Cutting fluid solid particle concentration measuring method and system

Abstract

The invention discloses a cutting fluid solid particle concentration measurement method and system, and relates to the technical field of particle concentration measurement, a time sliding window mode is used for carrying out fusion processing on an image sequence, synchronous acquisition of temperature and illumination intensity is combined, a dynamic temperature illumination normalized background image is constructed, and the concentration of solid particles in a cutting fluid is measured. The problems of brightness drift, image flicker and the like caused by environmental fluctuation are effectively eliminated, and the stability and adaptability of background modeling are remarkably improved. Through a disturbance identification and characteristic parameter extraction module, multi-dimensional characteristic extraction is carried out on non-particle disturbance such as bubbles, oil films and light reflection possibly occurring in the image, a disturbance characteristic set Rdr is formed, and the system can accurately identify and distinguish non-real particle signals. And through a dynamic difference denoising analysis module, a disturbance feature set Rdr is combined to carry out adaptive denoising on a difference image, a real particle structure is reserved, a non-particle interference region is eliminated, and a purified image with an enhanced edge is output, so that cleaner and more accurate basic data is provided for subsequent particle identification.

Description

Technical Field

[0001] The present invention relates to the technical field of particle concentration measurement, and specifically to a method and a system for measuring the solid particle concentration of cutting fluid. Background Art

[0002] As an important process medium in metal cutting, grinding and cooling, a large number of micron-sized solid particles are often mixed into cutting fluid during its use due to factors such as tool wear and workpiece peeling, which affect lubrication, cooling and machining accuracy. High-precision and real-time measurement of the solid particle concentration in cutting fluid has become an indispensable functional module in numerical control machining and high-end manufacturing processes, especially applicable to the "cutting fluid solid particle concentration measurement system" constructed based on image acquisition and visual analysis methods.

[0003] In current industrial practices, the online detection of the solid particle concentration in cutting fluid mostly relies on optical imaging and image recognition algorithms for particle recognition and counting. However, in actual use scenarios, the flow path of cutting fluid is complex, there is often an oil film floating on the surface, sudden changes in the reflective area, and microbubble impurities. These non-particle disturbance structures often appear as "bright spots" or "edge continuous blocks" in the image, and are easily misrecognized as real particles by traditional edge detection or connected domain algorithms, resulting in a high concentration estimation value.

[0004] Existing systems lack the ability to dynamically model background disturbances and cannot identify the essential differences between "stable background" and "real particle events" in the time dimension. Under the action of multiple factors such as the continuous flow of cutting fluid, continuous increase in temperature, uneven illumination, and change in interfacial tension, non-solid interference objects will stay in the image field of view for a long time, forming a "pseudo-particle aggregation effect", which is continuously misjudged as a particle signal, thereby causing the phenomenon of a high number of false particles and continuous deviation of the concentration estimation value. Summary of the Invention

[0005] Aiming at the deficiencies of the existing technology, the present invention provides a method and a system for measuring the solid particle concentration of cutting fluid, which solves the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention is realized through the following technical solutions: A cutting fluid solid particle concentration measurement system includes an original image acquisition module, a multi-frame time series fusion and background modeling module, a disturbance recognition and feature parameter extraction module, a dynamic differential denoising analysis module, a particle recognition and concentration estimation module, and a concentration anomaly feedback and self-correction module;

[0007] The original image acquisition module acquires the cutting fluid image stream per unit time through a sensor and an imaging array, obtains the image sequence It(x, y), and acquires the temperature T, illumination intensity L, and pixel mean μI attached to the image sequence It(x, y);

[0008] The multi-frame time series fusion and background modeling module performs background modeling on the image sequence It(x, y) based on the time sliding window Tw, extracts stable structure signals, and obtains the temperature and illumination normalized background image B(x, y);

[0009] The disturbance identification and feature parameter extraction module performs feature extraction on the acquired image sequence It(x, y) and the temperature illumination normalized background image B(x, y), extracts non-particle disturbance features, and fits them into the disturbance feature set Rdr;

[0010] The dynamic difference denoising analysis module analyzes the temperature and illumination normalized background image B(x, y) based on the disturbance feature set Rdr as the judgment basis to obtain the purified image D(x, y);

[0011] The particle identification and concentration estimation module identifies and counts particles according to the obtained purification map D(x, y) and calculates the concentration value Cest;

[0012] The concentration abnormality feedback and self-correction module detects the acquired concentration value Cest, obtains the concentration offset coefficient δc, and compares it with the preset concentration coefficient threshold Thc to determine whether it is abnormal, and automatically performs correction to obtain the concentration correction value Cadj.

[0013] Preferably, the original image acquisition module includes an imaging data acquisition unit and an auxiliary environmental parameter synchronous acquisition unit;

[0014] The imaging data acquisition unit uses the area array image sensor in the industrial-grade CMOS area array camera, in conjunction with a hyperspectral backlight source, to shoot the cutting fluid at a fixed acquisition window, collect the cutting fluid image stream per unit time, and form a continuous frame image sequence It(x, y);

[0015] Among them, the size of the image sequence It(x, y) is H*W;

[0016] The construction formula of the image sequence It(x, y) is: It(x, y), x∈[1,W], y∈[1,H];

[0017] The auxiliary environmental parameter synchronous acquisition unit synchronously records the temperature T, illumination intensity L and pixel mean μI during the image acquisition process;

[0018] The temperature T is detected and acquired by an integrated thermistor digital temperature sensor attached to the image acquisition window housing;

[0019] The illumination intensity L is obtained by measuring the illumination intensity of the imaging area using a high-precision digital photosensor;

[0020] The pixel mean μI is obtained by the following formula:

[0021]

[0022] Preferably, the multi-frame time-series fusion and background modeling module includes a sliding window image fusion unit and a temperature-illumination normalization correction unit;

[0023] The sliding window image fusion unit traces back the image frame sequence {It-i(x, y)}, i = 0, 1,..., Tw-1 within the time sliding window Tw from the current moment, and performs weighted fusion to obtain a preliminary background image Bo(x, y);

[0024] The preliminary background image Bo(x, y) is obtained through the following formula:

[0025]

[0026] In the formula, Z represents a normalization factor, It,i(x, y) represents the image sequence It(x, y) of the i-th frame within the time sliding window Tw, and ωi represents the preset weight value of the image sequence It(x, y) of the i-th frame;

[0027] The temperature-illumination normalization correction unit collects the temperature T and illumination intensity L corresponding to each frame of the image sequence It(x, y), establishes a photothermal normalization factor GRt, and performs normalization correction on the preliminary background image Bo(x, y) to obtain a temperature-illumination normalized background image B(x, y);

[0028] The photothermal normalization factor GRt is obtained through the following formula:

[0029]

[0030] In the formula, αT represents the sensitivity factor of temperature, αL represents the sensitivity factor of illumination, To represents the set reference temperature, and Lo represents the set reference illumination intensity;

[0031] The temperature-illumination normalized background image B(x, y) is obtained through the following formula:

[0032] B(x,y) = Bo(x,y) * GRt.

