A filter press coal cake residual detection method and system based on visual recognition

Abstract

The application discloses a filter press coal cake residual detection method and system based on visual identification: a polarization phase modulation imaging system is used for dynamic polarization imaging of the surface of a filter plate, image sequences are arranged in time sequence to form time sequence data of each pixel point, frequency domain analysis is performed on the time sequence data pixel by pixel, polarization characteristic parameter sets are extracted from the frequency domain signals, spatial frequency analysis is performed on the polarization characteristic parameter sets, a characteristic image is generated according to surface scattering characteristic parameters, a phase-sensitive detection method is used for signal enhancement processing of the time sequence data, polarization characteristic parameters are calculated according to the enhanced time sequence data, pixels of the characteristic image are classified according to the enhanced polarization characteristic parameter sets, each pixel point is classified as coal cake residual or non-residual, a binary residual distribution map is generated, and a filter press coal cake residual detection result is generated according to the residual distribution map and the enhanced polarization characteristic parameter sets. The application combines the phase-sensitive detection technology to cumulatively enhance the weak polarization response.

Description

Technical Field

[0001] This application relates to the field of visual inspection, and in particular to a method and system for detecting coal cake residue in a filter press based on visual recognition. Background Technology

[0002] In coal mining operations, filter presses are key equipment for dewatering clean coal, tailings, and raw coal slime. Filter presses apply high pressure to the coal slurry, forcing water through the filter cloth and forming a coal cake on the filter plate surface. During continuous production, it is crucial to promptly remove the coal cake from the filter plates; otherwise, residual coal cake will not only reduce the filtration efficiency of subsequent batches but also accelerate filter plate wear, affecting the equipment's lifespan. Therefore, accurately detecting the amount of residual coal cake on the filter plate surface is essential for ensuring the efficient and stable operation of the filter press.

[0003] However, visual detection of coal cake residue faces unique technical challenges. As a highly absorbent material, coal cake typically has a surface reflectivity of less than 5%, resulting in extremely weak reflected light intensity. In real industrial environments, this weak reflected signal is often drowned out by ambient noise, making it difficult for traditional image recognition methods to effectively distinguish coal cake residue from the filter plate surface. This is especially true for thin-layer coal cake residue, which, due to its small thickness and limited coverage area, produces an even weaker reflected signal, making it almost undetectable under low signal-to-noise ratio conditions.

[0004] Existing detection methods mainly rely on static imaging and simple grayscale thresholding. These methods have significant limitations when dealing with the low reflectivity of coal cakes: firstly, static imaging provides limited information and cannot fully utilize the differences in material polarization characteristics; secondly, in low signal-to-noise ratio environments, weak polarization signals are masked by noise, making polarization feature extraction difficult and resulting in low detection accuracy, particularly failing to reliably identify thin-layer residues that affect filtration efficiency. Therefore, a detection method capable of effectively extracting the polarization features of coal cakes under low signal-to-noise ratio conditions is urgently needed. Summary of the Invention

[0005] To address the challenge of accurately extracting polarization features from coal cake, a high-absorbing material, under low signal-to-noise ratio conditions, particularly the difficulty in detecting thin-layer residues, this application provides a visual recognition-based method and system for detecting coal cake residues in filter presses. This method transforms static polarization measurement into a dynamic detection process through continuous polarization modulation, extracts polarization information from the time-domain signal using Fourier analysis, and combines phase-sensitive detection technology to cumulatively enhance the weak polarization response.

[0006] One aspect of this application provides a visual recognition-based method for detecting coal cake residue in a filter press, comprising: S1, performing dynamic polarization imaging on the filter plate surface using a polarization phase modulation imaging system, acquiring an image sequence during the continuous change of polarization state, wherein each image in the image sequence corresponds to a polarization state, and arranging the image sequence in chronological order to form temporal data of each pixel. Where x and y are pixel coordinates, and t is time; S2, for time-series data Pixel-by-pixel frequency domain analysis is performed to convert the time-domain signal into a frequency-domain signal, and a set of polarization feature parameters is extracted from the frequency-domain signal. Among them, the set of polarization characteristic parameters Including polarization intensity parameters and polarization phase parameters S3, for the set of polarization characteristic parameters Spatial frequency analysis is performed to calculate the power spectral density distribution of polarization parameters. Surface scattering characteristic parameters are extracted based on the power spectral density distribution, and a feature image is generated based on these parameters. Feature Image The pixel value reflects the surface scattering characteristics of the corresponding position on the filter plate surface; S4, a phase-sensitive detection method is used to analyze the time-series data. Perform signal enhancement processing to generate enhanced timing data. Based on enhanced time-series data Calculate the polarization characteristic parameters to obtain the enhanced polarization characteristic parameter set. S5, based on the enhanced polarization characteristic parameter set For feature images Pixel classification is performed, classifying each pixel as either coal cake residue or non-residue, generating a binarized residue distribution map. S6, based on the residual distribution map and enhanced polarization characteristic parameter set The results of coal cake residue detection in the filter press are generated.

