A method for identifying the boundary of a spiral chute ore zone by dynamically adjusting weights and thresholds.

Abstract

This invention provides a method for identifying the boundary of a spiral chute ore zone by dynamically adjusting weights and thresholds, relating to the field of image processing technology. The method includes the following steps: acquiring spiral chute flow field images and their corresponding environmental parameters at different times, and preprocessing the images. The ore zone boundary contour is labeled manually and designated as a sample verification image. These sample verification images are mapped one-to-one with their corresponding sample images to form a training image set. The environmental parameters are normalized and subjected to sliding window filtering to construct an environmental influence parameter function. Using the sample images and corresponding environmental parameters from the training image set as input, and the sample verification images as labels, a deep learning model for edge detection is established. The weights and thresholds of the convolutional layers are dynamically adjusted, and the model is optimized through a loss function. Finally, the boundary of the spiral chute under test is identified. This scheme can achieve efficient boundary recognition under different environmental conditions, significantly improving the separation efficiency of spiral chutes.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, specifically to a method for identifying the boundary of a spiral chute mining zone by dynamically adjusting weights and thresholds. Background Technology

[0002] In recent years, with the rapid development of machine vision technology, the application of image processing algorithms in the field of mineral processing has provided new ideas for the automated sorting of spiral sluices. Existing research attempts to identify mineral zoning boundaries using traditional edge detection algorithms (such as Canny and Sobel) or deep learning-based edge detection models such as the Holistically-Nested Edge Detection (HED) network. However, the complex working conditions of existing spiral sluices pose a severe challenge to mineral zoning boundary detection: in existing technologies, the high-speed flow of slurry leads to blurred zoning boundaries, and traditional algorithms rely on static thresholds, making it difficult to adapt to gradient amplitude fluctuations caused by changes in flow velocity; secondly, uneven lighting and equipment vibration in industrial environments further exacerbate image noise, preventing effective improvement in the accuracy of existing algorithms. For example, although the optimized Canny algorithm improves noise resistance through adaptive dual thresholds, its linear filtering mechanism easily leads to the loss of high-frequency edge information; while deep learning methods based on HED networks perform well in multi-scale feature fusion, they lack robustness under dynamic conditions and cannot adjust detection sensitivity according to real-time environmental parameters.

[0003] To address the aforementioned issues, some patents propose adaptive threshold edge detection, which dynamically adjusts the threshold based on local image features to improve the robustness and accuracy of edge detection. For example, one method is based on local gradient statistics, determining high and low thresholds by calculating a histogram of gradient magnitudes within a local region. However, these adaptive thresholding methods still have limitations when dealing with the complex conditions of spiral chutes. They lack a systematic correlation between slurry operating environment parameters and edge detection thresholds, resulting in insufficient adaptability to high-speed flow scenarios. Therefore, in practical applications, further exploration of more effective adaptive thresholding methods is needed to improve the accuracy and efficiency of edge detection.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a method for identifying the boundary of a spiral chute ore zone by dynamically adjusting weights and thresholds, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A method for identifying the boundary of a spiral chute ore zone by dynamically adjusting weights and thresholds, comprising the following steps:

[0008] During the acquisition period, spiral chute flow field images of known spiral chute ore zone boundary contours at several different times are continuously acquired. At the same time, the environmental parameters corresponding to the acquisition of spiral chute flow field images are recorded. Each acquired spiral chute flow field image is preprocessed to obtain sample images. The environmental parameters include slurry flow velocity, light intensity and vibration intensity.

[0009] Based on the obtained sample images, the boundary contour of the spiral chute mining zone is marked in the sample images by manual labeling. The marked images are then binarized to obtain the corresponding binary edge images, which are denoted as sample verification images. The sample verification images are mapped one-to-one with the corresponding sample images to form a training image set.