[0033] Preferably, the disturbance recognition and feature parameter extraction module includes a structural feature extraction unit and a time-series adhesion analysis unit;

[0034] The structural feature extraction unit receives the image sequence It(x, y) and the temperature-illumination normalized background image B(x, y), performs analysis, and obtains a difference image Δt(x, y);

[0035] The difference image Δt(x, y) is obtained through the following formula:

[0036] Δt(x,y) = |It(x,y) - B(x,y)|;

[0037] Feature extraction is performed on the image sequence It(x, y) and the temperature and illuminance normalized background image B(x, y) to extract non-particle disturbance features, including morphological disturbance feature Fmo, light field anomaly index Fir, and liquid adhesion time drift factor Fvs;

[0038] The morphological disturbance feature Fmo is obtained through the following formula:

[0039]

[0040] In the formula, represents the set of boundary pixels of the region, represents the boundary gradient, and Nb represents the number of boundary pixel points;

[0041] The light field anomaly index Fir is obtained through the following formula:

[0042]

[0043] In the formula, Ωh represents the high-frequency region in the difference map, and σω 2 x,y represents the variance of the gray values of the pixels within the window.

[0044] Preferably, the time-series adhesion analysis unit tracks the abnormally high difference state of each pixel point in the image sequence It(x, y) within a continuous time window, determines whether there is an abnormal structure that has not changed for a long time, and obtains the liquid adhesion time drift factor Fvs;

[0045] The liquid adhesion time drift factor Fvs is obtained through the following formula:

[0046]

[0047] In the formula, Ωs represents the region of pixel points where high-frequency disturbances occur within the past sliding window, and r(x, y) represents the maximum time length for which the pixel point (x, y) is continuously in a disturbed state;

[0048] The obtained morphological disturbance feature Fmo, light field anomaly index Fir, and liquid adhesion time drift factor Fvs are fitted to obtain the disturbance feature set Rdr.

[0049] Preferably, the dynamic difference denoising analysis module includes a disturbance fusion discrimination unit and an adaptive purification map construction unit;

[0050] The disturbance fusion discrimination unit compares the obtained disturbance feature set Rdr with the preset morphological disturbance threshold Tmo, light field anomaly threshold Tir, and adhesion time drift threshold Tvs, obtains the disturbance significance function Ψ, and determines whether there is a large-area interference phenomenon in the current frame, and constructs the disturbance mask matrix M(x, y);

[0051] The perturbation significance function Ψ is obtained through the following formula:

[0052]

[0053] In the formula, λ1, λ2, and λ3 respectively represent the preset weight values of the morphological perturbation feature Fmo, the light field anomaly index Fir, and the liquid adhesion time drift factor Fvs;

[0054] The perturbation mask matrix M(x, y) is obtained through the following formula:

[0055]

[0056] In the formula, Δt(x, y) represents the difference map, and θd represents the local perturbation response judgment threshold;

[0057] When the value of the perturbation mask matrix M(x, y) is 1, it represents a normal area;

[0058] When the value of the perturbation mask matrix M(x, y) is 0, it represents a perturbation elimination area.

[0059] Preferably, the adaptive purification map construction unit analyzes the obtained perturbation mask matrix M(x, y), generates the purification map D(x, y), and retains the real particle boundary through the edge enhancement convolution kernel algorithm; preventing the edges of small particles from being smoothed or suppressed;

[0060] The purification map D(x, y) is obtained through the following formula:

[0061] D(x,y) = M(x,y) * Δt(x,y);

[0062] Filter and filter the obtained purification map D(x, y), and retain the real particle boundary;

[0063] The filter and filter formula is as follows:

[0064] D(x,y) ← D(x,y) + η * D(x,y) - GD(D(x,y));

[0065] In the formula, η represents the edge enhancement coefficient, and GD represents Gaussian kernel smoothing, which controls the over-edge blurring of the image.

[0066] Preferably, the particle recognition and concentration estimation module receives the purification map D(x, y), and uses connected component analysis for particle segmentation, extracts the particle regions one by one, and obtains the area Ak and the average scattering intensity Sk of particle k;

[0067] Set each independent region as And satisfy:

[0068] Rk = {(x, y) | D(x, y) > Tlx};

[0069] Wherein, Tlx represents the pixel brightness threshold;

[0070] The area Ak of particle k is obtained by the following formula:

[0071]

[0072] The average scattering intensity Sk is obtained by the following formula:

[0073]

[0074] Wherein, ωxy represents the weighting factor of the pixel pair for the scattering center symmetry;

[0075] The obtained area Ak and average scattering intensity Sk of each particle k are mapped to the particle concentration value Cest per unit volume;

[0076] The concentration value Cest is obtained by the following formula:

[0077]

[0078] Wherein, V represents the volume of the imaging corresponding liquid, M represents the total number of effective particles, β represents the non-linear particle scattering correction index, and ln represents the logarithmic function.

[0079] Preferably, the concentration anomaly feedback and self-correction module detects the concentration values Cest in different periods, obtains the concentration offset coefficient δc, and compares it with the preset concentration coefficient threshold Thc to judge the concentration state;

[0080] The concentration offset coefficient δc is obtained by the following formula:

[0081]

[0082] Wherein, Cw represents the concentration mean value, σw represents the concentration standard deviation, and E represents a positive constant to prevent division by zero;

[0083] The judgment of the concentration state is obtained by matching in the following manner:

[0084] When the concentration offset coefficient δc ≤ the concentration coefficient threshold Thc, it indicates that the concentration state is normal;

[0085] When the concentration offset coefficient δc > the concentration coefficient threshold Thc, it indicates that the concentration state is abnormal;

[0086] When the concentration state is abnormal, a concentration correction factor Kc is introduced to automatically compensate the concentration value Cest to obtain the concentration correction value Cadj;

[0087] The concentration correction factor Kc is obtained through the following formula:

[0088]

[0089] In the formula, ka represents the basic correction coefficient, pr represents the particle volume deviation rate adjustment factor, Ak represents the area of particle k, pA represents the average value of the area, and β represents the non-linear particle scattering correction index;

[0090] The concentration correction value Cadj is obtained through the following formula:

[0091]

[0092] The method for measuring the solid particle concentration of the cutting fluid includes the following steps:

[0093] Step 1: The original image acquisition module acquires the cutting fluid image stream within a unit time through the sensor and the imaging array, obtains the image sequence It(x, y), and acquires the temperature T, illumination intensity L, and pixel mean μI attached to the image sequence It(x, y);

[0094] Step 2: The multi-frame time series fusion and background modeling module performs background modeling on the image sequence It(x, y) based on the time sliding window Tw, extracts the stable structure signal, and obtains the temperature and illumination normalized background image B(x, y);

[0095] Step 3: The perturbation recognition and feature parameter extraction module extracts features from the obtained image sequence It(x, y) and the temperature and illumination normalized background image B(x, y), extracts the non-particle perturbation features, and fits them into the perturbation feature set Rdr;

[0096] Step 4: The dynamic differential denoising analysis module analyzes the temperature and illumination normalized background image B(x, y) based on the perturbation feature set Rdr as the judgment basis, and obtains the purified image D(x, y);

[0097] Step 5: The particle recognition and concentration estimation module performs particle recognition and counting based on the obtained purified image D(x, y), and calculates and obtains the concentration value Cest;

[0098] Step 6: The concentration anomaly feedback and self-correction module detects the obtained concentration value Cest, obtains the concentration deviation coefficient δc, compares it with the preset concentration coefficient threshold Thc, determines whether it is abnormal, and automatically performs correction to obtain the concentration correction value Cadj.

