Tailing coal ash content detection method and system based on space-frequency double-domain fusion

Abstract

The invention discloses a tail coal ash content detection method and system based on space-frequency dual-domain fusion. The method comprises the following steps: preparing a coal slime mixed solution and obtaining a tail coal liquid flow image data set; introducing a space-frequency double-domain fusion module and a state-space-based feature extraction module, and constructing a tail coal ash content detection model; and based on a tailing ash content detection model, performing tailing ash content detection on the tailing liquid flow image data set to obtain a concentration value and an ash content category of the coal slime mixed liquid. According to the method, the time required by model operation can be greatly shortened, and the ash value of the coal slime flotation tailings can be rapidly detected. The tail coal ash content detection method and system based on space-frequency double-domain fusion can be widely applied to the technical field of tail coal ash content detection.

Description

Technical Field

[0001] This invention relates to the field of tailings ash content detection technology, and in particular to a tailings ash content detection method and system based on spatial-frequency dual-domain fusion. Background Technology

[0002] Coal slime flotation is a crucial step in coal production. In traditional flotation processes, ash content detection mainly relies on manual sampling and analysis, as well as rapid ash analysis methods. This involves collecting samples from the production line and calculating the ash content by measuring the mass ratio before and after combustion. These methods are not only unreliable in terms of timeliness and difficulty in providing real-time guidance for production control, but also highly dependent on human experience. To overcome these shortcomings, existing online ash content detection methods mainly include radiometric ash determination, image processing ash determination, spectral analysis ash determination, and multi-source fusion ash determination.

[0003] Existing image processing methods for ash content determination have significant limitations. These methods rely solely on surface images or bubble images of the coal slime mixture for ash detection, failing to fully extract the internal information of the mixture and the intrinsic correlation between it and the tailings ash content. Furthermore, their core steps depend on manually designed features (such as color and grayscale features), and different working conditions often require different features, resulting in poor versatility. Simultaneously, the variable working conditions in actual production (such as foam state, background impurities, shooting angle, and foaming agent dosage) significantly affect the extraction effect of preset features, imposing stringent requirements on the image acquisition environment, ultimately leading to poor adaptability and robustness of the method.

[0004] The current method of ash determination using spectral analysis faces challenges in terms of real-time performance and detection stability. This method requires offline testing of coal slime mixture samples collected from the production line (outside the production process), with each measurement taking at least 3 minutes, which cannot meet the real-time control requirements of the flotation process. More importantly, the coal slime suspension exhibits significant randomness during sedimentation, leading to variations in the time required for each sedimentation and the resulting state. This uncertainty in the sedimentation process directly affects the stability of the collected spectral data, resulting in low accuracy of the model built upon it.

[0005] The gray measurement method that fuses absorption spectra with multimodal images can effectively improve detection accuracy. However, the absorption spectrum contains interference noise caused by the uncertainty of the solvent. Combined with the aforementioned unstable characteristics of the image, this can lead to a significant deviation of the detection results from the true value under certain circumstances. Furthermore, the method of directly fusing the two types of data without processing requires a large amount of computing power, which is difficult to meet the requirements of lightweight algorithms in industrial production. Summary of the Invention

[0006] To address the aforementioned technical problems, the present invention aims to provide a tailings ash content detection method and system based on space-frequency dual-domain fusion, which can significantly reduce the time required for model calculation and achieve rapid detection of tailings ash content.

[0007] The first technical solution adopted in this invention is: a method for detecting tailings ash content based on space-frequency dual-domain fusion, comprising the following steps: Prepare coal slime mixture and acquire tailings liquid flow image dataset; A space-frequency dual-domain fusion module and a state-space feature extraction module are introduced to construct a tailings ash content detection model; Based on the tailings ash detection model, tailings ash content was detected in the tailings liquid flow image dataset to obtain the concentration value and ash category of the coal slime mixture.

[0008] Furthermore, the step of preparing the coal slime mixture and acquiring the tailings liquid flow image dataset specifically includes: Coal slime mixtures with different ash contents and concentrations were prepared, and peristaltic pumps were used to drive the coal slime mixtures to form a circulating liquid flow. A transparent quartz tube is installed as a data acquisition window on the pipeline based on the circulating fluid flow. The light source is placed on one side of the acquisition window as a backlight, and the industrial camera is placed on the other side to acquire images of the circulating liquid flow, thus obtaining a tailings liquid flow image dataset.

[0009] Furthermore, the tailings ash content detection model specifically includes a grayscale transformation module, a location embedding module, a space-frequency dual-domain fusion module, a state-space feature extraction module, a Drop Path module, a feedforward neural network, and a SoftMax function. The grayscale transformation module, the location embedding module, the space-frequency dual-domain fusion module, the state-space feature extraction module, the Drop Path module, the feedforward neural network, and the SoftMax function are sequentially connected, wherein: The space-frequency dual-domain fusion module includes a first space-frequency dual-domain fusion module and a second space-frequency dual-domain fusion module; The state space-based feature extraction module includes a first state space-based feature extraction module and a second state space-based feature extraction module; The Drop Path module includes a first Drop Path module and a second Drop Path module; The first space-frequency dual-domain fusion module, the first state-space feature extraction module, the first Drop Path module, the second space-frequency dual-domain fusion module, the second state-space feature extraction module, and the second Drop Path module are connected in sequence.