[0007] The polarization phase modulation imaging system consists of a polarization modulator (such as an electro-optic modulator), a light source, an imaging device (high-speed camera), and a synchronization control unit. This system can continuously modulate the polarization state of a light beam incident on the filter surface, causing the polarization angle to change over time according to a preset function (such as a sine function), while simultaneously acquiring images of reflected light under different polarization states. Unlike traditional static polarization imaging, this system transforms polarization measurement from a single static acquisition into a continuous dynamic process.

[0008] Dynamic polarization imaging is an imaging method that continuously changes the polarization state of incident light while simultaneously acquiring a sequence of images. In this scheme, the polarization angle changes continuously within the range of 0° to 180° according to a sinusoidal law. N images are acquired within one modulation period, each image corresponding to a specific polarization angle. This dynamic process causes the grayscale value of each pixel to form a periodic time series as the polarization state changes, providing a foundation for subsequent frequency domain analysis and weak signal extraction.

[0009] polarization intensity parameters This characterizes the ability of a material surface to retain polarized light. In this scheme, the calculation formula is obtained by performing a Fourier transform on the time-series data. ,in, For the fundamental frequency component, This is the DC component. The value range is [0, 1]. Due to its rough surface structure, the coal cake exhibits a severe bias in its polarization solution. The value is low (typically <0.35); while a smooth filter plate surface can better maintain the polarization state. The value is relatively high (usually >0.6).

[0010] The polarization phase parameter φ(x, y) characterizes the phase delay of the polarization direction of the reflected light relative to the incident light, and is calculated using the following formula: The value ranges from [-π, π]. This parameter reflects the microstructural characteristics of the material surface. Due to its irregular surface structure, the phase parameter of the coal cake residue changes drastically in local areas, exhibiting a high phase standard deviation; while the phase parameter of the filter plate surface is relatively stable.

[0011] Feature Image A comprehensive feature image that incorporates surface scattering properties, obtained through a formula. Generate, where, For scattering intensity, Roughness index This is the normalized value of the roughness index. Each pixel value in this image quantitatively characterizes the surface scattering properties at the corresponding location. The coal cake residue area exhibits a high feature value due to its high roughness, which contrasts sharply with the clean filter plate with its low feature value.

[0012] The phase-sensitive detection method generates a reference signal that is in phase and frequency with the polarization modulation, performs correlation operations and low-pass filtering on the time-series data, and selectively extracts the signal component that matches the modulation frequency. In this scheme, the method can extract weak periodic polarization response signals from strong noise backgrounds, amplify the polarization characteristics of the coal cake that were originally submerged in noise, and significantly improve the signal-to-noise ratio.

[0013] Furthermore, S1, dynamic polarization imaging is performed on the filter surface using a polarization phase modulation imaging system, acquiring image sequences during the continuous change of polarization state. This includes: periodically modulating the polarization state of the incident light illuminating the filter surface using a polarization modulator in the polarization phase modulation imaging system, so that the polarization angle of the incident light follows a preset function. The image changes continuously with time t; the imaging device in the polarization phase modulation imaging system synchronously acquires images of the reflected light from the filter surface. The image acquisition and polarization modulation of the imaging device are synchronized in time. N images are acquired within one modulation period to form an image sequence, where the i-th image corresponds to the polarization angle. Extract the pixel grayscale values ​​at the same spatial location (x, y) from each image in the image sequence. Arrange the N grayscale values ​​according to the time sequence of image acquisition to form a one-dimensional time series of the corresponding pixel locations, which serves as the temporal data for each pixel. .

[0014] Furthermore, S2, for time series data Pixel-by-pixel frequency domain analysis is performed to convert the time-domain signal into a frequency-domain signal, and a set of polarization feature parameters is extracted from the frequency-domain signal. This includes: time series data The discrete Fourier transform is performed on the one-dimensional time series of each pixel position (x, y) to convert the time-domain signal into a frequency-domain signal. Where ω is the frequency; from the frequency domain signal Extract the frequency components related to the polarization modulation frequency, including: DC component. Characterized by average light intensity; fundamental frequency component ,in, The fundamental frequency for polarization modulation; based on the DC component. and fundamental frequency component Calculate polarization intensity parameters The calculation formula is: Based on fundamental frequency components Calculating polarization phase parameters in complex form The calculation formula is: ;

[0015] In particular, through time series data By performing a discrete Fourier transform, the weak polarization response that was originally submerged in broadband noise is converted into a specific frequency in the frequency domain. The sharp peak at that point. Since polarization modulation is periodic, the polarization response of the coal cake also exhibits the same periodicity, manifesting as a fundamental frequency component in the frequency domain. Random noise, on the other hand, exhibits a continuous distribution in the frequency domain without any specific frequency clustering. This frequency domain separation allows for accurate identification and extraction of polarization signals even when the signal-to-noise ratio in the time domain is -10dB.

[0016] In addition, extract specific frequency components. The process is essentially an ideal narrowband filter with a bandwidth of only 1 / N (where N is the number of sampling points). Compared to wideband measurements in the time domain, this narrowband extraction reduces noise power by a factor of N. Simultaneously, due to the fundamental frequency component... It contains polarization response information throughout the entire modulation period, which is equivalent to coherently accumulating multiple measurements, thus obtaining the signal strength. A 100-fold improvement. For a typical N=100 sampling, the signal-to-noise ratio can be improved by up to 20dB.