[0099] The present invention provides a method and a measurement system for measuring the solid particle concentration of the cutting fluid, having the following beneficial effects:

[0100] (1) When the system is running, the image sequence is fused using the time sliding window method, and combined with the synchronous acquisition of temperature and illumination intensity, a dynamic temperature-illumination normalized background image is constructed, effectively eliminating problems such as brightness drift and image flickering caused by environmental fluctuations, and significantly improving the stability and adaptability of background modeling. Through the perturbation recognition and feature parameter extraction module, multi-dimensional features of non-particle perturbations such as bubbles, oil films, and reflections that may appear in the image are extracted to form a perturbation feature set Rdr. The system can accurately identify and distinguish non-real particle signals, avoiding misjudgment and missed judgment in traditional methods.

[0101] Through the dynamic differential denoising analysis module, the differential image is adaptively denoised in combination with the perturbation feature set, retaining the real particle structure, removing the non-particle interference area, and outputting a purified image with enhanced edges, providing cleaner and more accurate basic data for subsequent particle recognition. Instead of using the traditional linear counting method, composite information such as area and scattering intensity is introduced for concentration estimation, which is closer to the physical state particle distribution characteristics, avoiding misestimation of large and small particles, and improving the overall measurement accuracy.

[0102] (2) Through the structural feature extraction unit, a differential image is constructed between the image sequence and the temperature-illumination normalized background image, and information such as boundary gradient, pixel fluctuation, and brightness contrast is further extracted to comprehensively analyze the morphological features, optical anomalies, and local intensity changes of non-particle perturbations in the image. For the first time, the image microstructure perturbation behavior is introduced into the preprocessing process of particle recognition, enhancing the observability of the perturbation source. By calculating the morphological perturbation feature Fmo and the optical field anomaly index Fir, this module can not only identify the perturbation structures similar to particles in shape, but also identify the asymmetric local high-frequency bright spots generated by oil film fluctuations and light interference. This dual-channel recognition mechanism can avoid the problem of misjudgment of single brightness or edge features in traditional systems, and significantly improve the identification accuracy of the system for complex visual interference.

[0103] (3) Through the multi-source discrimination mechanism constructed by the perturbation significance function, the system can identify interference areas such as reflection, oil film, and attached foam in real time, effectively preventing these areas from being misjudged as particle signals, and improving the accuracy and reliability of particle concentration estimation; the perturbation mask mechanism realizes the transformation of image purification processing from overall static processing to regional dynamic control, enabling the system to automatically identify and process complex disturbances without manually setting a large number of boundary conditions or manual calibration, greatly improving the intelligence of system operation. The edge enhancement mechanism retains and strengthens the edge structure of real particles, especially for particles with blurred edges, small size, and weak light scattering. It has a good detection enhancement effect, which helps the system to fully perceive the real distribution of particles and improve the integrity and continuity of overall particle recognition. Regardless of the conditions of uneven lighting, frequent light field disturbances, or drastic changes in the liquid level, the system can maintain the ability to accurately extract particle targets through dynamic mask and edge adjustment strategies to ensure stable output of concentration detection.

[0104] (4) By performing statistical analysis on the concentration values within multiple measurement cycles and calculating the deviation coefficient, the system can determine whether the current concentration value is in an abnormal state of fluctuation. This judgment mechanism effectively avoids the interference of short-term particle clusters or instantaneous lighting anomalies on concentration estimation, and improves the steady-state judgment ability of the system. When the system determines that the concentration state is abnormal, this embodiment introduces a concentration correction factor based on the degree of deviation of the particle area to automatically compensate for the original concentration estimate. The correction model simultaneously considers the area distribution, deviation rate and nonlinear correction coefficient of the particles, has high adaptability and system self-healing ability, and ensures the continuity and credibility of the concentration output. BRIEF DESCRIPTION OF THE DRAWINGS

[0105] Figure 1 It is a schematic diagram of the flow chart of the cutting fluid solid particle concentration measurement system of the present invention;

[0106] Figure 2 Schematic diagram of the steps of the method for measuring the concentration of solid particles in cutting fluid of the present invention;

[0107] Figure 3 This is a schematic diagram of the process of obtaining the solid particle concentration value of the cutting fluid of the present invention;

[0108] Figure 4 It is a line graph of the disturbance significance function of the present invention. DETAILED DESCRIPTION

[0109] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0110] Example 1

[0111] The present invention provides a cutting fluid solid particle concentration measurement system. Please refer to Figures 1 to 4 , which includes an original image acquisition module, a multi-frame time series fusion and background modeling module, a disturbance recognition and feature parameter extraction module, a dynamic differential denoising analysis module, a particle recognition and concentration estimation module, and a concentration anomaly feedback and self-correction module;

[0112] The original image acquisition module acquires the cutting fluid image stream within a unit time through a sensor and an imaging array, obtains the image sequence It(x, y), and acquires the temperature T, illumination intensity L, and pixel mean μI attached to the image sequence It(x, y);

[0113] The multi-frame time series fusion and background modeling module performs background modeling on the image sequence It(x, y) based on the time sliding window Tw, extracts the stable structure signal, and obtains the temperature-illumination normalized background image B(x, y);

[0114] The disturbance recognition and feature parameter extraction module performs feature extraction on the obtained image sequence It(x, y) and the temperature-illumination normalized background image B(x, y), extracts the non-particle disturbance features, and fits them into the disturbance feature set Rdr;

[0115] The dynamic differential denoising analysis module analyzes the temperature-illumination normalized background image B(x, y) based on the disturbance feature set Rdr as the judgment basis, and obtains the purified image D(x, y);

[0116] The particle recognition and concentration estimation module performs particle recognition and counting based on the obtained purified image D(x, y), and calculates and obtains the concentration value Cest;

[0117] The concentration anomaly feedback and self-correction module detects the obtained concentration value Cest, obtains the concentration offset coefficient δc, compares it with the preset concentration coefficient threshold Thc, judges whether it is abnormal, and automatically performs correction to obtain the concentration correction value Cadj.