[0010] Furthermore, the space-frequency dual-domain fusion module specifically includes a two-dimensional discrete fast Fourier transform module, an amplitude spectrum calculation module, a phase spectrum calculation module, a Gaussian filtering module, an inverse Fourier transform module, and an inversion module. The first output of the two-dimensional discrete fast Fourier transform module is connected to the input of the amplitude spectrum calculation module; the second output of the two-dimensional discrete fast Fourier transform module is connected to the input of the phase spectrum calculation module; the output of the phase spectrum calculation module is connected to the input of the Gaussian filtering module; the outputs of the amplitude spectrum calculation module and the Gaussian filtering module are connected to the input of the inverse Fourier transform module; and the output of the inverse Fourier transform module is connected to the input of the inversion module.

[0011] Furthermore, the state space-based feature extraction module specifically includes a first normalization module, a first linear mapping module, a two-dimensional convolution module, a first SiLU function, a second SiLU function, a state space block, a second linear mapping module, and a second normalization module. The output of the first normalization module is connected to the input of the first linear mapping module; the first output of the first linear mapping module is connected to the input of the two-dimensional convolution module; the output of the two-dimensional convolution module is connected to the first SiLU function; the second output of the first linear mapping module is connected to the second SiLU function; the first SiLU function is connected to the input of the state space block; the output of the state space block and the second SiLU function are connected to the input of the second linear mapping module; and the output of the second linear mapping module is connected to the input of the second normalization module.

[0012] Furthermore, the step of detecting the ash content of tailings liquid in the tailings liquid flow image dataset based on the tailings ash content detection model to obtain the concentration value and ash category of the coal slime mixture specifically includes: Input the tailings liquid flow image dataset into the tailings ash content detection model; The grayscale transformation module based on the tailings ash content detection model performs grayscale transformation processing on the tailings liquid flow image dataset to obtain the tailings liquid flow grayscale map. Based on the location embedding module of the tailings ash content detection model, the grayscale image of tailings liquid flow is scanned from left to right and the location is encoded and arranged to obtain the arranged grayscale image of tailings liquid flow. The spatial-frequency dual-domain fusion module based on the tailings ash content detection model performs spatial and frequency domain characterization fusion on the arranged tailings liquid flow grayscale map to obtain the fused tailings liquid flow grayscale map. A state-space feature extraction module based on the tailings ash content detection model extracts features from the fused tailings liquid flow grayscale map to obtain a tailings liquid flow feature map. The Drop Path module based on the tailings ash content detection model performs regularization processing on the tailings liquid flow feature map to obtain the regularized tailings liquid flow feature map. Based on the feedforward neural network of the tailings ash content detection model, feature learning and weight training are performed on the regularized tailings liquid flow feature map to obtain the probability class distribution of the tailings liquid flow feature map. Based on the SoftMax function of the tailings ash content detection model, the probability category distribution of the tailings liquid flow characteristic map is calculated to classify the concentration and ash content into categories, thereby obtaining the concentration value and ash content category of the coal slime mixture.

[0013] Furthermore, the spatial-frequency dual-domain fusion module based on the tailings ash content detection model performs spatial and frequency domain characterization fusion on the arranged tailings liquid flow grayscale map to obtain the fused tailings liquid flow grayscale map. This step specifically includes: The sorted tailings liquid flow grayscale image is input into the space-frequency dual-domain fusion module of the tailings ash content detection model; The two-dimensional discrete fast Fourier transform module based on the space-frequency dual-domain fusion module performs two-dimensional discrete fast Fourier transform on the arranged tailings liquid flow grayscale map to decompose it into amplitude map and phase map. The amplitude spectrum calculation module based on the space-frequency dual-domain fusion module performs frequency intensity calculation on the amplitude map to obtain the amplitude spectrum. The phase spectrum calculation module based on the space-frequency dual-domain fusion module calculates the frequency position information of the phase map to obtain the phase spectrum; A Gaussian filtering module based on the space-frequency dual-domain fusion module performs Gaussian filtering on the phase spectrum to obtain the filtered phase spectrum. The inverse Fourier transform module based on the space-frequency dual-domain fusion module performs inverse Fourier transform on the amplitude spectrum and the filtered phase spectrum to obtain the reconstructed spatial domain image. The inversion module based on the space-frequency dual-domain fusion module performs an inversion operation on the reconstructed spatial domain image to obtain the inverted spatial domain image. The inverted spatial domain image and the reconstructed spatial domain image are added together by their respective elements, and then multiplied together by their respective elements with the arranged tailings liquid flow grayscale image to obtain the fused tailings liquid flow grayscale image.