[0017] Finally, through calculation The amplitude of the polarization response is normalized to the average light intensity, eliminating the influence of uneven illumination. This normalization process allows for accurate characterization of the polarization properties even under conditions of extremely weak reflected light from the coal cake. The phase parameter... The extraction of phase information utilizes the characteristic of complex Fourier transform to preserve phase information, providing a quantitative basis for subsequent identification of phase fluctuations caused by irregular surfaces of coal cakes.

[0018] Furthermore, S3, for the set of polarization characteristic parameters Spatial frequency analysis is performed to calculate the power spectral density distribution of polarization parameters. Surface scattering characteristic parameters are extracted based on the power spectral density distribution, and a feature image is generated based on these parameters. Feature Image The pixel values ​​reflect the surface scattering characteristics at the corresponding location on the filter plate surface, including: the set of polarization feature parameters. Perform spatial frequency domain transformation to convert the polarization intensity parameters and polarization phase parameters The spatial spectrum is obtained by transforming from the spatial domain to the frequency domain; the power spectral density distribution is then calculated based on the spatial spectrum. Power spectral density distribution quantifies the energy distribution of polarization parameters at different spatial frequencies, reflecting the texture and roughness characteristics of the material surface. Statistical analysis of the power spectral density distribution extracts characteristic parameters characterizing surface scattering properties. Based on these surface scattering characteristic parameters, a feature image is constructed. By performing local analysis on the neighborhood of each pixel, the extracted scattering feature parameters are mapped to pixel values, thus improving the feature image. Each pixel value quantitatively characterizes the surface scattering properties at the corresponding location;

[0019] In particular, although the reflected light from coal cake residue is weak, its rough surface structure exhibits a unique power spectrum distribution in the spatial frequency domain. Through calculation... This process converts the spatial variation of polarization parameters into energy distribution. Even if the original polarization signal is weak, its spatial variation pattern will still form specific distribution characteristics in the power spectrum. The irregular surface of the coal cake generates a broadband power spectrum, while the smooth filter plate mainly concentrates in the low-frequency region.

[0020] Furthermore, statistical analysis is performed on the power spectral density distribution to extract characteristic parameters characterizing the surface scattering properties, including: a set of polarization characteristic parameters. Local windowing is performed, and analysis windows of a preset size slide in the spatial domain with a set step size. Frequency domain transformation is applied to the local polarization parameters at each window location to generate a local power spectral density distribution corresponding to the spatial location. Radial statistical analysis is then performed on each local power spectral density distribution, converting the two-dimensional power spectrum into a one-dimensional radial power spectrum through angle integration to eliminate direction dependence. A power-law model based on surface scattering theory is used to fit the radial power spectrum parameters, extracting characteristic parameters representing local scattering properties from the fitting results, including scattering intensity parameters and roughness index, establishing a mapping relationship between each window location and its local characteristic parameters. Finally, spatial interpolation methods are used to extend the characteristic parameters of discrete window locations to the entire image space, generating continuously distributed characteristic parameters so that each pixel location obtains parameter values ​​reflecting the scattering characteristics of its neighborhood.

[0021] Specifically, a sliding window is used for local frequency domain analysis, avoiding the signal dilution problem caused by global analysis. For thin coal cake residues, which cover a small area, a full-image Fourier transform would cause weak local features to be masked by a large background. By using a w×w local window, the power spectrum at each location only reflects the characteristics of its neighborhood, ensuring accurate characterization even for small-area thin residues. The choice of window size (larger than the minimum residue size but smaller than the typical residue size) ensures high detection sensitivity.

[0022] Furthermore, feature images are constructed based on surface scattering characteristic parameters. By performing local analysis on the neighborhood of each pixel, the extracted scattering feature parameters are mapped to pixel values, thus improving the feature image. Each pixel value quantitatively characterizes the surface scattering properties at the corresponding location, including: obtaining continuously distributed feature parameters, wherein the feature parameters include a scattering intensity parameter distribution. and roughness index distribution ; for roughness index distribution Normalization is performed to obtain the normalized roughness index distribution. ; Distribution of scattering intensity parameters With the normalized roughness index distribution Perform nonlinear fusion to construct a feature image: Where k is a preset enhancement coefficient, which amplifies the influence of roughness differences on eigenvalues ​​through an exponential function;

[0023] Furthermore, in S4, a phase-sensitive detection method is used to analyze the time-series data. Perform signal enhancement processing to generate enhanced timing data. Based on enhanced time-series data Calculate the polarization characteristic parameters to obtain the enhanced polarization characteristic parameter set. This includes: establishing a reference signal synchronized with the polarization state modulation of the polarization modulator. Time sampling and timing data of the reference signal Consistent, the frequency of the reference signal is equal to the fundamental frequency of the polarization modulation. Phase and polarization angle variation function Maintain a fixed relationship; based on a reference signal Time series data is demodulated through coherent demodulation. Signal enhancement processing is performed to obtain enhanced time-series data after noise suppression. ;

[0024] In particular, the core innovation of phase-sensitive detection lies in utilizing the deterministic periodic characteristics of polarization modulation, through comparison with a reference signal. Coherent demodulation achieves selective signal amplification. The physical essence of this process is that although the polarization response of the coal cake is weak, it maintains a strict phase-locked relationship with the polarization modulation, manifesting as a relationship with the modulation frequency. The modulation signal exhibits perfectly synchronized periodic changes; while various types of noise are random and have no phase correlation with the modulated signal. This is achieved by using time-series data... With reference signal By performing point-by-point multiplication and low-pass filtering, only components that are in phase and frequency with the reference signal can be coherently demodulated to generate a DC output, while all other frequency components (including broadband noise) are converted into AC signals and suppressed by the low-pass filter.