[0118] In this embodiment, the image sequence is fused and processed by using the time sliding window method, and combined with the synchronous acquisition of temperature and illumination intensity, a dynamic temperature-illumination normalized background image is constructed, effectively eliminating problems such as brightness drift and image flicker caused by environmental fluctuations, and significantly improving the stability and adaptability of background modeling. Through the disturbance recognition and feature parameter extraction module, multi-dimensional feature extraction is performed on non-particle disturbances such as bubbles, oil films, and reflections that may appear in the image, forming the disturbance feature set Rdr. The system can accurately identify and distinguish non-real particle signals, avoiding misjudgment and missed judgment in traditional methods.

[0119] Through the dynamic differential denoising analysis module, the differential image is adaptively denoised in combination with the perturbation feature set, preserving the real particle structure, removing the non-particle interference regions, and outputting a purified image with enhanced edges, providing cleaner and more accurate basic data for subsequent particle recognition. Instead of using the traditional linear counting method, composite information such as area and scattering intensity is introduced for concentration estimation, which is closer to the physical particle distribution characteristics, avoiding misestimation of large and small particles and improving the overall measurement accuracy.

[0120] Through background normalization modeling, perturbation feature recognition, and dynamic denoising, the system effectively filters out misjudgments caused by non-particle interferences, making the estimated particle concentration value more real and reliable, and avoiding false high readings caused by problems such as oil films and bubbles. Since the system can sense and compensate for environmental temperature and illumination fluctuations and has a dynamic self-correction mechanism, it can be applied to various cutting conditions, with stronger generalization ability and deployment flexibility. Avoiding frequent liquid change or false alarm behaviors caused by misjudging particle over-standard, the service life of the cutting fluid can be reasonably extended, while reducing maintenance frequency and material waste, saving operation costs for the factory. The particle concentration data output by the system has the characteristics of high precision and low fluctuation, and can provide stable and reliable data input for numerical control systems, automatic liquid replenishment systems, wear warning modules, etc., supporting subsequent decision-making logic.

[0121] Example 2

[0122] This example is an explanatory note based on Example 1. Please refer to Figure 1 , specifically: The original image acquisition module includes an imaging data acquisition unit and an auxiliary environmental parameter synchronous acquisition unit;

[0123] The imaging data acquisition unit uses the area array image sensor in an industrial-grade CMOS area array camera, in cooperation with a hyperspectral backlight source, to take pictures of the cutting fluid at a fixed acquisition window, acquiring the cutting fluid image stream per unit time and constituting a continuous frame image sequence It(x, y);

[0124] Among them, the size of the image sequence It(x, y) is H*W;

[0125] The construction formula of the image sequence It(x, y) is: It(x, y), x∈[1, W], y∈[1, H];

[0126] The auxiliary environmental parameter synchronous acquisition unit synchronously records the temperature T, illumination intensity L, and pixel mean μI during the image acquisition process;

[0127] Among them, the temperature T is obtained by detecting the temperature through an integrated thermistor digital temperature sensor attached to the outer shell of the image acquisition window;

[0128] The illumination intensity L is obtained by measuring the illumination intensity of the imaging area using a high-precision digital photosensitive sensor;

[0129] The pixel mean value μI is obtained by the following formula:

[0130]

[0131] The multi-frame time-sequence fusion and background modeling module includes a sliding window image fusion unit and a temperature and illumination normalization correction unit;

[0132] The sliding window image fusion unit traces back the image frame sequence {It-i(x, y)}, i = 0, 1,..., Tw-1 within the time sliding window Tw from the current moment, and performs weighted fusion to obtain the preliminary background image Bo(x, y);

[0133] The preliminary background image Bo(x, y) is obtained by the following formula:

[0134]

[0135] In the formula, Z represents the normalization factor, It,i(x, y) represents the image sequence It(x, y) of the i-th frame within the time sliding window Tw, and ωi represents the preset weight value of the image sequence It(x, y) of the i-th frame;

[0136] The temperature and illumination normalization correction unit collects the temperature T and illumination intensity L corresponding to each frame of the image sequence It(x, y), establishes the photothermal normalization factor GRt, and performs normalization correction on the preliminary background image Bo(x, y) to obtain the temperature and illumination normalized background image B(x, y);

[0137] The photothermal normalization factor GRt is obtained by the following formula:

[0138]

[0139] In the formula, αT represents the sensitivity factor of temperature, αL represents the sensitivity factor of illumination, To represents the set reference temperature, and Lo represents the set reference illumination intensity;

[0140] The temperature and illumination normalized background image B(x, y) is obtained by the following formula:

[0141] B(x,y) = Bo(x,y) * GRt.

[0142] In this embodiment, compared with the existing system that only relies on image data for particle recognition, this embodiment realizes the synchronous acquisition of the image sequence, temperature T, illumination intensity L, and pixel mean value μI by setting an auxiliary environmental parameter synchronous acquisition unit in the original image acquisition module. This design ensures that the subsequent processing module can perform image normalization modeling based on the multi-parameter collaboration at the same time point, avoiding the error accumulation caused by environmental changes.

[0143] In this embodiment, a time sliding window structure is adopted in multi-frame temporal fusion, and a preset weight factor is introduced for weighted fusion of image frames, thereby improving the system's ability to extract the historical change trend of images. This design can effectively suppress the distortion of the background model caused by single-frame anomalies, making the background extraction result more continuous and stable. By constructing a photothermal normalization factor, joint correction of temperature and illuminance is performed on the preliminary background image to achieve precise compensation for the brightness offset of images under different temperature fields and lighting conditions. This mechanism significantly enhances the robustness of the system, enabling the background modeling module to operate stably in complex environments such as day-night light changes, equipment heating, and workshop lighting interference.

[0144] Through multi-frame fusion within the time sliding window and weighted suppression of mutation frames, the temporal noise fluctuations of the background image are effectively smoothed, avoiding misidentifications caused by background changes, and contributing to the improvement of the accuracy of subsequent disturbance detection and particle extraction. The photothermal normalization processing mechanism compensates for the problem of inconsistent brightness caused by environmental interference in images, ensuring that the system can still output a stable and reliable background image as a reference under special working conditions such as high temperature, high brightness, and low light.

[0145] Embodiment 3

[0146] This embodiment is an explanatory description based on Embodiment 2. Please refer to Figure 1 and Figure 3 Specifically: The disturbance recognition and feature parameter extraction module includes a structural feature extraction unit and a temporal adhesion analysis unit;

[0147] The structural feature extraction unit receives the image sequence It(x, y) and the temperature-illuminance normalized background image B(x, y), analyzes them, and obtains the difference image Δt(x, y);

[0148] The difference image Δt(x, y) is obtained through the following formula:

[0149] Δt(x,y) = |It(x,y) - B(x,y)|;

[0150] Feature extraction is performed on the image sequence It(x, y) and the temperature-illuminance normalized background image B(x, y) to extract non-particle disturbance features, including morphological disturbance feature Fmo, light field anomaly index Fir, and liquid adhesion time drift factor Fvs;

[0151] The morphological disturbance feature Fmo is obtained through the following formula:

[0152]

[0153] In the formula, represents the set of boundary pixels of the region, represents the boundary gradient, and Nb represents the number of boundary pixel points;

[0154] The light field anomaly index Fir is obtained through the following formula:

[0155]

[0156] Wherein, Ωh represents the high-frequency region in the difference image, and σ 2 ωx,y represents the variance of the gray values of the pixels within the window.