[0014] Furthermore, the state-space feature extraction module based on the tailings ash content detection model extracts features from the fused tailings liquid flow grayscale map to obtain the tailings liquid flow feature map. This step specifically includes: The fused tailings liquid flow grayscale image is input into the state space feature extraction module of the tailings ash content detection model; The first normalization module based on the state space feature extraction module normalizes the fused tailings liquid flow grayscale map to obtain the normalized tailings liquid flow grayscale map. The first linear mapping module based on the state space feature extraction module performs linear mapping on the normalized tailings liquid flow grayscale map to obtain the tailings liquid flow grayscale map of the x vector and the tailings liquid flow grayscale map of the y vector. Based on the state space feature extraction module, the two-dimensional convolution module, the first SiLU function, and the state space block, the two-dimensional convolution operation is performed on the grayscale image of the tailings liquid flow of the x vector to extract local features. The hidden state is calculated based on the local features to obtain the output of the next state of the x vector. Based on the second SiLU function of the state space feature extraction module, the grayscale image of tailings liquid flow of the y vector is linearly mapped to obtain the nonlinear features of the y vector. Based on the state space feature extraction module, the second linear mapping module and the second normalization module sequentially fuse the output of the next state of the x vector and the nonlinear features of the y vector, perform linear mapping and root mean square normalization operations to obtain the tailings liquid flow feature map.

[0015] The second technical solution adopted in this invention is: a tailings ash content detection system based on space-frequency dual-domain fusion, comprising: The first module is used to prepare coal slime mixture and acquire tailings liquid flow image dataset; The second module is used to introduce the space-frequency dual-domain fusion module and the state-space feature extraction module to construct a tailings ash content detection model. The third module is used to detect the ash content of tailings based on the tailings ash content detection model, and to obtain the concentration value and ash category of the coal slurry mixture.

[0016] The beneficial effects of the method and system of this invention are as follows: This invention prepares a coal slime mixture and obtains a tailings flow image dataset. It further introduces a space-frequency dual-domain fusion module and a state-space-based feature extraction module to construct a tailings ash content detection model. Through the space-frequency dual-domain fusion module, the frequency domain representation of the image is introduced, and the frequency domain information is filtered to remove interference from illumination, concentration, and bubbles. Finally, the reconstructed spatial domain information is aligned and fused with the original spatial domain information in an orderly manner, enhancing the representation of coal slime particle details in the spatial domain. Through the state-space-based feature extraction module, an efficient combination of convolution and state space is introduced, and features at different scales are aligned and fused, improving the efficiency of feature extraction. Finally, based on the tailings ash content detection model, the tailings flow image dataset is used to detect the tailings ash content, obtaining the concentration value and ash category of the coal slime mixture. This significantly reduces the time required for model computation and enables rapid detection of tailings ash content in coal slime flotation. Attached Figure Description

[0017] Figure 1 This is a flowchart of the steps of a tailings ash content detection method based on space-frequency dual-domain fusion according to the present invention; Figure 2 This is a structural block diagram of a tailings ash content detection system based on space-frequency dual-domain fusion according to the present invention; Figure 3 This is a schematic diagram of a device for acquiring images of tailings liquid flow provided in a specific embodiment of the present invention; Figure 4 This is a schematic diagram of the tailings liquid flow image dataset provided in a specific embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of the tailings ash content detection model provided in a specific embodiment of the present invention; Figure 6 This is a schematic diagram of the structure of the space-frequency dual-domain fusion module provided in a specific embodiment of the present invention; Figure 7 This is a schematic diagram of the structure of the state space-based feature extraction module provided in a specific embodiment of the present invention. Detailed Implementation

[0018] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. The step numbers in the following embodiments are only for ease of explanation and do not limit the order of the steps. The execution order of each step in the embodiments can be adapted according to the understanding of those skilled in the art.

[0019] First, it should be noted that existing online ash content detection methods mainly include radiometric ash measurement, image processing ash measurement, spectral analysis ash measurement, and multi-source fusion ash measurement, among which: Image processing gray measurement method: First, an industrial camera is used to acquire images of bubbles or liquid surfaces during the coal slime flotation process under specific light sources and angles; then, manual feature extraction (such as color features, grayscale features, etc.) and preprocessing (such as grayscale transformation, histogram equalization, etc.) are performed; finally, the processed image is input into a convolutional neural network (CNN) for preset feature extraction and learning, and finally, gray content prediction is achieved.

[0020] Ash content determination by spectral analysis: First, coal slime mixture samples are collected from the production line and placed in quartz cuvettes. Then, a spectrometer is used to collect the absorption spectra of the mixture at different time points during the sedimentation process (e.g., 0s, 30s, 60s, 90s, 300s). Finally, these multiple sets of time-series spectral data are merged and input into a hybrid model combining a convolutional neural network (CNN) and a long short-term memory network (LSTM). This model utilizes CNN to extract spatial features and LSTM to capture the temporal changes in the sedimentation process, jointly completing the detection of ash content in the coal slime mixture.

[0021] Multi-source fusion ash content measurement method: This method uses multiple sets of sensors to collect two or more different types of data from the production site, extracts different features through deep learning, and fuses the features of the multi-source data through multimodal fusion technology to jointly characterize the ash content, thereby achieving the purpose of ash content detection.

