is an inherent correlation between the type of experiment and the applicable physical constraint rule, and the analysis process is intelligentized by establishing a type recognition and strategy loading mechanism. Secondly, the foreground area of different types of experiments has distinguishable characteristics in geometric structure and motion mode, and the physical laws followed by various experiments can provide reliable basis for data checking. Therefore, the idea of dynamically loading feature extraction strategy and physical constraint rule according to the recognition result is proposed. Further through experiment verification, the physical constraint rule is introduced into the data processing flow in the form of residual calculation, forming a closed-loop system of data checking and correction.

[0039] After introducing the basic idea of the present application, the embodiments of the present application will be specifically introduced with reference to the drawings.

[0040] Embodiment one:

[0041] As shown in Figure 1 A physical experiment data rapid analysis method based on image processing, comprising the following steps:

[0042] Processing the continuously collected original experiment image sequence to separate at least one foreground area representing the experiment object from the dynamic background;

[0043] Based on the image features of the foreground area, automatically identifying the experiment type of the physical experiment;

[0044] According to the identified experiment type, synchronously acquiring a unified configuration set; the unified configuration set is associated with a feature extraction strategy, a physical constraint rule and an initialization parameter;

[0045] According to the unified configuration set, performing in parallel:

[0046] According to the feature extraction strategy, the foreground image sequence of the foreground region is subjected to feature extraction, and initial time series data is generated;

[0047] According to the physical constraint rule, the initial time series data is verified, and a physical consistency residual is generated;

[0048] Based on the mode of the physical consistency residual, the pre-defined residual parameter mapping relationship is queried, the algorithm parameters of feature extraction or the positioning of the foreground region are dynamically adjusted, and iterative execution is performed until the physical consistency residual meets the preset convergence condition, and the corrected time series data is generated;

[0049] Based on the corrected time series data, the final physical quantity analysis result is output.

[0050] As shown in Figure 2 The working principle of the present application is as follows: the system first processes the continuously collected original experimental image sequence. Through dynamic background modeling technology, each frame of image is compared with the continuously updated background model, so as to accurately separate the foreground region representing the core experimental object (such as the interference fringes in the "Young's double-slit experiment", the suspension disc in the "three-line pendulum measurement of rotational inertia", the optical lever scale image in the "tensile method for measuring the Young's modulus of steel wire", etc.) from the dynamically changing experimental scene.

[0051] Subsequently, the system automatically identifies the type of physical experiment based on the image features of the separated foreground region. Specifically, by analyzing the geometric structure features (such as the specific configuration of the cross-shaped wire in "adjustment of spectrometer"), motion pattern features (such as the relative motion of the sound source and receiver in "Doppler effect test experiment") and spatial distribution features (such as the equipotential point distribution formed by the probe in "simulation method for measuring electrostatic field"), a multi-dimensional feature description vector is formed, and is matched with a pre-stored template library containing various experiments listed in the picture, so as to determine the specific experimental type. Once the identification is successful, the system synchronously acquires a pre-defined unified configuration set according to the identified experimental type. The unified configuration set is associated with the feature extraction strategy, the physical constraint rule and the algorithm initialization parameter strictly corresponding to this experimental type. The initialization parameter is the initial state or physical constant required for executing these rules, such as initial temperature, gravitational acceleration, medium density, etc. For example, for "magnetic field measurement of Helmholtz coil", the configuration set will bind the feature extraction strategy based on the sequence tracking of the deflection angle of the magnetic needle and the physical constraint rule based on the theoretical distribution model of the magnetic field; for "measurement of specific heat capacity of metal", it may bind the strategy based on the reading of the liquid column height of the thermometer and the physical constraint based on the cooling law.

[0052] The system executes two core processes in parallel according to the unified configuration set:

[0053] First, the image sequence of the foreground region is processed according to the binding feature extraction strategy, key parameters are extracted, and initial time series data (such as magnetic field intensity-position sequence or temperature-time sequence) are generated. The feature extraction strategy is a set of operation procedures and algorithms predefined for a specific experiment type to quantify the state of the experimental object from the image sequence. For example, for the "three-wire pendulum measurement of rotational inertia" experiment, the strategy will specifically specify the calculation of the torsion angle by tracking a specific marker point on the suspension disc; and for the "Young's double-slit experiment", the strategy will guide the system to analyze the spacing of the interference fringes to invert the wavelength of the light wave. Each strategy encapsulates a complete set of logic from image preprocessing, target positioning to parameter calculation.

[0054] Second, the initial time series data is verified for reasonableness according to the binding physical constraint rule, and the physical consistency residual C is generated by calculating the difference between the theoretical predicted value and the actual measured value. The physical constraint rule is a mathematical expression of the physical law or conservation quantity embedded in the system corresponding to the experiment type, used to check the reasonableness of the measured data. For example, for the "measurement of specific heat capacity of metal", the rule can be that the cooling process should conform to the differential form of Newton's cooling law.

[0055] Based on the pattern of the physical consistency residual C (such as the amplitude of the residual, the trend of change), the pre-defined residual parameter mapping relationship is queried to dynamically adjust the algorithm parameters of feature extraction (such as the sensitivity threshold of the tracking algorithm) or to fine-tune the positioning of the foreground region. This process will be iteratively executed until the physical consistency residual C meets the pre-set convergence condition, thereby finally generating a corrected time series data with higher data quality and more in line with physical laws.

[0056] The residual parameter mapping relationship is a pre-defined knowledge base or function that describes the correspondence between the "pattern of physical consistency residual" and the "algorithm parameter to be adjusted". It guides the system on how to dynamically adjust the feature extraction process according to the current residual characteristics. The mapping relationship can be a simple lookup table or a complex function. For example, when the physical consistency residual C exhibits high-frequency oscillation, the mapping relationship may indicate that the system increase the smoothing coefficient of the tracking algorithm; when the physical consistency residual C exhibits a persistent positive offset, the mapping relationship may indicate that the system fine-tune the scale factor in the calibration parameters. The pre-set convergence condition is used to determine whether the physical consistency residual C has been reduced to an acceptable level. Common convergence conditions include: absolute convergence, i.e. the absolute value of the physical consistency residual C is less than a pre-set first threshold (e.g. 0.01 physical units); relative convergence, i.e. the ratio of the residual norm of the current iteration to the initial residual norm is less than a pre-set second threshold (e.g. 1%); or the maximum number of iterations to prevent infinite loops.