[0157] The time-series adhesion analysis unit tracks the abnormally high difference state of each pixel point in the image sequence It(x, y) within a continuous time window, determines whether there is an abnormal structure that has not changed for a long time, and obtains the liquid adhesion time drift factor Fvs;

[0158] The liquid adhesion time drift factor Fvs is obtained through the following formula:

[0159]

[0160] Wherein, Ωs represents the area of pixel points where high-frequency disturbances appear within the past sliding window, and r(x, y) represents the maximum time length for which the pixel point (x, y) has been continuously in a disturbed state;

[0161] The obtained morphological disturbance feature Fmo, light field anomaly index Fir, and liquid adhesion time drift factor Fvs are fitted to obtain the disturbance feature set Rdr.

[0162] In this embodiment, the structural feature extraction unit constructs a difference image between the image sequence and the temperature and illumination normalized background image, and further extracts information such as boundary gradient, pixel fluctuation, and brightness contrast, comprehensively analyzes the morphological features, optical anomalies, and local intensity changes of non-particle disturbances in the image, and for the first time introduces the image microstructure disturbance behavior into the pre-processing process of particle recognition, enhancing the observability of the disturbance source. By calculating the morphological disturbance feature Fmo and the light field anomaly index Fir, this module can not only identify the disturbance structures similar to particles in shape, but also identify the asymmetric local high-frequency bright spots generated by oil film fluctuations and illumination interference. This dual-channel recognition mechanism can avoid the problem of misjudgment of single brightness or edge features in traditional systems, and significantly improve the identification accuracy of the system for complex visual interference.

[0163] With the help of the timing adhesion analysis unit, in this embodiment, the disturbance state of a single pixel point in multiple frames can be continuously tracked, and abnormal areas that persist and have unchanged positions can be identified, such as bubble attachment, oil film residue, etc. Such structures usually cannot be recognized by a single-frame image, but can be effectively excluded by identifying the dynamic behavior characteristics of a sliding time window, which reflects the anti-interference ability of the system in the time dimension. In this embodiment, three types of features - morphological features, light field anomalies, and time adhesion - are finally fused into a unified disturbance feature set to drive subsequent modules to perform differential map purification processing. This pre-structured "feature fitting - response control" provides a data basis for differential denoising and establishes a logical closed-loop from interference recognition to interference elimination.

[0164] Through comprehensive judgment of multi-dimensional features of structure, optics, and timing, it effectively avoids the problem of false particle recognition caused by oil films, light spots, adhered bubbles, etc., enabling subsequent particle concentration calculation to be based on a cleaner and more reliable image. This module completes disturbance recognition before particle recognition, directly blocking the transmission of artifacts in the data chain and avoiding abnormal fluctuations in concentration caused by system misjudgment, improving the credibility and industrial availability of the overall system output.

[0165] During the processing, uncontrollable interferences such as temperature changes, lighting reflections, and cross-mixing of lubricants occur frequently, and it is difficult for traditional methods to generalize and identify them. However, this module can adaptively identify various interference forms through time continuity judgment and multi-dimensional feature fusion mechanism, enhancing the stable operation ability of the system in high-complexity scenarios. The unified output disturbance feature set is not only used for internal processing of this module, but also provides a quantifiable and determinable analysis basis for subsequent dynamic differential denoising modules, establishing an efficient linkage logic of interference recognition - image purification - particle detection and improving the intelligent level of the overall system processing closed-loop.

[0166] Embodiment 4

[0167] This embodiment is an explanatory description based on Embodiment 3. Please refer to Figure 3 and Figure 4 , specifically: The dynamic differential denoising analysis module includes a disturbance fusion discrimination unit and an adaptive purification map construction unit;

[0168] The disturbance fusion discrimination unit compares the obtained disturbance feature set Rdr with the preset morphological disturbance threshold Tmo, light field anomaly threshold Tir, and adhesion time drift threshold Tvs, obtains the disturbance significance function Ψ, and determines whether there is a large-area interference phenomenon in the current frame, and constructs a disturbance mask matrix M(x, y);

[0169] The disturbance significance function Ψ is obtained through the following formula:

[0170]

[0171] In the formula, λ1, λ2, and λ3 respectively represent the preset weight values of the morphological perturbation feature Fmo, the light field anomaly index Fir, and the liquid adhesion time drift factor Fvs;

[0172] Specific example:

[0173] The preset morphological perturbation threshold Tmo = 1.0;

[0174] The light field anomaly threshold Tir = 1.0;

[0175] The adhesion time drift threshold Tvs = 1.0;

[0176] λ1, λ2, and λ3 are 0.4, 0.35, and 0.25 respectively;

[0177] Calculate and obtain the perturbation significance function Ψ:

[0178]

[0179] Table 1: Perturbation significance function acquisition table;

[0180]

[0181] The perturbation mask matrix M(x, y) is obtained through the following formula:

[0182]

[0183] In the formula, Δt(x, y) represents the difference map, and θd represents the local perturbation response judgment threshold;

[0184] When the value of the perturbation mask matrix M(x, y) is 1, it represents a normal region;

[0185] When the value of the perturbation mask matrix M(x, y) is 0, it represents a perturbation elimination region.

[0186] The adaptive purification map construction unit analyzes the obtained perturbation mask matrix M(x, y) to generate a purification map D(x, y), and retains the real particle boundary through the edge enhancement convolution kernel algorithm;

[0187] The purification map D(x, y) is obtained through the following formula:

[0188] D(x,y) = M(x,y) * Δt(x,y);

[0189] Filter the obtained purification map D(x, y) to retain the real particle boundary;

[0190] The filtering formula is as follows:

[0191] D(x,y)←D(x,y)+η*D(x,y)-GD(D(x,y));

[0192] Where η represents the edge enhancement coefficient and GD represents Gaussian kernel smoothing.

[0193] In this embodiment, a disturbance fusion discrimination unit is proposed, which forms a disturbance significance function by weighted fusion of morphological disturbance features, light field anomaly index and adhesion time drift factor, and further generates a disturbance mask matrix. This disturbance identification method can not only consider the coupling effect of multiple interference factors, but also comprehensively determine whether the local disturbance has "non-granular properties", thereby achieving more accurate interference discrimination and shielding.