[0025] The power of this coherent accumulation mechanism lies in its √N signal-to-noise ratio (SNR) improvement. For a typical 1kHz modulation frequency and an integration time of 0.1 seconds, this is equivalent to coherent accumulation over 100 modulation cycles, theoretically achieving a 10-fold SNR improvement. More importantly, this accumulation is coherent—the signal is linearly superimposed according to a defined phase relationship, while noise, due to its random phase, can only be superimposed according to power, thus achieving selective signal enhancement relative to noise. Experimental data shows that even when the original SNR is as low as -15dB (signal power is only 1 / 30th of the noise), polarization features can still be reliably extracted after phase-sensitive detection.

[0026] For the detection of thin-layer coal cake residue, phase-sensitive detection is of particular significance. The polarization response of a 0.1 mm thick residue may only be a few percentage points higher than that of a noisy substrate, making direct detection almost impossible. However, the narrow-band characteristics of phase-sensitive detection (equivalent bandwidth up to 0.01 Hz) compress the detection bandwidth by five orders of magnitude, correspondingly reducing the noise power to one ten-thousandth of its original value, allowing these extremely weak signals to emerge.

[0027] Furthermore, S5, based on the enhanced polarization characteristic parameter set For feature images Pixel classification is performed, classifying each pixel as either coal cake residue or non-residue, generating a binarized residue distribution map. This includes: extracting feature image values ​​for each pixel location (x, y). Enhance polarization intensity and enhance polarization phase According to the enhanced polarization phase parameters Calculate the local phase stability index, centered on pixel (x, y). Calculate the phase standard deviation within the window This characterizes the phase stability at that position; when and and If the condition is met, it is determined to be coal cake residue; otherwise, it is determined to be non-residue. For the feature image threshold, The polarization intensity threshold. The phase stability threshold is used to assign values ​​to pixels identified as coal cake residue. ; Assign values ​​to pixels determined to be non-residual ;

[0028] Specifically, the rough and irregular surface of the coal cake causes drastic changes in the polarization phase in local areas, while the phase is relatively stable on the smooth surface of the filter plate. Phase stability index It remains stable even under extremely low reflectivity conditions because it reflects relative changes rather than absolute intensity.

[0029] Furthermore, S6, according to the residual distribution map and enhanced polarization characteristic parameter set Generate filter press coal cake residue detection results, including: using a connected component labeling algorithm to... The pixels are divided into regions to obtain independent sets of residual regions. And calculate the geometric properties of each residual region; for each residual region Extract enhanced polarization intensity parameters within the region and enhance polarization phase parameters Calculate the regional polarization characteristic statistics; based on the geometric properties and polarization characteristic statistics of the residual area, generate the filter press coal cake residue detection results.

[0030] Another aspect of this application provides a visual recognition-based filter press coal cake residue detection system, comprising: a polarization phase modulation imaging device for dynamic polarization imaging of the filter plate surface, including a polarization modulator and an imaging device, wherein the polarization modulator periodically modulates the polarization state of the incident light irradiating the filter plate surface, and the imaging device synchronously acquires image sequences of reflected light from the filter plate surface, and arranges the image sequences in chronological order to form temporal data of each pixel. Frequency domain analysis module, for time series data Pixel-by-pixel frequency domain analysis is performed, converting the time-domain signal into a frequency-domain signal using Discrete Fourier Transform, and extracting parameters including polarization intensity from the frequency-domain signal. and polarization phase parameters The set of polarization characteristic parameters ;

[0031] The spatial frequency analysis module analyzes the set of polarization characteristic parameters. Spatial frequency analysis is performed, and surface scattering characteristic parameters are extracted through local windowing and power spectral density calculation, based on scattering intensity parameters. and roughness index Nonlinear fusion to generate feature images The phase-sensitive detection module uses a reference signal synchronized with polarization modulation to detect timing data. Perform coherent demodulation to generate enhanced time-series data. And calculate the set of enhanced polarization characteristic parameters based on the enhanced time series data. Pixel classification module, based on feature images and enhanced polarization characteristic parameter set A multi-dimensional feature vector is constructed, and each pixel is classified as coal cake residue or non-residue using a joint discrimination criterion, generating a binarized residue distribution map. The detection result generation module generates a residual distribution map. Connectivity analysis is performed, and combined with the enhanced polarization characteristic statistics within the region, to generate filter press coal cake residue detection results that include residual spatial distribution and physical properties.

[0032] Compared to existing technologies, the advantages of this application are:

[0033] (1) In traditional static measurement, the reflectivity of coal cake is less than 5%, resulting in extremely weak polarization signal that is almost indistinguishable from noise. However, this application transforms the originally constant weak signal into an alternating signal with a specific frequency through periodic polarization modulation. Even if the signal amplitude of a single measurement is still very small, its periodic variation pattern can be enhanced by the accumulation of multiple modulation cycles.