[0022] However, existing image processing methods for ash measurement rely solely on surface images or bubble images of coal slime mixtures for ash content detection, failing to fully extract the internal information of the mixture and the intrinsic correlation between the ash content of tailings. Spectroscopic analysis methods for ash measurement suffer from insufficient real-time performance and poor detection stability. Ash measurement methods that fuse absorption spectra with multimodal image analysis can lead to significant deviations from the true values ​​under certain conditions. Furthermore, methods that directly fuse two types of data without processing require substantial computational power, making it difficult to meet the requirements of lightweight algorithms in industrial production.

[0023] Based on this, this embodiment of the invention uses a spatial-frequency domain feature fusion method to weaken the impact of external factors such as unstable illumination and concentration differences on the internal image of coal slime mixture flow. Compared with the fusion of multi-source data, this embodiment of the invention only fuses the spatial domain features and frequency domain features of the image, which greatly reduces the computational power requirements of the fusion process and is more suitable for industrial production scenarios.

[0024] Reference Figure 1 This invention provides a method for detecting the ash content of tailings coal based on space-frequency dual-domain fusion, the method comprising the following steps: S100: Prepare coal slime mixture and acquire tailings liquid flow image dataset; Specifically, coal slime mixtures with different ash contents and concentrations are prepared, and a peristaltic pump is used to drive the coal slime mixture to form a circulating liquid flow. A transparent quartz tube is set on the pipeline based on the circulating liquid flow as a collection window. A light source is placed on one side of the collection window as a backlight, and an industrial camera is placed on the other side to collect images of the circulating liquid flow, thereby obtaining a tailings liquid flow image dataset.

[0025] In this embodiment, as Figure 3 As shown, a simulated industrial flotation process was used to produce 90 different coal slime mixtures with varying ash contents and concentrations. A peristaltic pump was used to drive the coal slime mixtures into a circulating flow, and a transparent quartz tube was installed on the circulation pipeline as a data acquisition window. A rectangular light source was placed on one side of the acquisition window as backlight, and an industrial camera was placed on the other side to acquire images of the tailings flow. The acquired image data was transmitted to a computer via USB communication and packaged into a flow image dataset, such as... Figure 4 As shown, this is for use in model development and training.

[0026] S200, introduces a space-frequency dual-domain fusion module and a state-space feature extraction module to construct a tailings ash content detection model; First, it should be noted that the embodiments of the present invention propose a deep learning model based on spatial-frequency dual-domain fusion, which improves efficiency and accuracy in image feature extraction and enhancement, and reduces computational overhead compared with existing models.

[0027] In this embodiment, as Figure 5 As shown, the tailings ash content detection model specifically includes a grayscale transformation module, a location embedding module, a spatial-frequency dual-domain fusion module, a state-space feature extraction module, a Drop Path module, a feedforward neural network, and a SoftMax function. The grayscale transformation module, the location embedding module, the spatial-frequency dual-domain fusion module, the state-space feature extraction module, the Drop Path module, the feedforward neural network, and the SoftMax function are sequentially connected. The spatial-frequency dual-domain fusion module includes a first spatial-frequency dual-domain fusion module and a second spatial-frequency dual-domain fusion module; the state-space feature extraction module includes a first state-space feature extraction module and a second state-space feature extraction module; the Drop Path module includes a first Drop Path module and a second Drop Path module; and the first spatial-frequency dual-domain fusion module, the first state-space feature extraction module, the first Drop Path module, the second spatial-frequency dual-domain fusion module, the second state-space feature extraction module, and the second Drop Path module are sequentially connected.

[0028] Furthermore, such as Figure 6 As shown, the space-frequency dual-domain fusion module specifically includes a two-dimensional discrete fast Fourier transform module, an amplitude spectrum calculation module, a phase spectrum calculation module, a Gaussian filtering module, an inverse Fourier transform module, and an inversion module. The first output of the two-dimensional discrete fast Fourier transform module is connected to the input of the amplitude spectrum calculation module; the second output of the two-dimensional discrete fast Fourier transform module is connected to the input of the phase spectrum calculation module; the output of the phase spectrum calculation module is connected to the input of the Gaussian filtering module; the outputs of the amplitude spectrum calculation module and the Gaussian filtering module are connected to the input of the inverse Fourier transform module; and the output of the inverse Fourier transform module is connected to the input of the inversion module.

[0029] like Figure 7As shown, the state space-based feature extraction module specifically includes a first normalization module, a first linear mapping module, a two-dimensional convolution module, a first SiLU function, a second SiLU function, a state space block, a second linear mapping module, and a second normalization module. The output of the first normalization module is connected to the input of the first linear mapping module. The first output of the first linear mapping module is connected to the input of the two-dimensional convolution module. The output of the two-dimensional convolution module is connected to the first SiLU function. The second output of the first linear mapping module is connected to the second SiLU function. The first SiLU function is connected to the input of the state space block. The output of the state space block and the second SiLU function are connected to the input of the second linear mapping module. The output of the second linear mapping module is connected to the input of the second normalization module.