[0057] Finally, based on the corrected time-series data, the system outputs the final physical quantity analysis results. These results may include calculated physical constants (such as grating constant and sound velocity), material properties (such as Young's modulus and viscosity coefficient), and fitting curves and reports characterizing the physical processes.

[0058] This application deeply integrates image processing and physical modeling, realizing end-to-end automated analysis from raw images to high-confidence physical quantity results, effectively improving the intelligence level of physical experimental data processing and the reliability of results.

[0059] This application further proposes that the specific steps for separating at least one foreground region representing the experimental object from the dynamic background of the original experimental image sequence include:

[0060] During system initialization, a segment of initial multi-frame images (e.g., 30 consecutive frames) without rapidly moving targets from the original experimental image sequence is acquired. The running average and variance at the pixel or block level are calculated to establish an initial background model. Specifically, the system establishes a background model by calculating the running average and standard deviation of the intensity values ​​at each pixel location or preset image block (e.g., an 8×8 pixel region) across all initial frames. This background model models each background point as a Gaussian distribution, where the mean represents the typical appearance of the point, and the standard deviation characterizes the inherent fluctuation range of the point due to noise or minor perturbations.

[0061] After the initial model is established, the system enters the real-time processing stage. For each new input frame of the image, the foreground is segmented by comparing the current intensity value of each pixel with the Gaussian distribution at the corresponding location in the background model. Specifically, if the current pixel intensity value... vs. background mean The absolute difference exceeds its standard deviation K times (i.e.) If the sensitivity coefficient K is high, then the pixel is considered to belong to the foreground region. The sensitivity coefficient K is a key parameter, and its empirical value is usually set between 2.5 and 3.5.

[0062] In subsequent processing, the update rate of the background model is adaptively adjusted based on the scene motion magnitude represented by the foreground region separation results. Scene Motion Magnitude (SMQ) is defined as the ratio of the total number of foreground pixels identified in the current frame to the total number of pixels in the image; it quantifies the drasticness of scene changes. Based on SMQ, the system dynamically calculates the model update rate α, which is calculated using the following formula:

[0063]

[0064] In this formula, and are preset minimum and maximum update rates, for example, 0.001 and 0.05, respectively;

[0065] is a decay constant, controlling the influence strength of SMQ on the update rate a, which can be set as 10 in general. When the scene is moving violently (SMQ value is large), the update rate a is automatically reduced to close to , effectively preventing the moving target from being "absorbed" into the background; when the scene is static or stable (SMQ value is small), the update rate a is increased to close to , enabling the background model to quickly adapt to slow environmental changes such as illumination changes.

[0066] The mean value μ and the standard deviation σ of the background model are then updated according to the calculated update rate a:

[0067]

[0068]

[0069] wherein, respectively represent the intensity mean value of a certain pixel or image block in the background model at time t and t+1, representing the appearance characteristics of the background at that position. During model initialization, are calculated from the initial image sequence.

[0070] respectively represent the intensity standard deviation of a certain pixel or image block in the background model at time t and t+1. They represent the noise level or inherent fluctuation range of the background at that position. During model initialization, they are calculated from the initial image sequence.

[0071] represents the actual intensity observation value of the new input image frame at the corresponding pixel or image block at time t.

[0072] The background model is not fixed, but is constantly updated with a controllable update rate a, according to the latest image observation data , to update the mean value μ and the intensity standard deviation σ of the model by weighted average. This enables the model to maintain stability for long-term static background and adapt to slow changes in the environment. After morphological post-processing (such as opening operation of erosion followed by dilation) to eliminate scattered noise and fill in holes, all pixels determined as foreground are finally aggregated to form a connected foreground region representing the physical experimental object.

[0073] Through the technical solution, the application can robustly and accurately separate the experimental target from the video stream containing complex environmental changes. The application effectively overcomes the defects of the traditional fixed update rate model in dealing with light mutations and continuous motion, lays a reliable foundation for subsequent experimental type intelligent recognition and high-precision physical feature extraction, and improves the applicability and accuracy of the entire analysis system in real experimental environments.

[0074] The application further proposes that the specific steps of automatically identifying the experimental type of the physical experiment based on the image features of the foreground region include:

[0075] For depth feature extraction of the foreground region, the system crops the image block containing the foreground region from each frame, and extracts image block data containing spatial distribution, motion trajectory and texture features from the foreground region: spatial distribution features (such as the relative position of the foreground region in the picture, the topological relationship between multiple objects), motion trajectory features (by tracking the foreground region in consecutive frames, the statistical quantities of displacement, velocity and acceleration are obtained) and texture features (such as the surface roughness of the foreground region, periodic patterns, which are particularly important when analyzing interference fringes or gratings).

[0076] To construct a robust feature representation, the system converts the above-mentioned original features into more discriminative descriptors. Specifically, the geometric invariant features and spatio-temporal motion features of the image block data are calculated. Geometric invariant features (such as scale invariant feature transform SIFT or direction gradient histogram HOG variants) are used to describe the shape and structure of the foreground object, ensuring that the recognition is not sensitive to the scale and rotation changes of the object. At the same time, spatio-temporal motion features (such as average motion direction and motion consistency based on dense optical flow calculation) are extracted to quantify the motion pattern of the object. All these features are spliced to generate a comprehensive multi-dimensional feature vector for experimental type recognition, thereby fully encoding the visual performance of the experiment.