[0010] The corresponding environmental parameters recorded when acquiring the spiral chute flow field image are normalized, and the normalized environmental parameters are subjected to sliding window filtering to obtain the precise values ​​of the environmental parameters at the corresponding time. Based on the precise values ​​of the environmental parameters, an environmental influence parameter function is generated, which includes a flow velocity influence function, an illumination influence function, and a vibration influence function.

[0011] A deep learning model for edge detection is established. The sample images in the training image set and the environmental parameters at the corresponding time are used as inputs, and the corresponding sample verification images in the training image set are used as labels to train the deep learning model for edge detection. The training specifically includes dynamically adjusting the fusion weights of each convolutional layer and the binarization threshold of the edge probability map based on the environmental influence parameter function, and optimizing the weight coefficients and scaling coefficients of each environmental parameter inversely through the loss function. This completes the multi-scale feature fusion of each convolutional layer and the binarization processing of the edge probability map of the output layer, and outputs a binary edge image.

[0012] After obtaining and preprocessing the environmental parameters and flow field images of the spiral chute to be identified, they are input into the trained edge detection deep learning model to identify the boundary of the spiral chute ore zone.

[0013] Furthermore, environmental parameters are collected by deploying velocity meters, light sensors, and vibration sensors on the surface of the spiral concentrator trough to obtain the slurry flow velocity v, light intensity l, and vibration intensity s during the collection period, where v ∈ [v...]. min , v max ], l∈[l min , l max ], s∈[s min s max ];v min and v maxThese represent the minimum and maximum flow velocities of the slurry during the sampling period, respectively. min and l max These represent the minimum and maximum light intensities during the data collection period, s. min and s max These represent the minimum and maximum vibration intensities during the data collection period, respectively.

[0014] Each acquired spiral chute flow field image is preprocessed to obtain sample images. The preprocessing includes: unifying image size, image enhancement and denoising preprocessing. Wavelet transform is used to denoise each spiral chute flow field image, and bilateral filtering is used to enhance each spiral chute flow field image.

[0015] The specific method for denoising using wavelet transform is as follows: The specific steps include: decomposing the spiral chute flow field image through wavelet transform to obtain wavelet coefficients of the image at different scales and directions; thresholding the wavelet coefficients, setting the low-amplitude wavelet coefficients to zero, and retaining the high-amplitude wavelet coefficients; performing inverse transform on the thresholded wavelet coefficients, and reconstructing the processed coefficients into an image to complete the image denoising process.

[0016] Bilateral filtering was used to enhance the details of the spiral chute flow field image. The specific formula used for the filtering transformation is as follows:

[0017]

[0018] In the formula, q is the coordinate vector in the image coordinate system, and I q Let B be the gray value at coordinate vector q. q grayscale value I q The grayscale value G after bilateral filtering transformation d and G r Both are Gaussian functions, where G d and G r The formula used is:

[0019]

[0020] In the formula, p is a coordinate vector in the image coordinate system, and I p Let σ be the gray value at coordinate vector p. d and σ r G d and G r The standard deviation.

[0021] Furthermore, the corresponding environmental parameters recorded when acquiring the spiral chute flow field images are normalized. The specific formula used for this normalization is as follows:

[0022]

[0023] In the formula, The normalized slurry flow rate, The normalized light intensity The normalized vibration intensity is represented by v, the collected slurry flow velocity data is represented by l, the collected light intensity data is represented by s, and the collected vibration intensity data is represented by s.

[0024] A sliding window filter is applied to the normalized environmental parameters to obtain the precise values ​​of the environmental parameters at the corresponding time points. The formula used to calculate the precise values ​​of the environmental parameters is as follows:

[0025]

[0026] In the formula, Let be the precise value of the slurry flow rate at time t. Let be the normalized slurry flow rate at time t. The normalized slurry flow rate at time t-1 This represents the precise value of the light intensity at time t. Let be the normalized light intensity at time t. The normalized light intensity at time t-1 This represents the precise value of the vibration intensity at time t. Let be the normalized vibration intensity at time t. Let be the normalized vibration intensity at time t-1, α be the smoothing coefficient, where 0 < α < 1, and t be the time variable within the acquisition period.