[0194] This embodiment uses a perturbation mask matrix to perform pixel-level judgment on the differential image, and can dynamically identify credible areas and perturbation elimination areas in the image, thereby retaining real particle signals and eliminating non-particle structures. This image selection processing method based on the discrimination mask provides a logically transparent and highly adaptive control mechanism for particle image purification. After image purification, traditional image processing often uses simple smoothing operations, which can easily cause blurred edges or unclear boundaries of small particles. This embodiment introduces an edge enhancement convolution kernel, combined with Gaussian filtering and edge enhancement adjustment, to retain and enhance the edges of real particles, thereby improving the accuracy of subsequent particle recognition and counting, and is particularly suitable for identifying high-density small particle groups or particle clusters with blurred edges.

[0195] Through the multi-source discrimination mechanism constructed by the perturbation significance function, the system can identify interference areas such as reflection, oil film, and attached foam in real time, effectively preventing these areas from being misjudged as particle signals, and improving the accuracy and credibility of particle concentration estimation; the perturbation mask mechanism realizes the transformation of image purification processing from overall static processing to regional dynamic control, enabling the system to automatically identify and process complex disturbances without manually setting a large number of boundary conditions or manual calibration, greatly improving the intelligence of system operation. The edge enhancement mechanism retains and strengthens the edge structure of real particles, especially for particles with blurred edges, small size, and weak light scattering. It has a good detection enhancement effect, which helps the system to fully perceive the real distribution of particles and improve the integrity and continuity of overall particle recognition. Whether under conditions of uneven lighting, frequent light field disturbances, or drastic changes in liquid level, the system can maintain the ability to accurately extract particle targets through dynamic mask and edge adjustment strategies to ensure stable output of concentration detection.

[0196] Example 5

[0197] This embodiment is explained in Example 4. Please refer to Figure 3, specifically: the particle recognition and concentration estimation module receives the purification image D(x, y), and uses connected component analysis to segment the particles, extracts the particle regions one by one, and obtains the area Ak and average scattering intensity Sk of particle k;

[0198] Set each independent region as and satisfy:

[0199] Rk = {(x, y) ∣ D(x, y) > Tlx};

[0200] In the formula, Tlx represents the pixel brightness threshold;

[0201] The area Ak of particle k is obtained by the following formula:

[0202]

[0203] The average scattering intensity Sk is obtained by the following formula:

[0204]

[0205] In the formula, ωxy represents the weighting factor of the pixel pair for the symmetry of the scattering center;

[0206] Map the obtained area Ak and average scattering intensity Sk of each particle k to the particle concentration value Cest per unit volume;

[0207] The concentration value Cest is obtained by the following formula:

[0208]

[0209] In the formula, V represents the liquid volume corresponding to the imaging, M represents the total number of effective particles, β represents the non-linear particle scattering correction index, and ln represents the logarithmic function.

[0210] The concentration anomaly feedback and self-correction module detects the concentration values Cest in different periods, obtains the concentration offset coefficient δc, and compares it with the preset concentration coefficient threshold Thc to judge the concentration state;

[0211] The concentration offset coefficient δc is obtained by the following formula:

[0212]

[0213] In the formula, Cw represents the concentration mean value, σw represents the concentration standard deviation, and E represents a positive constant to prevent division by zero;

[0214] Judge the concentration state by matching through the following method:

[0215] When the concentration offset coefficient δc ≤ the concentration coefficient threshold Thc, it indicates that the concentration state is normal;

[0216] When the concentration offset coefficient δc > the concentration coefficient threshold Thc, it indicates that the concentration state is abnormal;

[0217] When the concentration state is abnormal, a concentration correction factor Kc is introduced to automatically compensate the concentration value Cest to obtain the concentration correction value Cadj;

[0218] The concentration correction factor Kc is obtained through the following formula:

[0219]

[0220] In the formula, ka represents the basic correction coefficient, pr represents the particle volume deviation rate adjustment factor, Ak represents the area of particle k, pA represents the average value of the area, and β represents the non-linear particle scattering correction index;

[0221] The concentration correction value Cadj is obtained through the following formula:

[0222]

[0223] In this embodiment, by setting a brightness threshold to determine the particle regions in the purification image, and using connected component analysis to extract the particle regions one by one, it is ensured that each particle has closure and structural integrity, avoiding problems such as regional fracture, adhesion, and mis-segmentation that occur in traditional global grayscale difference recognition, and improving the accuracy and independence of particle contour recognition. The average scattering intensity of each particle is introduced, combined with its geometric area, and the concentration is calculated through a non-linear mapping model. This composite parameter estimation method more realistically reflects the influence of particle size distribution on the concentration, making the measurement results more rigorous in terms of physical meaning, especially having advantages in the presence of a large number of small particles or irregularly shaped particles.

[0224] By statistically analyzing the concentration values within multiple measurement cycles and calculating the offset coefficient, the system can determine whether the current concentration value is in a fluctuating abnormal state. This judgment mechanism effectively avoids the interference of short-term particle clusters or instantaneous illumination anomalies on concentration estimation, and improves the steady-state judgment ability of the system. When the system determines that the concentration state is abnormal, this embodiment introduces a concentration correction factor based on the degree of particle area deviation to automatically compensate the original concentration estimation value. This correction model simultaneously considers the area distribution, deviation rate, and non-linear correction coefficient of the particles, has high self-adaptability and system self-healing ability, and ensures the continuity and credibility of the concentration output.

[0225] The connected component analysis method ensures the spatial integrity and separability of the particle regions, reduces recognition errors caused by adhesion or noise spots, and improves the basic quality of particle parameter extraction. The fusion of light scattering intensity and particle area not only enhances the perception accuracy of complex particle structures but also improves the responsiveness of concentration estimation to changes in particle physical properties, showing particular superiority in scenarios with coexisting particles of multiple sizes or significant differences in scattering characteristics. The concentration offset monitoring mechanism effectively identifies trend offsets or short-term anomalies during system operation, enabling the system to have a feed-forward anomaly monitoring capability and reducing unnecessary maintenance or downtime operations.

[0226] Example 6

[0227] For the method of measuring the concentration of solid particles in cutting fluid, please refer to Figure 2 , specifically: it includes the following steps:

[0228] Step 1: The original image acquisition module acquires the cutting fluid image stream within a unit time through a sensor and an imaging array, obtains the image sequence It(x, y), and acquires the temperature T, illumination intensity L, and pixel mean μI attached to the image sequence It(x, y).

[0229] Step 2: The multi-frame temporal fusion and background modeling module performs background modeling on the image sequence It(x, y) based on the time sliding window Tw, extracts the stable structure signal, and obtains the temperature-illumination normalized background image B(x, y).

[0230] Step 3: The perturbation recognition and feature parameter extraction module performs feature extraction on the obtained image sequence It(x, y) and the temperature-illumination normalized background image B(x, y), extracts the non-particle perturbation features, and fits them into the perturbation feature set Rdr.

[0231] Step 4: The dynamic differential denoising analysis module analyzes the temperature-illumination normalized background image B(x, y) based on the perturbation feature set Rdr as a judgment basis, and obtains the purified image D(x, y).