[0077] Through clustering analysis and template matching of the multi-dimensional feature vector, the experimental type identification and recognition confidence are output. Specifically, the determination of the experimental type is completed through a pre-constructed classification model. The training process of the model is as follows:

[0078] In the offline stage, 10,000 video clips (10-30 seconds each) covering 20 typical physical experiment types are collected for feature vectors, and manually labeled with real experiment type labels. Using these labeled data, a support vector machine (SVM) classifier is trained. The core of SVM is to find a hyperplane that can optimally separate feature vectors of different classes in the feature space. The training conditions include using a radial basis function (RBF) as a kernel function to handle non-linear separable problems, optimizing hyperparameters through grid search (penalty coefficient V∈[0.1,1,10,100], kernel function coefficient γ∈[0.001,0.01,0.1,1]) and 5-fold cross-validation, and finally selecting the parameter combination with the highest classification accuracy (such as V=10, γ=0.1). The training process iterates 500-1000 times until the hyperplane converges, in order to achieve high accuracy on the training set while maintaining good generalization ability.

[0079] In the online recognition stage, the system inputs the real-time extracted multi-dimensional feature vector into the trained SVM model. The model not only outputs the most likely experiment type label (for example, "judged as'standing wave experiment on string'"), but also outputs a recognition confidence, which is usually derived from the functional gap between the sample feature vector and the SVM classification hyperplane. To ensure the reliability of the recognition result, the system sets a confidence threshold (for example, 0.85). The setting of the confidence threshold is based on the performance evaluation of the model on the reserved validation set, and the threshold point that can achieve the best balance between accuracy and recall rate is selected. When the highest confidence output is higher than the confidence threshold, the result is valid; if it is lower than the confidence threshold, the system can determine it as "unknown experiment" or trigger the re-recognition mechanism.

[0080] Through the above technical solution, the present application realizes fast, automatic and accurate recognition of various physical experiment types, overcomes the limitations of traditional methods relying on manual preset or parameter adjustment, and improves the autonomy and adaptability of the system in the face of diversified experimental scenarios.

[0081] The present application further proposes that the specific steps of feature extraction according to the feature extraction strategy include:

[0082] Temporal difference analysis is performed on the image sequence of the foreground region to obtain inter-frame motion information;

[0083] Spatial gradient analysis is performed on the single frame image of the foreground region to obtain edge contour information;

[0084] Specifically, two analyses are performed in parallel: one is temporal difference analysis, in which the system calculates the pixel-level difference of the foreground region between consecutive frames. By comparing the t-th frame with the t-1-th (or t-n-th) frame, a binary motion mask can be obtained, in which the highlighted regions represent the parts that have shifted between frames. Connected component analysis and centroid calculation on these regions can yield the inter-frame motion information of the experimental object, including the instantaneous velocity vector and the motion direction. The other is spatial gradient analysis, in which the system uses Sobel or Canny edge detection operators to calculate the image gradient of the foreground region in each frame, thus accurately extracting the edge contour information of the experimental object.

[0085] According to the target motion speed reflected by the inter-frame motion information, the fusion weight of the inter-frame motion information and the edge contour information is dynamically adjusted, and the experimental object is located and tracked based on the fused information, and the position or shape parameters reflecting the state of the experimental object based on pixel coordinates are output as key parameters.

[0086] The system defines a fusion weight factor ω, the value of which is dynamically adjusted according to the target motion speed v calculated by the temporal difference analysis. The adjustment strategy is as follows: when the target motion speed v is high (for example, v > 150 pixels per second), it means that the target displacement is large, and the contour may be blurred between frames, at which time the system gives higher weight to the temporal motion information (for example, ω is close to 0.8) to ensure the continuity and real-time of tracking. When the target motion speed v is low (for example, v < 20 pixels per second) or close to static, the spatial contour information is more stable and accurate, at which time the system gives higher weight to the edge contour information (for example, ω is close to 0.2). The setting of the speed threshold , is based on the statistical analysis of the speed distribution of the target in various typical experimental scenarios (such as fast and slow swinging of a pendulum, starting and stopping of a slider), and is an empirical value determined to cover most motion states. , The final fused information can be represented as:

[0087]

[0088]

[0089] wherein, and are the normalized motion mask information and edge contour information, respectively.

[0090] ​​​Based on the fused information F, the system adopts a tracking algorithm such as correlation filtering or mean shift to continuously and stably locate the experimental object in the image sequence. The object boundary box center coordinates, angle or specific contour point coordinates output by the tracker are taken as key parameters (such as position, inclination) reflecting the state of the experimental object and are output in real time to generate initial time series data.

[0091] The feature extraction strategy of the present application skillfully balances the robustness of the tracking algorithm under high-speed motion and the accuracy under low-speed or stationary state. It effectively overcomes the inherent defects of simply relying on the differential method to easily lose tracking when the target is stationary or simply relying on the contour method to be inaccurate in positioning due to motion blur when the target moves quickly, so that it can provide continuous, stable and accurate key parameter sequences for subsequent physical quantity calculation under various complex experimental conditions, laying a solid foundation for high-reliability data analysis.

[0092] The present application further proposes that, before generating the initial time series data, a calibration and unit conversion step is also included, specifically comprising:

[0093] Obtain pre-stored or dynamically detected scale information from the original experimental image sequence. There are two sources of scale information: one is the pre-stored calibration method, that is, before the experiment starts, a reference object with a known physical size (for example, a calibration plate or a specific marker with a length L (such as 10.0 cm)) is placed in the measurement plane of the experimental scene and its image is collected. The system locates the reference object through image recognition technology (such as corner detection) and calculates its pixel length P in the image. Thus, the pixel-physical unit conversion scaling factor S = L / P (unit: mm / pixel) can be calculated. This scaling factor S will be stored for subsequent conversion of all frames. The second is the dynamic detection method, which is suitable for using known size static objects (such as the scale of experimental instruments or specific parts of clamps) in the scene as natural scales. The system dynamically calculates and updates the scaling factor S by real-time detection and recognition of the features of these objects in the image to adapt to possible slight movement or zoom of the camera.