[0027] Furthermore, based on the precise values ​​of environmental parameters, environmental impact parameter functions are generated and constructed. These environmental impact parameter functions include flow velocity impact functions, illumination impact functions, and vibration impact functions, wherein the specific expressions of the environmental impact parameter functions are as follows:

[0028]

[0029]

[0030] In the formula, The value of the function relating flow velocity. The value of the illumination effect function. Let σ be the vibration influence function value, and σ be the illumination influence factor, where 0.1≤σ≤0.3.

[0031] Furthermore, the training of the edge detection deep learning model specifically includes dynamically adjusting the fusion weights of each convolutional layer and the binarization threshold of the edge probability map, completing the multi-scale feature fusion of each convolutional layer and the binarization processing of the edge probability map of the output layer. The specific logic for completing the multi-scale feature fusion of each convolutional layer is as follows:

[0032] The sample image is input into the edge detection deep learning model for convolutional processing. A corresponding edge probability map is generated after each convolutional layer. The fusion weights of each convolutional layer are dynamically adjusted based on environmental parameters. Through a multi-scale feature fusion mechanism, the edge probability maps of each layer are weighted and fused to obtain a fused edge probability map. The formula used to calculate the fused edge probability map is as follows:

[0033]

[0034] Among them, Y f (x, y) represents the fused edge probability map, where x is the x-coordinate of a pixel in the fused edge probability map, y is the y-coordinate of a pixel in the fused edge probability map, and X is the activation function, F. i W represents the edge probability map output by the i-th convolutional layer. i is the fusion weight of the i-th convolutional layer, m is the total number of convolutional layers in the edge detection deep learning model, and i is the index of the convolutional layer, where i∈[1,m];

[0035] The fusion weights W of the i-th convolutional layer i The specific formula used for the calculation is as follows:

[0036]

[0037] In the formula, ω v ω l and ω s These are the weighting coefficients for slurry flow rate, light intensity, and vibration intensity, respectively, k. i For the i-th marginal probability map F i The corresponding scaling factor is used to adjust the steepness of the weight distribution;

[0038] The formula used to calculate the activation function χ is as follows:

[0039]

[0040] Here, r represents the weighted input of the neuron.

[0041] Furthermore, the specific logic for generating the binary edge map is as follows: A binarization threshold is dynamically generated based on environmental parameters; the edge probability maps fused from each convolutional layer are then binarized to generate the final binary edge map. The formula used to calculate the final binary edge map is:

[0042]

[0043] In the formula, Y b (x,y) is a binary edge map, where T b The binarization threshold is T, where the binarization threshold is T. b The formula used for the calculation is:

[0044]

[0045] In the formula, T base The base threshold is 0.05 ≤ T. base ≤0.25.

[0046] Furthermore, the output binary edge verification image is compared with the corresponding sample verification image after manually marking and binarizing the spiral chute ore zone boundary contour, and the loss function is calculated. The formula for calculating the loss function is as follows:

[0047]

[0048] In the formula, φ is the loss function, and Y v (x,y) is a sample verification image after manually marking the boundary contour of the spiral chute ore zone and binarizing it;

[0049] The weighting coefficients and scaling coefficients of each environmental parameter are optimized in reverse based on the loss function. The formula used to adjust the weighting coefficients of each environmental parameter is as follows:

[0050]

[0051] In the formula, ω v ′、ω l ′ and ω s ′ represents the weighting coefficients for the optimized slurry flow rate, light intensity, and vibration intensity, respectively, and k i F' is the optimized i-th marginal probability map. i The corresponding scaling factor, where η is the learning rate.