[0232] Step 5: The particle recognition and concentration estimation module performs particle recognition and counting based on the obtained purified image D(x, y), and calculates and obtains the concentration value Cest.

[0233] Step 6: The concentration anomaly feedback and self-correction module detects the obtained concentration value Cest, obtains the concentration offset coefficient δc, compares it with the preset concentration coefficient threshold Thc, determines whether it is abnormal, and automatically performs correction to obtain the concentration correction value Cadj.

[0234] In this embodiment, by synchronously collecting temperature, illuminance, and the average pixel brightness of the image while collecting the image sequence, the system can sense and model non-particle factors such as light fluctuations and thermal environment changes, avoiding interference with the image structure stability caused by environmental changes and breaking the limitation of relying solely on single-image information in the traditional method. By extracting the background structure through multi-frame temporal fusion and combining with the temperature-illuminance normalization operation, the background image can adapt to the imaging conditions under the current working conditions, significantly enhancing the robustness of the system in interference scenarios such as light variation and flow fluctuation, and constructing a stable and reliable "reference baseline" image.

[0235] By extracting structural features, light field anomalies, and temporal adhesion behaviors to form a perturbation feature set, it provides a numerical basis for interference recognition for the system, which can accurately distinguish pseudo-particle signals such as bubbles, oil films, and reflections, and solve the key problem of frequent misjudgments in traditional image recognition. Setting up a perturbation judgment mechanism and mask image construction enables the system to automatically eliminate interference regions according to spatial regions, only retaining the credible particle response image, establishing an image purification process combining dynamic difference and edge enhancement, and improving the accuracy and integrity of subsequent particle recognition. By determining the concentration state through the concentration offset coefficient and introducing a concentration correction factor for automatic compensation, the system has the ability to autonomously identify anomalies and automatically and stably output, effectively coping with valuation deviations caused by environmental mutations, sudden increases in particles, etc., and significantly improving the stability and reliability of the system during industrial continuous operation.

[0236] A multi-dimensional constraint mechanism and dynamic adjustment means are constructed from the source data collection to the final estimation output, avoiding common interference misjudgments and data jump phenomena, and ensuring that the concentration data is continuous, stable, and close to the real value. For common situations such as uneven illumination, liquid reflection, and temperature fluctuation in the factory workshop, this method incorporates compensation and recognition mechanisms in multiple steps, capable of adapting to various non-ideal environments, ensuring the stability of the measurement results and the robustness of the system operation.

[0237] Through the concentration anomaly detection and automatic correction process, the system can maintain the credibility of the data output by itself, avoiding behaviors such as incorrect liquid replacement and incorrect maintenance caused by frequent false alarms or misjudgments, and reducing the operation cost and management burden.

[0238] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A cutting fluid solid particle concentration measurement system, characterized in that: It includes an original image acquisition module, a multi-frame time series fusion and background modeling module, a disturbance recognition and feature parameter extraction module, a dynamic differential denoising analysis module, a particle recognition and concentration estimation module, and a concentration anomaly feedback and self-correction module; The original image acquisition module acquires the cutting fluid image stream within a unit time through a sensor and an imaging array, obtains the image sequence It(x, y), and acquires the temperature T, illumination intensity L, and pixel mean μI attached to the image sequence It(x, y); The multi-frame time series fusion and background modeling module performs background modeling on the image sequence It(x, y) based on the time sliding window Tw, extracts the stable structure signal, and obtains the temperature-illumination normalized background image B(x, y); The disturbance recognition and feature parameter extraction module extracts features from the obtained image sequence It(x, y) and the temperature-illumination normalized background image B(x, y), extracts non-particle disturbance features, and fits them into the disturbance feature set Rdr; The dynamic differential denoising analysis module analyzes the temperature-illumination normalized background image B(x, y) based on the disturbance feature set Rdr as a judgment basis, and obtains the purified image D(x, y); The particle recognition and concentration estimation module performs particle recognition and counting based on the obtained purified image D(x, y), and calculates and obtains the concentration value Cest; The concentration anomaly feedback and self-correction module detects the obtained concentration value Cest, obtains the concentration offset coefficient δc, compares it with the preset concentration coefficient threshold Thc, determines whether it is abnormal, and automatically performs correction to obtain the concentration correction value Cadj.

2. The cutting fluid solid particle concentration measurement system according to claim 1, wherein: The original image acquisition module includes an imaging data acquisition unit and an auxiliary environment parameter synchronous acquisition unit; The imaging data acquisition unit takes pictures of the cutting fluid at a fixed acquisition window through the area image sensor in the industrial-grade CMOS area array camera, cooperated with the hyperspectral backlight source, acquires the cutting fluid image stream within a unit time, and constitutes a continuous frame image sequence It(x, y); Among them, the size of the image sequence It(x, y) is H*W, where H represents the height of the image sequence It(x, y), and W represents the width of the image sequence It(x, y); The construction formula of the image sequence It(x, y) is: It(x, y), x∈[1,W], y∈[1,H]; The auxiliary environment parameter synchronous acquisition unit synchronously records the temperature T, illumination intensity L, and pixel mean μI during the image acquisition process; Among them, the temperature T is obtained by detecting the temperature through the integrated thermistor digital temperature sensor attached to the outer shell of the image acquisition window; The illumination intensity L is obtained by measuring the illumination intensity of the imaging area using a high-precision digital photosensitive sensor; The pixel mean μI is obtained through the following formula:

3. The cutting fluid solid particle concentration measurement system according to claim 2, characterized in that: The multi-frame time series fusion and background modeling module includes a sliding window image fusion unit and a temperature-illumination normalization correction unit; The sliding window image fusion unit traces back the image frame sequence {It-i(x, y)} within the time sliding window Tw from the current moment, i = 0, 1,..., Tw-1, and performs weighted fusion to obtain the preliminary background image Bo(x, y); The preliminary background image Bo(x, y) is obtained through the following formula: Wherein, Z represents a normalization factor, It, and i(x, y) represent the image sequence It(x, y) of the i-th frame within the time sliding window Tw, and ωi represents the preset weight value of the image sequence It(x, y) of the i-th frame; The temperature-illumination normalization correction unit collects the temperature T and illumination intensity L corresponding to each frame of the image sequence It(x, y), establishes the optothermal normalization factor GRt, performs normalization correction on the preliminary background image Bo(x, y), and obtains the temperature-illumination normalized background image B(x, y); The optothermal normalization factor GRt is obtained through the following formula: Wherein, αT represents the sensitivity factor of temperature, αL represents the sensitivity factor of illumination, To represents the set reference temperature, and Lo represents the set reference illumination intensity; The temperature-illumination normalized background image B(x, y) is obtained through the following formula: B(x,y) = Bo(x,y) * GRt.