[0034] (2) The introduction of phase-sensitive detection technology enables the system to extract weak signals of specific frequencies from a background of strong noise. By generating a reference signal synchronized with polarization modulation, the system can selectively amplify the signal component with the same modulation frequency while suppressing noise at other frequencies. This frequency selectivity allows even the extremely weak polarization response generated by a thin layer of coal cake residue with a thickness of only 0.1 mm to be effectively identified. Fourier analysis further transforms the accumulated signal in the time domain to the frequency domain, where the polarization response is represented by a peak at a specific frequency. Even under conditions where the signal-to-noise ratio is as low as -10 dB, polarization characteristic parameters can still be accurately extracted. Attached Figure Description

[0035] This application will be further described by way of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting; in these embodiments, the same reference numerals denote the same structures, wherein:

[0036] Figure 1 This is an exemplary flowchart of a visual recognition-based method for detecting coal cake residue in a filter press, according to some embodiments of this application.

[0037] Figure 2 This is an exemplary flowchart illustrating the generation of timing data for each pixel according to some embodiments of this application;

[0038] Figure 3 This is an exemplary flowchart illustrating the generation of polarization intensity parameters and polarization phase parameters according to some embodiments of this application;

[0039] Figure 4 This is an exemplary flowchart illustrating the construction of a feature image according to some embodiments of this application;

[0040] Figure 5 This is an exemplary flowchart illustrating the detection results of coal cake residue according to some embodiments of this application. Detailed Implementation

[0041] The methods and systems provided in the embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0042] like Figure 1As shown, a polarization phase modulation imaging system is used to perform dynamic polarization imaging on the filter plate surface. Image sequences are acquired during the continuous change of polarization state, with each image in the sequence corresponding to a polarization state. The image sequences are then arranged in chronological order to form temporal data for each pixel. Where x and y are pixel coordinates, and t is time; for time-series data Pixel-by-pixel frequency domain analysis is performed to convert the time-domain signal into a frequency-domain signal, and a set of polarization feature parameters is extracted from the frequency-domain signal. Among them, the set of polarization characteristic parameters Including polarization intensity parameters and polarization phase parameters For the set of polarization characteristic parameters Spatial frequency analysis is performed to calculate the power spectral density distribution of polarization parameters. Surface scattering characteristic parameters are extracted based on the power spectral density distribution, and a feature image is generated based on these parameters. Feature Image The pixel values ​​reflect the surface scattering characteristics at corresponding locations on the filter plate surface; a phase-sensitive detection method is used to analyze the time-series data. Perform signal enhancement processing to generate enhanced timing data. Based on enhanced time-series data Calculate the polarization characteristic parameters to obtain the enhanced polarization characteristic parameter set. Based on the enhanced polarization characteristic parameter set For feature images Pixel classification is performed, classifying each pixel as either coal cake residue or non-residue, generating a binarized residue distribution map. According to the residual distribution map and enhanced polarization characteristic parameter set The results of coal cake residue detection in the filter press are generated.

[0043] like Figure 2 As shown, S1: Dynamic polarization detection is performed on the filter plate surface using a polarization phase modulation imaging system. The polarization phase modulation imaging system includes an electro-optic modulator and a high-speed camera, wherein: the electro-optic modulator modulates the polarization state of the light beam incident on the filter plate surface, so that the polarization angle... The polarization state of the incident light changes continuously over time according to a preset function; the modulation frequency is 1 kHz; the polarization state of the incident light changes continuously within the range of 0° to 180° according to a sinusoidal law; the polarization angle... Where f is the modulation frequency and t is time; a high-speed camera synchronized with the electro-optic modulator acquires images of reflected light from the filter surface, and N images are acquired within one polarization modulation period T to form an image sequence. ; Image sequence The grayscale values ​​of pixels (x, y) at the same location in each image are extracted in chronological order of acquisition time to form the temporal data of that pixel. , where t∈[0,T].

[0044] like Figure 3 As shown, S2: Perform Fourier analysis on the polarization response curve to extract polarization characteristic parameters: for time series data The discrete Fourier transform is performed on the one-dimensional time series of each pixel position (x, y) to convert the time-domain signal into a frequency-domain signal. , where ω is the frequency;

[0045] From frequency domain signals Extract the frequency components related to the polarization modulation frequency, including: DC component. Characterized by average light intensity; fundamental frequency component ,in, The fundamental frequency for polarization modulation;

[0046] Based on DC component and fundamental frequency component Calculate polarization intensity parameters The calculation formula is: ;

[0047] Based on fundamental frequency component Calculating polarization phase parameters in complex form The calculation formula is: ;

[0048] Among them, polarization intensity parameters The polarization degree reflects the pixel location and ranges from [0, 1]; polarization phase parameter It reflects the polarization direction of the pixel position, and its value ranges from [-π, π].