[0030] S300. Based on the tailings ash content detection model, tailings ash content is detected in the tailings liquid flow image dataset to obtain the concentration value and ash category of the coal slime mixture.

[0031] S310. Input the tailings liquid flow image dataset into the tailings ash content detection model; S320, a grayscale transformation module based on the tailings ash content detection model, performs grayscale transformation processing on the tailings liquid flow image dataset to obtain a grayscale map of the tailings liquid flow; S330, The position embedding module based on the tailings ash content detection model performs position encoding and arrangement of the tailings liquid flow grayscale map from left to right to obtain the arranged tailings liquid flow grayscale map; S340, Spatial-frequency dual-domain fusion module based on tailings ash content detection model, performs spatial and frequency domain characterization fusion on the arranged tailings liquid flow grayscale map to obtain the fused tailings liquid flow grayscale map. Specifically, the arranged grayscale image of the tailings liquid flow is input into the space-frequency dual-domain fusion module of the tailings ash content detection model; the two-dimensional discrete fast Fourier transform module based on the space-frequency dual-domain fusion module performs a two-dimensional discrete fast Fourier transform on the arranged grayscale image of the tailings liquid flow, decomposing it into an amplitude map and a phase map; the amplitude spectrum calculation module based on the space-frequency dual-domain fusion module calculates the frequency intensity of the amplitude map to obtain the amplitude spectrum; the phase spectrum calculation module based on the space-frequency dual-domain fusion module calculates the frequency position information of the phase map to obtain the phase spectrum; and the Gaussian spectrum calculation module based on the space-frequency dual-domain fusion module... The filtering module performs Gaussian filtering on the phase spectrum to obtain the filtered phase spectrum. The inverse Fourier transform module based on the space-frequency dual-domain fusion module performs inverse Fourier transform on the amplitude spectrum and the filtered phase spectrum to obtain the reconstructed spatial domain image. The phase inversion module based on the space-frequency dual-domain fusion module inverts the reconstructed spatial domain image to obtain the inverted spatial domain image. The inverted spatial domain image and the reconstructed spatial domain image are added element-wise, and then multiplied element-wise with the arranged tailings liquid flow grayscale image to obtain the fused tailings liquid flow grayscale image.

[0032] In this embodiment, the module is a single-input, single-output module, and the input is a pixel size of... The image is transformed in the frequency domain, and the frequency and spatial information are jointly represented to achieve feature enhancement. The input image is processed as a matrix using a two-dimensional discrete fast Fourier transform according to the following formula, mapping the image from the spatial domain to the frequency domain and decomposing it into an amplitude map and a phase map, where the amplitude spectrum... Represents frequency intensity and phase spectrum Location information indicating frequency.

[0033] ; in, Indicates pixel position; Represents the grayscale value (0-225) at a specific location; Represents frequency components, For horizontal frequency, The vertical frequency.

[0034] After the Fast Fourier Transform, a high-pass Gaussian filter is applied to the decomposed phase spectrum to remove high-frequency components from the original image, i.e., background interference. The Gaussian filter is shown below: ; in, For frequency domain points To the center point Euclidean distance, Represents the height and width of the image in pixels; This indicates the cutoff frequency, which determines the boundary position where the filter begins to significantly retain high-frequency components.

[0035] The filtered phase spectrum and the original amplitude spectrum are then subjected to inverse Fourier transform for spatial domain information reconstruction, converting the frequency domain information into an image. The conversion process is shown below: ; in, Indicates pixel position; Represents the grayscale value (0-225) at a specific location; Represents frequency components, For horizontal frequency, The vertical frequency.

[0036] Finally, the reconstructed spatial domain image will be... Perform pixel inversion operation to obtain and will and The image is fused by adding the elements in pairs and then multiplying them with the input image in pairs. The fusion process is as follows: ; in, The image input to the spatial-frequency dual-domain fusion module is represented as a matrix. The output image after processing by the space-frequency dual-domain fusion module is represented in matrix form. This indicates the addition of the matrix elements. This indicates element-wise multiplication of a matrix.

[0037] S350, a state-space feature extraction module based on the tailings ash content detection model, extracts features from the fused tailings liquid flow grayscale map to obtain a tailings liquid flow feature map. Specifically, the fused tailings liquid flow grayscale image is input into the state-space feature extraction module of the tailings ash content detection model; the first normalization module of the state-space feature extraction module normalizes the fused tailings liquid flow grayscale image to obtain a normalized tailings liquid flow grayscale image; the first linear mapping module of the state-space feature extraction module performs a linear mapping on the normalized tailings liquid flow grayscale image to obtain... Grayscale diagram of tailings liquid flow vector and The vector-based grayscale image of tailings liquid flow; a two-dimensional convolution module based on the state-space feature extraction module, the first SiLU function, and the state-space block, for... The grayscale image of the tailings liquid flow vector is subjected to a two-dimensional convolution operation to extract local features. Hidden states are then calculated based on these local features. The output of the next state in the vector inference; the second SiLU function based on the state space feature extraction module, for... The grayscale image of the tailings liquid flow vector is linearly mapped to obtain the nonlinear characteristics of the y-vector; based on the second linear mapping module and the second normalization module of the state space feature extraction module, the nonlinear characteristics of the y-vector are obtained. The output of the next state in the vector reasoning and The nonlinear characteristics of the vector are successively fused, linearly mapped, and normalized by root mean square to obtain the tailings liquid flow characteristic map.