[0094] Using the scale information, after obtaining a reliable conversion scaling factor S, the pixel-based position or shape parameters are converted into physical quantity measurement values with physical units. For example, a centroid coordinate is converted from a pixel position to a physical position ; an angle calculated based on pixel coordinates (such as the angle corresponding to the steel wire elongation calculated by detecting the deflection of the optical lever) is converted into a physical angle in degrees according to its geometric relationship combined with the scaling factor S.

[0095] After this conversion, all key parameters are given real physical units. The system then uses these physical quantity measurements (rather than the original pixel values) to construct the initial time series data. For example, the generated data sequence is "time t (s) - displacement x (mm)" or "time t (s) - angle Θ (°)", which enables subsequent data processing and physical law verification to be performed in real physical dimensions.

[0096] By introducing a flexible and reliable calibration mechanism and unit conversion process, the present application fundamentally guarantees the physical reality and dimensional consistency of the extracted data, enabling subsequent data checking and correction based on physical constraint rules to have a clear physical meaning and comparability, effectively avoiding the problem of distorted physical laws and calculation results that cannot be connected with the actual physical world due to unknown scales.

[0097] The present application further proposes that after generating the physical consistency residual C through physical constraint rules, the algorithm parameters of the feature extraction are dynamically adjusted, specifically including one or more of the following ways:

[0098] In the foreground region of the time frame corresponding to the abnormal data point, sub-pixel level positioning fitting calculation is performed again. When the physical consistency residual C exhibits isolated and large amplitude spikes, the system determines that this is a transient positioning anomaly. At this time, the system traces back to the time frame corresponding to the abnormal data point and starts sub-pixel level positioning fitting calculation in the foreground region it has segmented. Specifically, the system abandons simple pixel level centroid calculation and instead uses algorithms based on gray weight or surface fitting (such as two-dimensional Gaussian surface fitting) to reposition the center of gravity or edge of the target. For example, for a circular experimental object, by fitting the gray distribution of its edge, the positioning accuracy can be improved from 1 pixel to 0.1 pixel or even higher, effectively correcting the large errors of a single point caused by partial occlusion, noise or motion blur.

[0099] Using the values of adjacent high-confidence data points in the initial time series data, the abnormal data segment is interpolated and replaced. When the physical consistency residual C exhibits small continuous abnormalities or data loss, the system enables an interpolation and replacement mechanism based on high-confidence neighborhood data. The system first assigns a confidence score to each point in the initial time series data according to the size of the physical consistency residual C. The smaller the physical consistency residual C, the higher the confidence. Then, for the short period of data judged to be abnormal, the system uses its adjacent high-confidence data points to reconstruct the data in that period using appropriate interpolation algorithms (such as cubic spline interpolation to maintain curve smoothness). The size of the neighborhood window used for interpolation can be adaptively adjusted according to the length of the abnormal segment, for example, for an abnormal segment of length N frames, the high-confidence data of M frames (M ≥ N) before and after it is usually selected as the basis for interpolation to ensure the accuracy of the reconstruction.

[0100] Further, after generating the physical consistency residuals C by the physical constraint rule, a positioning of the foreground region is further modified dynamically, specifically: based on the systematic bias reflected by the physical consistency residuals C, an online micro-calibration process is triggered to re-estimate the mapping parameter from pixel to physical unit. When the physical consistency residuals C presents a persistent, one-way systematic bias (for example, a scale change caused by a slight displacement of the camera), the system will trigger the online micro-calibration process. This process is based on a core observation: the physical constraint rule (such as the law of conservation of energy, the equation of uniformly accelerated motion) holds in the real world, so the systematic bias must be due to the misalignment of the measurement model parameters (mainly the pixel-physical unit conversion scale factor S, which establishes the spatial scale mapping relationship between the image pixel coordinate system and the real physical world coordinate system, and is the core parameter for converting image measurement values to real physical quantities.). The system collects a segment of high-confidence time-series data, substitutes it into the physical model corresponding to the experimental type (for example, substitutes the displacement-time data into the uniformly accelerated motion formula), and uses optimization algorithms such as least squares to back-calculate a new scale factor S' that best satisfies the physical model under the current data. The new scale factor S' generates new parameters that will replace the old parameters for subsequent conversion of image coordinates to physical quantities, thereby fundamentally correcting the systematic bias.

[0101] The present application constructs a powerful data self-healing system. It can intelligently distinguish and effectively handle different types of defects such as transient noise, local data loss, and systematic model bias, not only filling in missing data, but also performing online calibration at the measurement principle level, not only smoothing on the surface, but also being highly credible in physical essence, improving the robustness of the entire analysis system in a non-ideal experimental environment and the scientific value of the output results.

[0102] As shown in Figure 3 The present application further proposes specific steps for automatically identifying the experimental type of a physical experiment, including:

[0103] The image features are constructed into a multi-dimensional feature vector containing geometric structure features, motion pattern features, and optical features. The multi-dimensional feature vector integrates three types of key information: geometric structure features, such as the Hu moment, aspect ratio, and contour complexity of the foreground region, for describing the static configuration of the experimental device; motion pattern features, such as the curvature of the target trajectory, the statistical properties (mean, variance) of the velocity-time curve, and the periodicity of the motion, for characterizing the dynamics of the experimental object; and optical features, such as the overall brightness, contrast, or response in a specific color channel of the foreground region, which is crucial for distinguishing experiments that rely on optical phenomena such as interference and diffraction. All features are normalized before concatenation to eliminate the effects of dimension.

[0104] The similarity score of the multi-dimensional feature vector and each template vector in the pre-stored experimental template library is calculated; a set of initial similarity scores is obtained by calculating the cosine similarity between the multi-dimensional feature vector and each template vector in the pre-stored experimental template library. The construction of the template library is derived from offline training: for each type of experiment to be identified (such as "Young's double-slit experiment", "three-line pendulum measurement of rotational inertia", etc.), a large number of sample videos under different conditions are collected, and their feature vectors are extracted and their average feature vector (or multiple typical templates obtained by clustering) is calculated as the template of this type of experiment.