[0049] like Figure 4 As shown, S3: Generate a feature image based on surface scattering characteristic parameters. : Set of polarization characteristic parameters polarization intensity parameters and polarization phase parameters Perform two-dimensional Fourier transforms to obtain the spatial spectrum of the polarization intensity parameters. Spatial spectrum of polarization phase parameters Where u and v are spatial frequency coordinates;

[0050] Calculation of surface scattering power spectral density distribution based on spatial spectrum: Spatial texture characteristics characterizing polarization intensity; It characterizes the spatial variation of polarization phase; among which, the power spectral density reflects the energy distribution of different spatial frequency components in surface scattering;

[0051] Extracting surface scattering characteristic parameters from power spectral density distribution: assembling polarization characteristic parameters Divide the area into sliding windows of size w×w with a window step size of s. Perform a two-dimensional Fourier transform on the local polarization parameters within each window position (i, j) to obtain the local power spectral density distribution. ;

[0052] Power spectral density distribution for each local window Perform polar coordinate transformation and calculate radial average: Where f is the radial frequency and θ is the direction angle;

[0053] Radial average power spectrum for each window Perform power-law fitting: ; Obtain the local scattering intensity at each window location. and local roughness index ;

[0054] The feature parameters at the center of the window are obtained through bilinear interpolation. Mapping back to the original resolution generates a continuous scattering intensity map. and roughness distribution map Each pixel value reflects the local scattering characteristics of its neighborhood at that location;

[0055] Among them, the coal cake residue area exhibits localized high roughness values. The cleaned filter plate area exhibits localized low roughness values. By setting an appropriate window size w that is larger than the minimum residual size but smaller than the typical residual size, accurate positioning of discretely distributed coal cake residue can be achieved.

[0056] Constructing feature images based on surface scattering feature parameters This includes: acquiring scattering intensity maps generated through local window analysis and spatial interpolation. and roughness distribution map Each pixel location (x, y) contains the local scattering intensity value and roughness index value within the w×w neighborhood of that location;

[0057] Roughness distribution map Normalization is performed to map the roughness index to the [0, 1] interval: ;in, and These represent the minimum and maximum values ​​of the roughness distribution, respectively.

[0058] Construct a feature image by combining scattering intensity and normalized roughness: Where k is an adjustment parameter, ranging from 2 to 8, controlling the degree to which roughness enhances the feature values; where the feature image... The roughness difference was enhanced by using an exponential function, making the characteristic value of the coal cake residue area (high β value) significantly higher than that of the clean filter plate area (low β value), thus realizing a quantitative visualization characterization of surface scattering properties; the selection of the local analysis window size w ensured that the feature image could accurately reflect the spatial distribution of coal cake residue at different scales.

[0059] S4: Employ phase-sensitive detection methods for time-series data. Perform signal enhancement processing: Generate a reference signal synchronized with polarization modulation. Reference signal and timing data With the same time sampling points, the frequency and phase of the reference signal are consistent with the polarization modulation signal;

[0060] Temporal data for each pixel location (x, y) With reference signal Perform the relevant calculations: ; obtain the modulated correlation signal ;

[0061] For relevant signals Low-pass filtering is performed to remove high-frequency noise components, extract signal components coherent with polarization modulation, and generate enhanced time-series data. ;

[0062] Enhanced time series data Perform pixel-by-pixel frequency domain analysis: Perform Discrete Fourier Transform on the enhanced time-series data of each pixel to obtain the enhanced frequency domain signal. Extract the enhanced DC component. and fundamental frequency component ; Calculate the enhanced polarization intensity parameters ; Calculate the enhanced polarization phase parameters ; Obtain the set of enhanced polarization characteristic parameters ;

[0063] Among them, phase-sensitive detection selectively enhances the signal component synchronized with polarization modulation through correlation operation of the reference signal, suppresses random noise and incoherent interference, and improves the signal-to-noise ratio of polarization characteristic parameters.

[0064] S5: Image segmentation based on polarization feature parameters: from an enhanced set of polarization feature parameters Extracting enhanced polarization intensity parameters and enhance polarization phase parameters , as a discriminative feature for pixel classification;

[0065] Combined feature images To enhance polarization feature parameters, construct a multidimensional classification feature vector: for each pixel location (x, y), extract feature image values. Enhance polarization intensity and enhance polarization phase ; Calculate the standard deviation of the enhanced polarization phase within the local window Characterizes phase stability;

[0066] Pixel classification based on multidimensional features: when and and If the condition is met, it is determined to be coal cake residue; otherwise, it is determined to be non-residue. For the feature image threshold, The polarization intensity threshold. This is the phase stability threshold; specifically... ,in, For feature images The global mean, Standard deviation, For adjustment coefficients, Value range: [0.5, 2.0], adjusted according to residual contrast. Polarization intensity threshold. Phase stability threshold .

[0067] Generate a binarized residual distribution map Assign values ​​to pixels identified as coal cake residue. ; Assign values ​​to pixels determined to be non-residual Among them, multidimensional feature classification comprehensively utilizes surface scattering characteristics and enhanced polarization characteristics, and coal cake residue exhibits high feature values ​​due to its rough surface. Low polarization degree Its high phase instability contrasts sharply with that of clean filter plates.