[0038] In this embodiment, the module is a single-input, single-output module, and the input is a pixel size of... The feature map is used to perform feature extraction through convolution and state space computation.

[0039] First, the input is normalized to scale the data to between 0 and 1, facilitating subsequent processing and preventing overfitting. Second, the normalized vector is linearly mapped to... , Within two subspaces. Vectors are subjected to 2D convolution to extract local features. These local features are then input into the SSM (Synchronous Multiplication Model) to calculate the hidden state. The hidden state of each input is memorized, and the output representation of the next state is inferred. The inference process is as follows: ; The vectors then directly obtain nonlinear features through the SiLU function, the activation function of which is shown below: ; Finally, features from x and y are fused to improve the model's expressive power. The fused features are linearly mapped back to the original space, added symmetrically to the input vector, and then normalized using root mean square to obtain the extracted feature output.

[0040] S360, the Drop Path module based on the tailings ash content detection model, performs regularization processing on the tailings liquid flow characteristic map to obtain the regularized tailings liquid flow characteristic map. S370. Based on the feedforward neural network of the tailings ash content detection model, feature learning and weight training are performed on the regularized tailings liquid flow feature map to obtain the probability class distribution of the tailings liquid flow feature map. S380. Based on the tailings ash content detection model, the SoftMax function is used to calculate the probability category distribution of the tailings liquid flow characteristic map and classify the concentration and ash content to obtain the concentration value and ash content category of the coal slime mixture.

[0041] In this embodiment, firstly, the tailings liquid flow image with an input image size of 2048*2048 pixels is converted into a grayscale image through logarithmic grayscale transformation. The conversion process is as follows: ; in, These are the pixel values ​​after logarithmic transformation; To adjust the contrast constant; These are the original pixel values.

[0042] Then, the grayscale image is divided into 256 blocks of 128*128 pixels each, and their positions are encoded and arranged in a top-to-bottom, left-to-right scanning manner. Next, a spatial-frequency dual-domain fusion module is used to fuse the spatial and frequency domain representations of each block, eliminating high-frequency noise interference while enhancing the image. After this part, the number and size of the blocks remain unchanged. Then, each block is input into a state-space based feature extraction module for feature extraction. The extracted features are 32*32 pixels in size and have 128 features. Then, a random Drop Path operation is performed, resetting the weights of some extracted branches to zero to prevent overfitting. The Drop Path coefficient is 0.5, meaning that after this operation, the number of features becomes 64. Further spatial-frequency fusion and feature extraction are then performed, and the 32 8*8 feature maps are input into a feedforward neural network for feature learning and weight training. Finally, the SoftMax function is used to calculate the concentration and gray category, as shown in the calculation process below, and the final output is the concentration and gray category.

[0043] ; in, Represents the SoftMax function; For the input vector of the th One element; For vector dimensions; is the base of the natural logarithm.

[0044] In summary, this invention proposes a gray detection method based on spatial-frequency dual-domain fusion. It not only considers the representation of image features in the spatial domain but also introduces frequency domain representation. Image processing is performed in the frequency domain to remove high-frequency noise, and a fusion algorithm is used to jointly represent the reconstructed features in the frequency domain with the original spatial features. This method achieves fast and accurate detection while maintaining strong anti-interference capabilities and significantly reducing computational overhead.

[0045] In summary, the embodiments of the present invention differ from the prior art in the following technical features: 1) A spatial-frequency dual-domain fusion module is proposed. By introducing the frequency domain representation of the image, the frequency domain information is filtered to remove interference such as illumination, concentration and bubbles. Finally, the reconstructed spatial domain information is aligned and fused with the original spatial domain information in an orderly manner to enhance the representation of coal slime particle details in the spatial domain.

[0046] 2) A state-space-based feature extraction module is proposed. By introducing an efficient combination of convolution and state space, and aligning and fusing features of different scales, the efficiency of feature extraction is improved.

[0047] 3) A gray detection model based on spatial-frequency dual-domain fusion is proposed. By combining feature extraction with the orderly fusion of spatial and frequency domain features, the computational cost of the model can be significantly reduced while maintaining high accuracy and strong robustness.

[0048] Therefore, the embodiments of the present invention have the following advantages compared with the prior art: 1) Low dependence on experience: This embodiment of the invention utilizes a deep learning model. After the model is trained, it can be deployed to the work site for real-time data collection and detection. Compared with traditional methods, no human intervention is required in the detection process, thus reducing the dependence on experience.

[0049] 2) Fast detection speed: The embodiments of the present invention utilize the representation of different domains of image data for fusion, which greatly reduces the time required for model calculation and enables rapid detection of ash content in coal slime flotation tailings.

[0050] 3) Strong robustness: The embodiments of the present invention use a Gaussian filter to filter out high-frequency information in the image by processing the image frequency domain information, thereby reducing the influence of factors such as illumination and density on image features.