[0052] Compared with the prior art, the present invention has the following beneficial effects:

[0053] First, in the data preprocessing stage, recording and normalizing environmental parameters followed by sliding window filtering effectively eliminates noise and improves the accuracy of these parameters. This enhances the quality of the input data, providing more reliable foundational data and enabling the model to learn and identify more accurately under different environmental conditions. Second, the environmental impact parameter function constructed based on these parameters effectively captures the influence of factors such as flow velocity, illumination, and vibration on the boundary of the spiral chute ore zone, improving the model's adaptability and generalization ability. Furthermore, during model training, dynamically adjusting the fusion weights of the convolutional layers and the binarization threshold of the edge probability map, through multi-scale feature fusion, makes the output edge probability map more accurate, generating high-quality binary edge images. This results in more stable recognition results that better reflect the actual boundary features of the ore zone. Finally, by comparing with manually labeled verification images, the loss function is calculated, and the weight coefficients of the environmental parameter convolutional layers are optimized in reverse, further improving the model's accuracy and reliability. This not only achieves efficient boundary recognition under different environmental conditions but also provides accurate data support for subsequent mineral processing, ultimately leading to a significant improvement in the separation efficiency of the spiral chute. Attached Figure Description

[0054] Figure 1 This is a schematic diagram of the overall method flow of the present invention. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0056] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0057] Example:

[0058] Please see Figure 1 , the present invention provides a technical solution:

[0059] A method for identifying the boundary of a spiral chute ore zone by dynamically adjusting weights and thresholds, comprising the following steps:

[0060] Step 1: During the acquisition period, continuously acquire spiral chute flow field images of the known spiral chute ore zone boundary contour at several different times. At the same time, record the environmental parameters corresponding to the acquisition of the spiral chute flow field images. Preprocess each acquired spiral chute flow field image to obtain sample images. The environmental parameters include slurry flow velocity, light intensity, and vibration intensity.

[0061] Environmental parameters are collected by deploying flow meters, light sensors, and vibration sensors on the surface of the spiral concentrator trough to obtain the slurry flow velocity v, light intensity l, and vibration intensity s during the collection period, where v ∈ [v...]. min , v max ], l∈[l min , l max ], s∈[s min s max ];v min and v max These represent the minimum and maximum flow velocities of the slurry during the sampling period, respectively. min and l max These represent the minimum and maximum light intensities during the data collection period, s. min and s max These represent the minimum and maximum vibration intensities within the sampling period, respectively, with the sampling interval set to 0.5 seconds.

[0062] Each acquired spiral chute flow field image is preprocessed to obtain sample images. The preprocessing includes: unifying image size, image enhancement, and denoising. Wavelet transform is used to denoise each spiral chute flow field image, and bilateral filtering is used to enhance each spiral chute flow field image.

[0063] The specific method for unifying image size is as follows: Select a uniform target size, such as width and height (e.g., 512x512 pixels or 256x256 pixels). This size should be chosen based on subsequent processing requirements and the availability of computing resources. Import the images to be processed using a programming language (such as Python) or image processing software (such as OpenCV, PIL, etc.) to obtain the original width and height of each image for scaling. Choose an appropriate scaling method according to specific needs. Common scaling methods include: Nearest Neighbor Interpolation: simple and fast, but may cause jagged edges; Bilinear Interpolation: considers the weighted average of the surrounding four pixels, with better results; Bicubic Interpolation: considers the 16 nearest pixels, usually providing smoother results.

[0064] A wavelet transform denoising method is used to denoise the spiral chute flow field image. The specific steps of the wavelet transform denoising method include: decomposing the spiral chute flow field image through wavelet transform to obtain wavelet coefficients of the image at different scales and directions; thresholding the wavelet coefficients, setting the low-amplitude wavelet coefficients to zero and retaining the high-amplitude wavelet coefficients; and performing an inverse transform on the thresholded wavelet coefficients to reconstruct the image from the processed coefficients, thus completing the image denoising process.