[0068] like Figure 5 As shown, S6: Generate detection results: Use the 8-connected neighborhood labeling algorithm to identify all Connected regions, assigning a unique label to each independent residual region. Perform morphological opening operations on the marked regions to eliminate areas smaller than 10 ... The noise points, among which, Pixels; calculate the geometric features of each connected component: area Centroid coordinates Circumscribed rectangle, aspect ratio, and roundness; for each residual region Extracting enhanced polarization intensity within the region and enhance polarization phase ;

[0069] Calculate the polarization characteristic statistics of the residual region: average degree of polarization: Standard deviation of polarization: Phase consistency: ;

[0070] Determine the residual state based on polarization characteristics: when When, it is determined to be a thick residue; when At that time, it was determined to be a moderate residue; when At that time, it was determined to be a thin residue;

[0071] The foregoing illustrative description of the present application and its embodiments is not restrictive and can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application. The accompanying drawings are only one embodiment of the present application, and the actual structure is not limited thereto. Therefore, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the present application, such designs should fall within the scope of protection of this application. Furthermore, the word "comprising" does not exclude other elements or steps, and the word "a" preceding an element does not exclude the inclusion of "a plurality" of that element. Terms such as "first," "second," etc., are used to indicate names and do not indicate any specific order.

Claims

1. A method for detecting residual coal cake in a filter press based on visual recognition, characterized in that, include: S1, Dynamic polarization imaging of the filter plate surface is performed using a polarization phase modulation imaging system. Image sequences are acquired during the continuous change of polarization state. Each image in the image sequence corresponds to a polarization state. The image sequence is arranged in chronological order to form the temporal data of each pixel. Where x and y are pixel coordinates, and t is time; S2, for time series data Pixel-by-pixel frequency domain analysis is performed to convert the time-domain signal into a frequency-domain signal, and a set of polarization feature parameters is extracted from the frequency-domain signal. Among them, the set of polarization characteristic parameters Including polarization intensity parameters and polarization phase parameters ; S3, for the set of polarization characteristic parameters Spatial frequency analysis is performed to calculate the power spectral density distribution of polarization parameters. Surface scattering characteristic parameters are extracted based on the power spectral density distribution, and a feature image is generated based on these parameters. Feature Image The pixel value reflects the surface scattering characteristics of the corresponding position on the filter plate surface; S4, using a phase-sensitive detection method for time-series data. Perform signal enhancement processing to generate enhanced timing data. Based on enhanced time-series data Calculate the polarization characteristic parameters to obtain the enhanced polarization characteristic parameter set. ; S5, based on the enhanced polarization characteristic parameter set For feature images Pixel classification is performed, classifying each pixel as either coal cake residue or non-residue, generating a binarized residue distribution map. ; S6, based on the residual distribution map and enhanced polarization characteristic parameter set The results of coal cake residue detection in the filter press are generated.

2. The method for detecting residual coal cake in a filter press based on visual recognition according to claim 1, characterized in that: S1, Dynamic polarization imaging of the filter plate surface is performed using a polarization phase modulation imaging system, acquiring image sequences during the continuous change of polarization state, including: The polarization state of the incident light illuminating the filter surface is periodically modulated by the polarization modulator in the polarization phase modulation imaging system, so that the polarization angle of the incident light follows a preset function. It changes continuously with time t; The imaging device in the polarization phase modulation imaging system synchronously acquires images of the reflected light from the filter surface. The image acquisition and polarization modulation are synchronized in time. N images are acquired within one modulation period to form an image sequence, where the i-th image corresponds to a polarization angle. ; Extract the pixel grayscale value at the same spatial location (x, y) from each image in the image sequence. Arrange the N grayscale values ​​according to the time sequence of image acquisition to form a one-dimensional time series of the corresponding pixel location, which serves as the temporal data of each pixel. .

3. The method for detecting residual coal cake in a filter press based on visual recognition according to claim 2, characterized in that: S2, Extracting the set of polarization feature parameters from the frequency domain signal. ,include: For time series data The discrete Fourier transform is performed on the one-dimensional time series of each pixel position (x, y) to convert the time-domain signal into a frequency-domain signal. , where ω is the frequency; From frequency domain signals Extracting frequency components related to the polarization modulation frequency, including: DC component It characterizes the average light intensity; Fundamental frequency component ,in, The fundamental frequency for polarization modulation; Based on DC component and fundamental frequency component Calculate polarization intensity parameters The calculation formula is: ; Based on fundamental frequency component Calculating polarization phase parameters in complex form The calculation formula is: .

4. The method for detecting residual coal cake in a filter press based on visual recognition according to claim 2, characterized in that: S3, Generate a feature image based on surface scattering characteristic parameters. ,include: For the set of polarization characteristic parameters Perform spatial frequency domain transformation to convert the polarization intensity parameters and polarization phase parameters Transform from the spatial domain to the frequency domain to obtain the corresponding spatial spectrum; Power spectral density distribution calculated based on spatial spectrum The power spectral density distribution quantifies the energy distribution of polarization parameters at different spatial frequencies, reflecting the texture and roughness characteristics of the material surface; Statistical analysis of the power spectral density distribution was performed to extract characteristic parameters that characterize the surface scattering properties. Constructing feature images based on surface scattering characteristic parameters By performing local analysis on the neighborhood of each pixel, the extracted scattering feature parameters are mapped to pixel values, thus improving the feature image. Each pixel value quantitatively characterizes the surface scattering properties at the corresponding location.