[0105] The similarity scores are logically verified based on predefined experimental type discrimination rules; these experimental type discrimination rules encode domain knowledge, for example: "if identified as'simple pendulum experiment', the main direction of its motion trajectory should be approximately perpendicular to the direction of gravity"; "if identified as 'light equal-thickness interference', the foreground area should present parallel stripes of light and dark". Only when the highest scoring candidate type passes the verification of these logical rules, the identification result is considered to be initially valid.

[0106] Finally, the system outputs the experimental type identifier and a quantitative type matching confidence. The confidence is determined by two factors: one is the similarity score with the best matching template, and the other is the degree of passing the logical rule verification. When the similarity score is higher than the type recognition threshold and passes the logical verification, the corresponding experimental type identifier and type matching confidence are output;

[0107] Among them, the type matching confidence serves as a weight factor and participates in the recommendation calculation of the unified configuration set. For example, when the type matching confidence is very high (such as higher than 0.95), the system will completely load the "standard" configuration set preset for this type; when the type matching confidence is medium, the system may integrate some more universal or robust alternative parameters based on the standard configuration to form a "weighted" configuration set to cope with the uncertainty in identification.

[0108] Through the above technical solutions, the application realizes high-precision and high-reliability identification of physical experiment types, so that the entire system can exhibit excellent adaptability and decision flexibility when facing borderline fuzzy or poor experimental scenarios.

[0109] The application further proposes a mode based on physical consistency residual C, which further includes establishing an anomaly classification model based on physical constraint rules, specifically including:

[0110] Intelligent diagnosis and classification of data anomalies, according to the spatial and temporal distribution characteristics and amplitude of the physical consistency residual C, the identified abnormal data is divided into transient noise type, system deviation type and tracking loss type. Among them, the transient noise type is manifested as isolated in the time series, amplitude mutation of the peak, the residual derivative is maximum, which is usually caused by random noise, temporary occlusion or light reflection in a single image. The system deviation type is manifested as the physical consistency residual C has a persistent non-zero mean, that is, the overall data deviates from the predicted curve of the physical model, which is often caused by the initial error of the camera pose or the calibration parameter (such as the pixel-physical unit conversion scale factor S). The characteristics of the tracking loss type are that the physical consistency residual C increases sharply or even the data is interrupted in a period of time, which is usually caused by fast target motion leading to motion blur, being completely occluded by other objects or moving out of the field of view.

[0111] For the transient noise type, an adaptive filtering algorithm based on neighborhood confidence is used for smoothing. Specifically, it assigns a confidence to each data point in the sequence, and the neighborhood points with smaller physical consistency residual C have higher weights. The filtering output is no longer a simple moving average, but a weighted average of neighborhood points, where the weight is proportional to the confidence of each point. This ensures that while smoothing noise, the true and highly reliable physical signal characteristics can be maximally retained.

[0112] For the system deviation type, the parameter self-calibration process is started, and the mapping relationship between pixels and physical units is recalculated. The parameter self-calibration process uses the time series data in a window period (for example, the last 100 frames) to substitute into the core physical equation corresponding to the current experimental type (for example, for uniform acceleration straight line motion, the formula ( ) is used). Through optimization algorithms such as least squares method, the new pixel-physical unit mapping parameters (such as the new scale factor S') that best satisfy the physical equation are solved. This process essentially uses the physical law itself as an absolute reference to re-calibrate the measurement system online, thereby fundamentally eliminating the systematic error source.

[0113] For the tracking loss type, a foreground region relocation mechanism based on motion trajectory prediction is started in the subsequent image frames. The foreground region relocation mechanism uses the motion state (such as position, velocity, acceleration) of the target in a period of time before tracking loss to predict the most likely position and range of the target in the subsequent frames by establishing a kinematic model (such as uniform speed model, uniform acceleration model or Kalman filter). The system then searches globally in this predicted area using more sensitive image matching or feature detection algorithms, thereby quickly and accurately recapturing and locking the target, restoring tracking and data flow.

[0114] The application avoids signal distortion or response lag that may be introduced by a single filtering method, can take the most appropriate compensation measures for data defects of different origins, thereby improving the accuracy and efficiency of data repair in a complex and variable real experimental environment, and ensuring that the final output physical quantity time series data has high integrity and physical authenticity.

[0115] The application further proposes outputting the final physical quantity analysis result. The goal of this stage is not only to calculate the physical quantity, but also to generate a comprehensive analysis report that integrates data, analysis, and quality evaluation, which can be directly understood and used by users. Specifically, it includes:

[0116] Curve fitting is performed on the corrected time series data, and the system automatically selects the most matching mathematical model according to the identified experimental type. For example, for simple harmonic vibration, a sine function is used for fitting; for uniform acceleration motion, a quadratic polynomial is used for fitting; for damped vibration, an exponential decay function is used for fitting. The fitting process usually uses robust regression algorithms (such as RANSAC or weighted least squares) to further reduce the influence of potential outliers, generating a smooth and standardized physical quantity time series curve that clearly reveals the internal law of the evolution of the physical quantity over time.

[0117] Subsequently, the system automatically extracts key feature parameters from the physical quantity time series curve; these key feature parameters are the core indicators for quantifying physical processes, mainly including extreme points (such as peak value, valley value, used to calculate amplitude, maximum speed, etc.), rate of change (obtained by calculating the first or second derivative of the curve, used to represent instantaneous speed, acceleration or reaction rate) and steady-state value (such as the final reading of the system after reaching equilibrium or the average value of periodic motion). The extraction of these parameters relies on mathematical analysis of the fitted curve function, such as accurately calculating the extreme points by taking the derivative and setting it to zero, ensuring the accuracy of the extraction.