[0051] In summary, compared with the prior art, the embodiments of the present invention have significant advantages in speed and accuracy, and reduce the process of human intervention, eliminating many interfering factors that may affect the accuracy of the results during the manual collection or analysis of samples, thereby improving the efficiency, accuracy and robustness of detection, and significantly reducing computational overhead.

[0052] Reference Figure 2 A tailings ash content detection system based on space-frequency dual-domain fusion includes: The first module 201 is used to prepare coal slime mixture and acquire tailings liquid flow image dataset; The second module 202 is used to introduce the space-frequency dual-domain fusion module and the state-space feature extraction module to construct a tailings ash content detection model. The third module 203 is used to detect the ash content of tailings based on the tailings ash content detection model, and obtain the concentration value and ash category of the coal slurry mixture.

[0053] The content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0054] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims

1. A tail coal ash content detection method based on space-frequency dual-domain fusion, characterized in that, The method comprises the following steps: Preparation of coal slime mixed liquid and acquisition of tail coal liquid flow image data set; Introducing an empty frequency dual domain fusion module and a state space feature extraction module to construct a tail coal ash content detection model; Based on the tail coal ash content detection model, the tail coal ash content of the tail coal liquid flow image data set is detected, and the concentration value and ash classification of the coal slime mixed liquid are obtained.

2. The tail coal ash content detection method based on space-frequency dual-domain fusion according to claim 1, characterized in that, The step of preparing coal slime mixed liquid and acquiring tail coal liquid flow image data set specifically includes: Preparation of coal slime mixed liquid with different ash content and different concentration, and driving the coal slime mixed liquid to form a circulating liquid flow by using a peristaltic pump; A transparent quartz tube is arranged on the pipeline based on the circulating liquid flow as a collection window; Place the light source on one side of the collection window as backlight, and place the industrial camera on the other side to collect images of the circulating liquid flow to obtain the tail coal liquid flow image data set.

3. The tail coal ash content detection method based on space-frequency dual-domain fusion according to claim 2, characterized in that, The tail coal ash content detection model specifically includes a grayscale transformation module, a position embedding module, an empty frequency dual domain fusion module, a state space feature extraction module, a Drop Path module, a feedforward neural network, and a SoftMax function. The grayscale transformation module, the position embedding module, the empty frequency dual domain fusion module, the state space feature extraction module, the Drop Path module, the feedforward neural network, and the SoftMax function are connected in sequence. The empty frequency dual domain fusion module includes a first empty frequency dual domain fusion module and a second empty frequency dual domain fusion module. The state space feature extraction module includes a first state space feature extraction module and a second state space feature extraction module. The Drop Path module includes a first Drop Path module and a second Drop Path module.

4. The tail coal ash content detection method based on space-frequency dual-domain fusion according to claim 3, characterized in that, The first empty frequency dual domain fusion module, the first state space feature extraction module, the first Drop Path module, the second empty frequency dual domain fusion module, the second state space feature extraction module, and the second Drop Path module are connected in sequence. The empty frequency dual domain fusion module specifically includes a two-dimensional discrete fast Fourier transform module, an amplitude spectrum calculation module, a phase spectrum calculation module, a Gaussian filter module, an inverse Fourier transform module, and an inversion module. The first output end of the two-dimensional discrete fast Fourier transform module is connected with the input end of the amplitude spectrum calculation module. The second output end of the two-dimensional discrete fast Fourier transform module is connected with the input end of the phase spectrum calculation module. The output end of the phase spectrum calculation module is connected with the input end of the Gaussian filter module. The output end of the amplitude spectrum calculation module, the output end of the Gaussian filter module, and the input end of the inverse Fourier transform module are connected. The output end of the inverse Fourier transform module is connected with the input end of the inversion module.

5. The tailings ash content detection method based on space-frequency dual-domain fusion according to claim 4, characterized in that, The state space feature extraction module specifically comprises a first normalization module, a first linear mapping module, a two-dimensional convolution module, a first SiLU function, a second SiLU function, a state space block, a second linear mapping module, and a second normalization module, wherein the output end of the first normalization module is connected with the input end of the first linear mapping module, the first output end of the first linear mapping module is connected with the input end of the two-dimensional convolution module, the output end of the two-dimensional convolution module is connected with the first SiLU function, the second output end of the first linear mapping module is connected with the second SiLU function, the input end of the state space block is connected with the first SiLU function, the output end of the state space block, the second SiLU function, and the input end of the second linear mapping module are connected, and the output end of the second linear mapping module is connected with the input end of the second normalization module.