5. The method for detecting residual coal cake in a filter press based on visual recognition according to claim 4, characterized in that: Statistical analysis of the power spectral density distribution was performed to extract characteristic parameters representing the surface scattering properties, including: For the set of polarization characteristic parameters The local window is divided, and the analysis window of a preset size slides in the spatial domain with a set step size. The local polarization parameters at each window position are transformed in the frequency domain to generate the local power spectral density distribution corresponding to the spatial position. Radial statistical analysis is performed on each local power spectral density distribution, and the two-dimensional power spectrum is converted into a one-dimensional radial power spectrum by angle integration to eliminate direction dependence; The radial power spectrum is fitted with parameters based on the power law model of surface scattering theory. Characteristic parameters that characterize local scattering properties, including scattering intensity parameters and roughness index, are extracted from the fitting results. The mapping relationship between each window position and its local characteristic parameters is established. By using spatial interpolation, the feature parameters of discrete window locations are extended to the entire image space to generate continuously distributed feature parameters, so that each pixel location obtains a parameter value that reflects the scattering characteristics of its neighborhood.

6. The method for detecting residual coal cake in a filter press based on visual recognition according to claim 4, characterized in that: Constructing feature images based on surface scattering characteristic parameters ,include: Obtain the characteristic parameters of the continuous distribution, where the characteristic parameters include the distribution of scattering intensity parameters. and roughness index distribution ; For roughness index distribution Normalization is performed to obtain the normalized roughness index distribution. ; Distribution of scattering intensity parameters With the normalized roughness index distribution Perform nonlinear fusion to construct a feature image: Where k is a preset enhancement coefficient, which amplifies the influence of roughness differences on eigenvalues ​​through an exponential function.

7. The method for detecting residual coal cake in a filter press based on visual recognition according to claim 4, characterized in that: S4, using a phase-sensitive detection method for time-series data. Perform signal enhancement processing to generate enhanced timing data. ,include: Establish a reference signal synchronized with the polarization state modulation of the polarization modulator. Time sampling and timing data of the reference signal Consistent, the frequency of the reference signal is equal to the fundamental frequency of the polarization modulation. Phase and polarization angle variation function Maintain a fixed relationship; Based on reference signal Time series data is demodulated through coherent demodulation. Signal enhancement processing is performed to obtain enhanced time-series data after noise suppression. .

8. The method for detecting residual coal cake in a filter press based on visual recognition according to claim 4, characterized in that: S5, Generate a binarized residual distribution map ,include: For each pixel location (x, y), extract the feature image value. Enhance polarization intensity and enhance polarization phase ; Based on enhanced polarization phase parameters Calculate the local phase stability index, centered on pixel (x, y). Calculate the phase standard deviation within the window This characterizes the phase stability at that position; when and and At that time, it was determined to be coal cake residue; Otherwise, it is determined to be non-residual; in, Threshold for feature image The polarization intensity threshold. This is the phase stability threshold; Assign values ​​to pixels identified as coal cake residue. ; Assign values ​​to pixels that are determined to be non-residual. .

9. The method for detecting residual coal cake in a filter press based on visual recognition according to claim 4, characterized in that: S6 generates the filter press coal cake residue detection results, including: Using a connected component labeling algorithm to The pixels are divided into regions to obtain independent sets of residual regions. And calculate the geometric properties of each residual region; For each residual area Extract enhanced polarization intensity parameters within the region and enhance polarization phase parameters Calculate the regional polarization characteristic statistics; Based on the geometric properties and polarization characteristic statistics of the residual area, the detection results of filter press coal cake residue are generated.

10. A filter press coal cake residue detection system based on visual recognition, characterized in that, include: A polarization phase modulation imaging device performs dynamic polarization imaging on the surface of a filter plate. It includes a polarization modulator and an imaging device. The polarization modulator periodically modulates the polarization state of the incident light illuminating the filter plate surface, while the imaging device simultaneously acquires image sequences of the reflected light from the filter plate surface. These image sequences are then arranged chronologically to form temporal data for each pixel. ; Frequency domain analysis module, for time series data Pixel-by-pixel frequency domain analysis is performed, converting the time-domain signal into a frequency-domain signal using Discrete Fourier Transform, and extracting parameters including polarization intensity from the frequency-domain signal. and polarization phase parameters The set of polarization characteristic parameters ; The spatial frequency analysis module analyzes the set of polarization characteristic parameters. Spatial frequency analysis is performed, and surface scattering characteristic parameters are extracted through local windowing and power spectral density calculation, based on scattering intensity parameters. and roughness index Nonlinear fusion to generate feature images ; The phase-sensitive detection module uses a reference signal synchronized with polarization modulation to detect timing data. Perform coherent demodulation to generate enhanced time-series data. And calculate the set of enhanced polarization characteristic parameters based on the enhanced time series data. ; Pixel classification module, based on feature images and enhanced polarization characteristic parameter set A multi-dimensional feature vector is constructed, and each pixel is classified as coal cake residue or non-residue using a joint discrimination criterion, generating a binarized residue distribution map. ; The detection result generation module generates a residual distribution map. Connectivity analysis is performed, and combined with the enhanced polarization characteristic statistics within the region, to generate filter press coal cake residue detection results that include residual spatial distribution and physical properties.

Images