4. The cutting fluid solid particle concentration measurement system according to claim 3, wherein: The disturbance recognition and feature parameter extraction module includes a structural feature extraction unit and a temporal adhesion analysis unit; The structural feature extraction unit receives the image sequence It(x, y) and the temperature-illumination normalized background image B(x, y), performs analysis, and obtains the difference map Δt(x, y); The difference map Δt(x, y) is obtained through the following formula: Δt(x,y) = |It(x,y) - B(x,y)|; Feature extraction is performed on the image sequence It(x, y) and the temperature-illumination normalized background image B(x, y) to extract non-particle disturbance features, including morphological disturbance feature Fmo, light field anomaly index Fir, and liquid adhesion time drift factor Fvs; The morphological disturbance feature Fmo is obtained through the following formula: In the formula, represents the set of regional boundary pixels, represents the boundary gradient, and Nb represents the number of boundary pixel points; The light field anomaly index Fir is obtained through the following formula: where, Ωh represents the high-frequency region in the difference image, and σ 2 ωx,y represents the variance of the gray values of the pixels within the window.

5. The cutting fluid solid particle concentration measurement system according to claim 4, wherein: The temporal adhesion analysis unit tracks the abnormally high difference state of each pixel point in the image sequence It(x, y) within the continuous time window, determines whether there is an abnormal structure that has not changed for a long time, and obtains the liquid adhesion time drift factor Fvs; The liquid adhesion time drift factor Fvs is obtained through the following formula: Wherein, Ωs represents the pixel point area where high-frequency disturbances occur within the past sliding window, and r(x, y) represents the maximum time length for which the pixel point (x, y) is continuously in a disturbed state; The obtained morphological disturbance feature Fmo, light field anomaly index Fir, and liquid adhesion time drift factor Fvs are fitted to obtain the disturbance feature set Rdr.

6. The cutting fluid solid particle concentration measurement system according to claim 5, wherein: The dynamic difference denoising analysis module includes a disturbance fusion discrimination unit and an adaptive purification map construction unit; The disturbance fusion discrimination unit compares the obtained disturbance feature set Rdr with the preset morphological disturbance threshold Tmo, light field anomaly threshold Tir, and adhesion time drift threshold Tvs, obtains the disturbance significance function Ψ, and determines whether there is a large-area interference phenomenon in the current frame, and constructs the disturbance mask matrix M(x, y); The disturbance significance function Ψ is obtained through the following formula: Wherein, λ1, λ2, and λ3 respectively represent the preset weight values of the morphological disturbance feature Fmo, light field anomaly index Fir, and liquid adhesion time drift factor Fvs; The disturbance mask matrix M(x, y) is obtained through the following formula: Where, Δt(x, y) represents the difference image, and θd represents the local disturbance response judgment threshold; When the value of the disturbance mask matrix M(x, y) is 1, it represents the normal area; When the value of the disturbance mask matrix M(x, y) is 0, it represents the disturbance elimination area.

7. The cutting fluid solid particle concentration measurement system according to claim 6, characterized in that: The adaptive purification map construction unit analyzes the obtained disturbance mask matrix M(x, y), generates the purification map D(x, y), and retains the real particle boundary through the edge enhancement convolution kernel algorithm; The purification map D(x, y) is obtained through the following formula: D(x,y) = M(x,y) * Δt(x,y); Filter the obtained purification map D(x, y) to retain the real particle boundary; The filtering formula is as follows: D(x,y) ← D(x,y) + η * D(x,y) - GD(D(x,y)); Where, η represents the edge enhancement coefficient, and GD represents Gaussian kernel smoothing.

8. The cutting fluid solid particle concentration measurement system according to claim 7, wherein: The particle recognition and concentration estimation module receives the purification map D(x, y), and uses connected component analysis to segment particles, extracts the particle regions one by one, and obtains the area Ak and average scattering intensity Sk of particle k; Set each independent region to Purify the graph D(x, y), and satisfy: Rk = {(x, y) | D(x, y) > Tlx}; Where, Tlx represents the pixel brightness threshold; The area Ak of particle k is obtained through the following formula: The average scattering intensity Sk is obtained through the following formula: Where, ωxy represents the weighting factor of the pixel pair's symmetry to the scattering center; Map the obtained area Ak and average scattering intensity Sk of each particle k to the particle concentration value Cest per unit volume; The concentration value Cest is obtained through the following formula: Where, V represents the liquid volume corresponding to the imaging, M represents the total number of effective particles, β represents the non-linear particle scattering correction index, and ln represents the logarithmic function.

9. The cutting fluid solid particle concentration measurement system according to claim 8, characterized in that: The concentration anomaly feedback and self-correction module detects the concentration values Cest in different periods, obtains the concentration offset coefficient δc, and compares it with the preset concentration coefficient threshold Thc to judge the concentration state; The concentration offset coefficient δc is obtained through the following formula: Where, Cw represents the concentration mean, σw represents the concentration standard deviation, and E represents a positive constant to prevent division by zero; Judge the concentration state through the following matching method: When the concentration offset coefficient δc ≤ the concentration coefficient threshold Thc, it means the concentration state is normal; When the concentration offset coefficient δc > the concentration coefficient threshold Thc, it means the concentration state is abnormal; When the concentration state is abnormal, introduce the concentration correction factor Kc to automatically compensate the concentration value Cest to obtain the concentration correction value Cadj; The concentration correction factor Kc is obtained through the following formula: Where, ka represents the basic correction coefficient, pr represents the particle volume deviation rate adjustment factor, Ak represents the area of particle k, pA represents the average value of the area, and β represents the non-linear particle scattering correction index; The concentration correction value Cadj is obtained through the following formula:

10. A method for measuring the concentration of solid particles in a cutting fluid, which is applied to the cutting fluid solid particle concentration measurement system according to any one of claims 1 to 9, and is characterized in that: Including the following steps: Step 1: The original image acquisition module collects the cutting fluid image stream within a unit time through sensors and imaging arrays, obtains the image sequence It(x, y), and collects the temperature T, illumination intensity L, and pixel mean μI attached to the image sequence It(x, y). Step 2: The multi-frame temporal fusion and background modeling module performs background modeling on the image sequence It(x, y) based on the time sliding window Tw, extracts the stable structure signal, and obtains the temperature-illumination normalized background image B(x, y). Step 3: The disturbance recognition and feature parameter extraction module extracts features from the obtained image sequence It(x, y) and the temperature-illumination normalized background image B(x, y), extracts the non-particle disturbance features, and fits them into the disturbance feature set Rdr. Step 4: The dynamic differential denoising analysis module analyzes the temperature-illumination normalized background image B(x, y) based on the disturbance feature set Rdr as the judgment basis, and obtains the purified image D(x, y). Step 5: The particle recognition and concentration estimation module performs particle recognition and counting based on the obtained purified image D(x, y), and calculates to obtain the concentration value Cest. Step 6: The concentration anomaly feedback and self-correction module detects the obtained concentration value Cest, obtains the concentration offset coefficient δc, compares it with the preset concentration coefficient threshold Thc, determines whether it is abnormal, and automatically performs correction to obtain the concentration correction value Cadj.

Images