6. The tail coal ash content detection method based on space-frequency dual-domain fusion according to claim 5, characterized in that, The tail coal ash content detection model is used to detect the tail coal liquid flow image data set to obtain the concentration value and the ash classification of the slime mixed liquid, and the specific steps include: The tail coal liquid flow image data set is input into the tail coal ash content detection model; The tail coal liquid flow image data set is subjected to gray scale conversion processing based on the gray scale conversion module of the tail coal ash content detection model to obtain a tail coal liquid flow gray scale image; The tail coal liquid flow gray scale image is subjected to position coding arrangement from left to right based on the position embedding module of the tail coal ash content detection model to obtain an arranged tail coal liquid flow gray scale image; The arranged tail coal liquid flow gray scale image is subjected to space domain and frequency domain representation fusion based on the space-frequency dual-domain fusion module of the tail coal ash content detection model to obtain a fused tail coal liquid flow gray scale image; The fused tail coal liquid flow gray scale image is subjected to feature extraction based on the state space feature extraction module of the tail coal ash content detection model to obtain a tail coal liquid flow feature map; The tail coal liquid flow feature map is subjected to regularization processing based on the Drop Path module of the tail coal ash content detection model to obtain a regularized tail coal liquid flow feature map; The regularized tail coal liquid flow feature map is subjected to feature learning and weight training based on the feedforward neural network of the tail coal ash content detection model to obtain a probability class distribution of the tail coal liquid flow feature map; The probability class distribution of the tail coal liquid flow feature map is calculated based on the SoftMax function of the tail coal ash content detection model to obtain the concentration value and the ash classification of the slime mixed liquid.

7. The tailings ash content detection method based on space-frequency dual-domain fusion according to claim 6, characterized in that, The space-frequency dual-domain fusion module of the tail coal ash content detection model is used to perform space domain and frequency domain representation fusion on the arranged tail coal liquid flow gray scale image to obtain a fused tail coal liquid flow gray scale image, and the specific steps include: The arranged tail coal liquid flow gray scale image is input into the space-frequency dual-domain fusion module of the tail coal ash content detection model; The arranged tail coal liquid flow gray scale image is subjected to two-dimensional discrete fast Fourier transform based on the two-dimensional discrete fast Fourier transform module of the space-frequency dual-domain fusion module to obtain an amplitude graph and a phase graph; The amplitude graph is subjected to frequency intensity calculation based on the amplitude spectrum calculation module of the space-frequency dual-domain fusion module to obtain an amplitude spectrum; A phase spectrum calculation module based on the space-frequency dual-domain fusion module is configured to calculate the position information of the frequency of the phase image to obtain a phase spectrum; A Gaussian filtering module based on the space-frequency dual-domain fusion module is configured to perform Gaussian filtering processing on the phase spectrum to obtain a filtered phase spectrum; An inverse Fourier transform module based on the space-frequency dual-domain fusion module is configured to perform inverse Fourier transform on the amplitude spectrum and the filtered phase spectrum to obtain a reconstructed spatial domain image; An inverse phase module based on the space-frequency dual-domain fusion module is configured to perform an inverse phase operation on the reconstructed spatial domain image to obtain an inverse-phase spatial domain image; The inverse-phase spatial domain image and the reconstructed spatial domain image are subjected to an element-by-element addition operation, and then subjected to an element-by-element multiplication operation with the arranged tail coal liquid stream grayscale image to obtain a fused tail coal liquid stream grayscale image.

8. The tail coal ash content detection method based on space-frequency dual-domain fusion according to claim 7, characterized in that, The state space feature extraction module based on the tail coal ash content detection model is configured to perform feature extraction on the fused tail coal liquid stream grayscale image to obtain a tail coal liquid stream feature map, and the state space feature extraction module based on the tail coal ash content detection model comprises: The fused tail coal liquid stream grayscale image is input into the state space feature extraction module based on the tail coal ash content detection model; The first normalization module based on the state space feature extraction module is configured to perform normalization processing on the fused tail coal liquid stream grayscale image to obtain a normalized tail coal liquid stream grayscale image; The first linear mapping module based on the state space feature extraction module is configured to perform linear mapping on the normalized tail coal liquid stream grayscale image to obtain an x-vector tail coal liquid stream grayscale image and a y-vector tail coal liquid stream grayscale image; The two-dimensional convolution module, the first SiLU function and the state space block based on the state space feature extraction module are configured to perform two-dimensional convolution operation on the x-vector tail coal liquid stream grayscale image to extract local features, and perform hidden state calculation according to the local features to obtain an output of the inference next state of the x-vector; The second SiLU function based on the state space feature extraction module is configured to perform linear mapping on the y-vector tail coal liquid stream grayscale image to obtain a nonlinear feature of the y-vector; The second linear mapping module and the second normalization module based on the state space feature extraction module are configured to sequentially perform fusion, linear mapping and root mean square normalization operation on the output of the inference next state of the x-vector and the nonlinear feature of the y-vector to obtain a tail coal liquid stream feature map.

9. A tail coal ash content detection system based on space-frequency dual-domain fusion, characterized in that, The method comprises the following modules: A first module is configured to prepare a coal slime mixed liquid and obtain a tail coal liquid stream image dataset; A second module is configured to introduce a space-frequency dual-domain fusion module and a state space feature extraction module to construct a tail coal ash content detection model; A third module is configured to perform tail coal ash content detection on the tail coal liquid stream image dataset based on the tail coal ash content detection model to obtain a concentration value of the coal slime mixed liquid and a gray classification.

Images