[0065] Bilateral filtering was used to enhance the details of the spiral chute flow field image. The specific formula used for the filtering transformation is as follows:

[0066]

[0067] In the formula, q is the coordinate vector in the image coordinate system, and I q Let B be the gray value at coordinate vector q. q grayscale value I q The grayscale value G after bilateral filtering transformation d and G r Both are Gaussian functions, where G d and G r The formula used is:

[0068]

[0069] In the formula, p is a coordinate vector in the image coordinate system, and I p Let σ be the gray value at coordinate vector p. d and σ r G d and G r The standard deviation.

[0070] Step 2: Based on the obtained sample images, the boundary contour of the spiral chute mining zone is marked in the sample images by manual labeling. The marked images are then binarized to obtain the corresponding binary edge images, which are denoted as sample verification images. The sample verification images are mapped one-to-one with the corresponding sample images to form a training image set.

[0071] The specific steps for manually labeling the boundary contours of the spiral chute ore zone in training images are as follows: Selecting a suitable image annotation tool is the first step in manual annotation. Commonly used image annotation tools include LabelMe and VGG Image Annotator (VIA). When annotating the boundary contours of the spiral chute ore zone, common annotation types include polygon annotation: suitable for irregular boundary contours, using polygons to accurately depict the boundary of the ore zone. Rectangular annotation: if the boundary of the ore zone is relatively regular, a rectangular box can be used for annotation.

[0072] The specific manual annotation process includes: importing randomly selected sample images into the annotation tool; ensuring the images have sufficient resolution and quality to clearly identify the ore zone boundaries; using the selected annotation tool, manually drawing the boundary contours based on the actual boundaries of the spiral chute, including: label classification: labeling the ore zone boundary as a category, such as "spiral chute ore zone boundary"; after annotation, exporting the annotated data to a standard format, such as Pascal VOC, COCO format, or a custom JSON format, for subsequent use in model training.

[0073] Step 3: Normalize the corresponding environmental parameters recorded when acquiring the spiral chute flow field image, and apply sliding window filtering to the normalized environmental parameters to obtain the accurate values ​​of the environmental parameters at the corresponding time. Based on the accurate values ​​of the environmental parameters, generate and construct environmental influence parameter functions, which include flow velocity influence function, illumination influence function and vibration influence function.

[0074] The environmental parameters recorded during the acquisition of spiral chute flow field images are normalized. The specific formula used for this normalization is as follows:

[0075]

[0076] In the formula, The normalized slurry flow rate, Normalized light intensity The normalized vibration intensity is represented by v, the collected slurry flow velocity data is represented by l, the collected light intensity data is represented by s, and the collected vibration intensity data is represented by s.

[0077] A sliding window filter is applied to the normalized environmental parameters to obtain the precise values ​​of the environmental parameters at the corresponding time points. The formula used to calculate the precise values ​​of the environmental parameters is as follows:

[0078]

[0079]

[0080]

[0081] In the formula, Let be the precise value of the slurry flow rate at time t. Let be the normalized slurry flow rate at time t. The normalized slurry flow rate at time t-1 This represents the precise value of the light intensity at time t. Let be the normalized light intensity at time t. The normalized light intensity at time t-1 This represents the precise value of the vibration intensity at time t. Let be the normalized vibration intensity at time t. Let be the normalized vibration intensity at time t-1, α be the smoothing coefficient, where 0 < α < 1, and t be the time variable within the acquisition period.

[0082] Based on the precise values ​​of environmental parameters, environmental impact parameter functions are generated and constructed. These functions include flow velocity impact functions, illumination impact functions, and vibration impact functions. The specific expressions for these environmental impact parameter functions are as follows:

[0083]

[0084]

[0085]

[0086] In the formula, The value of the function relating flow velocity. The value of the illumination effect function. Let σ be the vibration influence function value, and σ be the illumination influence factor, where 0.1≤σ≤0.3.