[0118] In order to intuitively display the quality control and reliability of the data analysis process, the system visualizes the physical consistency residual C generated during the entire processing process along the time axis, generating a residual map that is displayed synchronously with the physical quantity time series curve. In this residual map, the physical consistency residual C is represented as a set of bar charts or curves that change over time, with their color or height mapping the size of the physical consistency residual C (for example, green represents low residual, red represents high residual). More importantly, the system establishes a bidirectional association mapping between the abnormal section in the residual map and the corresponding time point in the physical quantity time series curve. This means that when users review the final report, if they observe a red abnormal interval on the residual map, they can directly locate to the corresponding time period on the physical quantity curve to understand the data performance in that period, and thus clearly understand where the data has been significantly corrected by the system and the necessity of the correction.

[0119] Based on the identified experimental type, the corresponding report template is selected, and the physical quantity time curve, key feature parameters, residual map and correlation mapping are structured and packaged to generate the final physical quantity analysis result. The final physical quantity analysis result can be a PDF document, a structured data file (such as JSON format) or an interactive chart directly presented in the user interface.

[0120] By accurately extracting physical features from the smoothed fitting curve and combining residual visualization and correlation mapping with clear physical meaning, the application improves the transparency of data analysis and the reliability of the results through an integrated result presentation method, enabling users to quickly grasp the core laws of the physical process and clearly understand each key step of data processing, providing a solid and intuitive basis for scientific decision-making and teaching evaluation.

[0121] The following is a specific embodiment of a physical experiment data rapid analysis method based on image processing:

[0122] In a university physics laboratory, students use the system described in the invention to perform Young's modulus measurement experiments. The system continuously collects image sequences of the complete experimental scene including the optical lever and telescope scale group through a high-speed industrial camera (sampling rate 120 frames / second) fixed directly above the experimental table.

[0123] In the initial stage, the system performs dynamic background modeling on the collected image sequence. By calculating the running average and variance of the pixels of the initial 30 frames of images (about 0.25 seconds) with a sensitivity coefficient K=3.0, an initial background model is established. When the students start loading the weights, the optical lever scale image is shifted, and the system accurately extracts the moving area of the scale image as the foreground area through foreground separation.

[0124] Based on the image features of the foreground area, the system starts the experimental type recognition process. By extracting the linear motion trajectory features (displacement direction consistent with the direction of gravity), geometric structure features (elongated strip-shaped bright spots) and spatial distribution features (located in the center of the telescope field of view), a feature vector containing 12 dimensions is constructed. Calculate the similarity score of this vector with the pre-stored experimental template library, where the "stretching method to measure Young's modulus" template has a score of 0.92 (threshold 0.85), and through logical verification (scale image movement synchronized with weight loading), the system confirms the experimental type and outputs the type matching confidence 0.95.

[0125] According to the recognition result, the configuration module immediately loads the unified configuration set customized for this experiment, which includes: scale image reading extraction strategy based on sub-pixel edge detection, physical constraint rules based on Hooke's law and optical lever principle, and initialization parameters such as steel wire initial length , optical lever arm length D=0.258m, etc.

[0126] The system enters the parallel processing stage. The feature extraction thread performs sub-pixel level positioning on the image sequence according to the loaded strategy. When the fifth weight (mass m = 1.000 kg) is loaded, the image of the ruler appears blurred in the single frame image due to slight camera shake caused by the movement of the laboratory personnel, resulting in an abnormal displacement value read directly from the ruler: the measured ruler reading deviates from the reference position by Δl = 8.45 cm (the average displacement increment corresponding to the first four weights is about 1.82 cm). At the same time, the physical verification thread runs synchronously, which calculates the physical consistency residual C according to the expected form of Hooke's law (within the elastic limit, the displacement increment should be proportional to the load). For the fifth data point, the expected displacement increment is about 1.82 cm, and the physical consistency residual C generated by the abnormal value 8.45 cm increases sharply to C , far exceeding the preset convergence condition (the preset first threshold = 0.5 cm).

[0127] The verification module is triggered immediately, and the residual parameter mapping relationship is queried based on this residual pattern to determine that this is a transient positioning abnormality, and the feature extraction parameters are dynamically adjusted: the two-dimensional Gaussian surface fitting-based sub-pixel repositioning algorithm is started in the foreground area corresponding to the abnormal frame. After recalculating, the accurate displacement of the ruler image reading of this frame is corrected to Δl = 1.79 cm. The system iteratively executes until the physical consistency residuals C of the subsequent data points are all stable below 0.5 cm, satisfying the convergence condition, and finally generates a corrected displacement-load time series data that is smooth and consistent with the linear relationship of Hooke's law.

[0128] Based on this high-quality data, the system calls the Young's modulus calculation formula: , where the steel wire diameter d = 0.520 mm. Substituting the corrected data, the Young's modulus E of the steel wire is calculated to be E = 2.02 × 10 11 Pa, which is highly consistent with the standard reference value (about 2.00 × 10 11 Pa).

[0129] Through the automatic identification of the experimental type, the system can intelligently load the most suitable analysis strategy and physical knowledge. When encountering single-frame data anomalies caused by environmental interference, the system does not simply discard them as bad points, but sensitively discovers the problem through physical consistency verification and triggers a self-healing mechanism to perform sub-pixel level fine-tuning from the source, ensuring the accuracy and reliability of the final physical quantity (Young's modulus). The entire process does not require human intervention, achieving full-process automation and intelligentization from image acquisition to physical quantity output, improving the efficiency of physical experiment data processing and the scientific value of the results.