[0087] Step 4: Establish a deep learning model for edge detection. Use sample images from the training image set and environmental parameters at corresponding times as input, and use the corresponding sample verification images from the training image set as labels to train the deep learning model for edge detection. The training specifically includes dynamically adjusting the fusion weights of each convolutional layer and the binarization threshold of the edge probability map based on the environmental influence parameter function, and optimizing the weight coefficients and scaling coefficients of each environmental parameter inversely through the loss function to complete the multi-scale feature fusion of each convolutional layer and the binarization processing of the edge probability map of the output layer, and outputting a binary edge image.

[0088] The training specifically includes dynamically adjusting the fusion weights of each convolutional layer and the binarization threshold of the edge probability map, completing multi-scale feature fusion of each convolutional layer and binarization of the edge probability map of the output layer. The specific logic for completing multi-scale feature fusion of each convolutional layer is as follows:

[0089] The sample image is input into the edge detection deep learning model for convolutional processing. A corresponding edge probability map is generated after each convolutional layer. The fusion weights of each convolutional layer are dynamically adjusted based on environmental parameters. Through a multi-scale feature fusion mechanism, the edge probability maps of each layer are weighted and fused to obtain a fused edge probability map. The formula used to calculate the fused edge probability map is as follows:

[0090]

[0091] Among them, Y f (x, y) represents the fused edge probability map, where x is the x-coordinate of a pixel in the fused edge probability map, y is the y-coordinate of a pixel in the fused edge probability map, and χ is the activation function, F. i W represents the edge probability map output by the i-th convolutional layer. i is the fusion weight of the i-th convolutional layer, m is the total number of convolutional layers in the edge detection deep learning model, and i is the index of the convolutional layer, where i∈[1,m];

[0092] The fusion weights W of the i-th convolutional layer i The specific formula used for the calculation is as follows:

[0093]

[0094] In the formula, ω v ω l and ω s These are the weighting coefficients for slurry flow rate, light intensity, and vibration intensity, respectively, k. i For the i-th marginal probability map F i The corresponding scaling factor is used to adjust the steepness of the weight distribution.

[0095] The formula used to calculate the activation function χ is as follows:

[0096]

[0097] Here, r represents the weighted input of the neuron.

[0098] The specific logic for generating the binary edge map is as follows: A binarization threshold is dynamically generated based on environmental parameters. The edge probability maps fused from each convolutional layer are then binarized to generate the final binary edge map. The formula used to calculate the final binary edge map is:

[0099]

[0100] In the formula, Y b (x,y) is a binary edge map, where T b The binarization threshold is T, where the binarization threshold is T. b The formula used for the calculation is:

[0101]

[0102] In the formula, T base The base threshold is 0.05 ≤ T. base ≤0.25.

[0103] The output binary edge verification image is compared with the corresponding sample verification image after manually marking and binarizing the spiral chute ore zone boundary contour, and a loss function is calculated. The formula for calculating the loss function is as follows:

[0104]

[0105] In the formula, φ is the loss function, and Y v (x,y) is the sample verification image corresponding to the manually marked spiral chute ore zone boundary contour and binarized;

[0106] The weighting coefficients and scaling coefficients of each environmental parameter are optimized in reverse based on the loss function. The formula used to adjust the weighting coefficients of each environmental parameter is as follows:

[0107]

[0108] In the formula, ω v ′、ω l ′ and ω s ′ represents the weighting coefficients for the optimized slurry flow rate, light intensity, and vibration intensity, respectively, and k i F' is the optimized i-th marginal probability map. i The corresponding scaling factor, where η is the learning rate.

[0109] Step 5: After obtaining the environmental parameters and flow field images of the spiral chute to be identified and preprocessing them, input them into the trained edge detection deep learning model to identify the boundary of the spiral chute ore zone.

[0110] The optimized edge detection deep learning model employs a reinforcement learning algorithm to limit the magnitude of policy updates, maintaining training stability and efficiency. The calculation formula used in the reinforcement learning algorithm is as follows:

[0111]

[0112] In the formula, max θ E represents the maximum magnitude of the policy update, π θ (a|s) represents the probability of choosing action a in state u when the policy network parameters are θ, and π old (a|u) represents the probability distribution of the old policy network choosing action a in state u, where a is the increase or decrease of the weight coefficient, state u represents the magnitude of the loss function φ, A(u,a) is the advantage function, and ∈ is the cutoff coefficient used to limit the magnitude of policy updates and avoid drastic fluctuations, where 0.05≤∈≤0.3. The clip operation is used to limit the magnitude of policy updates, ensuring that the proportion of the new policy relative to the old policy is not too large during the update process, thereby preventing the model from becoming unstable due to excessive policy updates.

[0113] The process involves acquiring and preprocessing the environmental parameters and flow field images of the spiral chute to be identified, then inputting them into an optimized edge detection deep learning model to identify the boundary of the spiral chute's ore zone. Specifically, after acquiring the environmental parameters and flow field images of the spiral chute, the environmental parameters are normalized and filtered using a sliding window to obtain precise values ​​for the corresponding environmental parameters. Similarly, the flow field images are preprocessed using the same method and then input into the optimized edge detection deep learning model. Using the optimized parameters, the model outputs a binary image of the flow field of the spiral chute to be identified.

[0114] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0115] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0116] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0117] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. A method for identifying the boundary of a spiral chute ore zone by dynamically adjusting weights and thresholds, characterized in that, The specific steps include: During the acquisition period, spiral chute flow field images of known ore zone boundary contours at several different times are continuously acquired. At the same time, the environmental parameters corresponding to the acquisition of spiral chute flow field images are recorded. Each acquired spiral chute flow field image is preprocessed to obtain sample images. The environmental parameters include slurry flow velocity, light intensity and vibration intensity. Based on the obtained sample images, the boundary contour of the spiral chute mining zone is marked in the sample images by manual labeling. The marked images are then binarized to obtain the corresponding binary edge images, which are denoted as sample verification images. The sample verification images are mapped one-to-one with the corresponding sample images to form a training image set. The corresponding environmental parameters recorded when acquiring the spiral chute flow field image are normalized, and the normalized environmental parameters are subjected to sliding window filtering to obtain the accurate values ​​of the environmental parameters at the corresponding time. Based on the accurate values ​​of the environmental parameters, an environmental influence parameter function is constructed, which includes a flow velocity influence function, an illumination influence function, and a vibration influence function. A deep learning model for edge detection is established. The sample images in the training image set and the environmental parameters at the corresponding time are used as inputs, and the corresponding sample verification images in the training image set are used as labels to train the deep learning model for edge detection. The training specifically includes dynamically adjusting the fusion weights of each convolutional layer and the binarization threshold of the edge probability map based on the environmental influence parameter function, and optimizing the weight coefficients and scaling coefficients of each environmental parameter inversely through the loss function. This completes the multi-scale feature fusion of each convolutional layer and the binarization processing of the edge probability map of the output layer, and outputs a binary edge image. After obtaining and preprocessing the environmental parameters and flow field images of the spiral chute to be identified, they are input into the trained edge detection deep learning model to identify the boundary of the spiral chute ore zone.

2. The method for identifying the boundary of a spiral chute ore zone by dynamically adjusting weights and thresholds according to claim 1, characterized in that: Environmental parameters were collected by deploying flow meters, light sensors, and vibration sensors on the surface of the spiral chute to obtain the slurry flow rate during the collection period. Light intensity and vibration intensity ,and , , ; and These represent the minimum and maximum flow velocities of the slurry during the sampling period, respectively. and These represent the minimum and maximum light intensity during the data collection period, respectively. and These represent the minimum and maximum vibration intensities during the data collection period, respectively.

Images