[0130] Example Two:

[0131] As Figure 4and Figure 5 As shown in the above-mentioned one kind based on image processing's physical experiment data fast analysis method, a kind of physical experiment data fast analysis system based on image processing, comprising:

[0132] Preprocessing module, for the original experimental image sequence of continuous acquisition is handled, from original experimental image sequence dynamic background separates out at least one foreground area representing experimental object;

[0133] Identification module, for the image feature of foreground area, automatically identifies the experimental type of physical experiment;

[0134] Configuration module, for according to the experimental type of identification, synchronous acquisition uniform configuration set;Uniform configuration set is associated with binding feature extraction strategy, physical constraint rule and initialization parameter;

[0135] Parallel processing module, for according to uniform configuration set, parallel execution:

[0136] According to the feature extraction strategy, the foreground image sequence of foreground area is extracted, and initial time series data is generated;

[0137] According to the physical constraint rule, the initial time series data is verified, and physical consistency residual is generated;

[0138] Verification module, for based on the mode of physical consistency residual, query pre-defined residual parameter mapping relationship, dynamically adjust the algorithm parameter of feature extraction or the positioning of foreground area, and iteratively execute until physical consistency residual meets preset convergence condition, and generate modified time series data;

[0139] Output module, for based on the time series data of modification, output final physical quantity analysis result.

[0140] The hardware basis of the present application includes: image acquisition unit, usually by fixed in the upper or side of experimental scene high-speed or high-resolution video camera is realized, is responsible for continuous acquisition original experimental image sequence;Data processing unit, by one or more embedded computers or servers is constituted, is equipped with sufficient CPU, GPU and memory resources, for carrying and running each software module described below;And result output unit, it can be computer display, touch screen or through network interface connected remote terminal, for presenting final physical quantity analysis result.

[0141] On the above-mentioned hardware basis, the system works through the following modules cooperatively:

[0142] The preprocessing module runs on the data processing unit, and its core function is to execute dynamic background modeling and foreground segmentation algorithm. It receives the original data stream from the image acquisition unit, establishes and continuously updates the background model by calculating the statistical characteristics (such as mean and variance) of pixels or image blocks, and then accurately separates the moving experimental object (such as the pendulum of a simple pendulum or a moving trolley) from the dynamic background, and outputs the foreground region mask.

[0143] The recognition module receives the foreground region sequence from the preprocessing module. It extracts the geometric structure and motion trajectory of the foreground region, constructs a feature vector, and matches it with the pre-stored experimental template library, thereby automatically identifying the type of physical experiment currently being conducted (such as "simple harmonic vibration experiment" or "inclined plane acceleration motion experiment").

[0144] The configuration module synchronously calls a corresponding unified configuration set from the system configuration database according to the experimental type identifier output by the recognition module. The unified configuration set is a data package that internally associates and binds the feature extraction strategy suitable for this experiment (such as tracking the center of mass or the edge), the physical constraint rule (such as the mathematical expression of the law of conservation of energy), and the algorithm initialization parameters (such as the initial position and mass).

[0145] The parallel processing module is the computing core of the system, which executes two processes in parallel according to the information provided by the unified configuration set. Process one processes the foreground image sequence according to the feature extraction strategy, and outputs the initial time series data (such as position-time sequence). Process two simultaneously verifies the initial time series data according to the physical constraint rule, and calculates the physical consistency residual C, which quantifies the deviation of the data from the physical law.

[0146] The verification module monitors the pattern of the physical consistency residual C, and queries the pre-set knowledge base of residual parameter mapping relationships. According to the mapping relationship, it dynamically issues instructions to adjust the algorithm parameters of the feature extraction module (such as the size of the search window) or requests the preprocessing module to reposition the foreground region. Through iterative operation, until the physical consistency residual C meets the pre-set convergence condition (such as the residual norm being less than a threshold), the final output is the corrected time series data with high confidence.

[0147] The output module receives the corrected time series data, selects the appropriate visualization template according to the experimental type, integrates the physical quantity time series curve, key feature parameters (such as period, acceleration), and quality assessment report, and finally generates and displays the final physical quantity analysis result on the result output unit.

[0148] Take the experiment of "Michelson interferometer measurement of small length" as an example: after the system starts, the image acquisition unit (high-definition CCD camera) begins to shoot the video of the change of the interference ring. The preprocessing module accurately separates the foreground area of the interference ring stripe from the video stream. The identification module determines that the experiment type is "Michelson interferometer experiment" according to the characteristics of the concentric circular ring and the center throughput law. The configuration module loads the corresponding unified configuration set, which includes the "stripe center positioning and ring counting" feature extraction strategy and the "isoclinal interference formula" physical constraint rule. The parallel processing module outputs the interference ring center coordinates and the ring number change sequence (initial timing data) in real time according to the strategy, and synchronously verifies the sequence using the interference formula to generate physical consistency residuals. When the moving mirror displacement causes the ring center to be temporarily blurred, the check module detects that the residual error increases abnormally, and immediately triggers the sub-pixel level stripe center repositioning algorithm, and dynamically adjusts the ring counting logic parameters. After iteration and correction, the residual error converges to within the preset threshold.

[0149] Based on the corrected ring number-displacement data, the system automatically calculates the helium-neon laser wavelength value of 632.8nm, and outputs a complete experiment report containing the wavelength measurement result, the interference ring change curve and the uncertainty analysis, realizing the full-process automatic analysis of the interference measurement.

[0150] The present application realizes universality through experiment type self-recognition and strategy self-adaptation, reduces the threshold of manual operation and data processing, and provides an efficient and reliable solution for the digitization and intelligentization of scientific experiment teaching and research.

[0151] The technical scope of the present application is not limited to the content in the above description, and those skilled in the art can make various modifications and changes to the above embodiments without departing from the technical idea of the present application, and these modifications and changes should all belong to the protection scope of the present application.

Claims

1. A rapid analysis method for physical experimental data based on image processing, characterized in that: Includes the following steps: The continuously acquired raw experimental image sequence is processed to separate at least one foreground region representing the experimental object from the dynamic background; Based on the image features of the foreground region, the experiment type of the physical experiment is automatically identified; Based on the identified experiment type, a unified configuration set is obtained synchronously; the unified configuration set is associated with and bound to feature extraction strategies, physical constraint rules, and initialization parameters. Based on the unified configuration set, execute in parallel: Based on the feature extraction strategy, features are extracted from the foreground image sequence of the foreground region to generate initial time-series data; Verify the initial time-series data according to the physical constraint rules, and generate physical consistency residuals; Based on the pattern of the physical consistency residual, the predefined residual parameter mapping relationship is queried, the algorithm parameters of the feature extraction or the positioning of the foreground region are dynamically adjusted, and the process is iterated until the physical consistency residual meets the preset convergence condition, thereby generating corrected time series data. Based on the corrected time series data, the final physical quantity analysis results are output.

2. The method for rapid analysis of physical experimental data based on image processing according to claim 1, characterized in that: The specific steps for separating at least one foreground region representing the experimental object from the dynamic background of the original experimental image sequence include: Obtain the initial multi-frame images of the original experimental image sequence, calculate the running average and variance at the pixel level or block level, and establish a background model; In subsequent processing, the update rate of the background model is adaptively adjusted based on the scene motion represented by the separation results of the foreground region.

3. The rapid analysis method for physical experiment data based on image processing according to claim 1, characterized in that: The specific steps for automatically identifying the experiment type of the physical experiment based on the image features of the foreground region include: Extract image patch data containing spatial distribution, motion trajectory, and texture features from the foreground region; Calculate the geometric invariant features and spatiotemporal motion features of the image patch data to generate a multi-dimensional feature vector for experimental type identification; By performing cluster analysis and template matching on the multi-dimensional feature vectors, the experiment type identifier and recognition confidence level are output.

4. The method for rapid analysis of physical experimental data based on image processing according to claim 1, characterized in that: The specific steps for feature extraction based on the feature extraction strategy include: Temporal difference analysis is performed on the image sequence of the foreground region to obtain inter-frame motion information; Spatial gradient analysis is performed on a single frame image of the foreground region to obtain edge contour information; Based on the target motion speed reflected by the inter-frame motion information, the fusion weight of the inter-frame motion information and the edge contour information is dynamically adjusted. The experimental object is located and tracked based on the fused information, and the position or shape parameters of the experimental object are output.

5. The rapid analysis method for physical experiment data based on image processing according to claim 4, characterized in that: Before generating the initial time series data, calibration and unit conversion steps are also included, specifically: Obtain pre-stored or dynamically detected scale information from the original experimental image sequence; Using the scale information, the position or shape parameter is converted into a physical quantity measurement value with physical units, and the initial time series data is generated based on the physical quantity measurement value.

6. The method for rapid analysis of physical experimental data based on image processing according to claim 1, characterized in that: Dynamically adjusting the algorithm parameters for feature extraction specifically includes one or more of the following methods: Within the foreground region of the time frame corresponding to the abnormal data point, recalculate the subpixel-level positioning and fitting. The abnormal data segments are replaced by interpolation using the values ​​of adjacent high-confidence data points in the initial time series data.

7. The rapid analysis method for physical experimental data based on image processing according to claim 1, characterized in that: The specific steps for automatically identifying the experiment type of the physical experiment include: The image features are constructed into a multi-dimensional feature vector that includes geometric structure features, motion pattern features, and optical features; Calculate the similarity score between the multi-dimensional feature vector and each template vector in the pre-stored experimental template library; Logical verification is performed on the similarity score based on predefined experiment type discrimination rules; When the similarity score is higher than the type recognition threshold and passes the logical verification, the corresponding experimental type identifier and type matching confidence score are output. The type matching confidence score is used as a weighting factor in the recommendation calculation of the unified configuration set.

8. The rapid analysis method for physical experiment data based on image processing according to claim 1, characterized in that: The pattern based on physical consistency residuals also includes establishing an anomaly classification model based on the physical constraint rules, specifically including: The identified abnormal data are categorized into transient noise, systematic bias, and tracking loss types. For the aforementioned instantaneous noise type, an adaptive filtering algorithm based on neighborhood confidence is used for smoothing. For the aforementioned system deviation type, initiate the parameter self-calibration process to recalculate the mapping relationship between pixels and physical units; For the aforementioned tracking loss type, a foreground region relocalization mechanism based on motion trajectory prediction is initiated in subsequent image frames.

9. The method for rapid analysis of physical experimental data based on image processing according to claim 1, characterized in that: The final physical quantity analysis results are output, including: Curve fitting is performed on the corrected time series data to generate standardized physical quantity time series curves; Key feature parameters are extracted from the time series curves of the physical quantities; the key feature parameters include extreme points, rates of change, and steady-state values. The physical consistency residuals are visualized and encoded along the time axis to generate a residual map that is displayed synchronously with the time series curve of the physical quantity; Establish a correlation mapping between the abnormal segments in the residual spectrum and the corresponding time points in the time series curves of the physical quantities; Based on the identified experiment type, the corresponding report template is selected, and the physical quantity time series curve, the key feature parameters, the residual spectrum, and the correlation mapping are encapsulated in a structured manner to generate the final physical quantity analysis results.

10. A rapid analysis system for physical experiment data based on image processing, characterized in that: A rapid analysis method for physical experimental data based on image processing, as described in any one of claims 1 to 9, includes: The preprocessing module is used to process the continuously acquired raw experimental image sequence and separate at least one foreground region representing the experimental object from the dynamic background of the raw experimental image sequence. The identification module is used to automatically identify the type of physical experiment based on the image features of the foreground region. The configuration module is used to synchronously obtain a unified configuration set according to the identified experiment type; the unified configuration set is associated with and bound to feature extraction strategies, physical constraint rules and initialization parameters; The parallel processing module is used to execute in parallel according to the unified configuration set: Based on the feature extraction strategy, features are extracted from the foreground image sequence of the foreground region to generate initial time-series data; Verify the initial time-series data according to the physical constraint rules, and generate physical consistency residuals; The verification module is used to query the predefined residual parameter mapping relationship based on the pattern of the physical consistency residual, dynamically adjust the algorithm parameters of the feature extraction or the positioning of the foreground region, and iterate until the physical consistency residual meets the preset convergence condition to generate corrected time series data. The output module is used to output the final physical quantity analysis results based on the corrected time series data.

Images