Method for recovering ore spots from low-power ore images, electronic device, and storage medium

Abstract

The application belongs to the technical field of artificial intelligence, and particularly relates to a low-power ore image ore point recovery method, an electronic device and a storage medium. A first metal wedge image and a first ore image set are collected under a first data collection condition, a second metal wedge image and a second ore image set are collected under a second data collection condition, and a training data set is constructed. The first data collection condition includes a first X-ray power and a first conveying speed, the second data collection condition includes a second X-ray power and a second conveying speed, the first X-ray power is greater than the second X-ray power, and the first conveying speed is less than the second conveying speed. A convolutional neural network model is trained based on the training data set to obtain an ore point recovery model. An ore image to be recovered is collected under the second data collection condition, and the ore image to be recovered is input into the ore point recovery model to obtain an ore point recovery image, thereby realizing ore point recovery of a low-power ore image.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of artificial intelligence, and particularly relates to a low-power ore image ore point recovery method, an electronic device and a storage medium. BACKGROUND

[0002] In high-altitude areas, with the increase of altitude, atmospheric pressure and air density gradually decrease. This environmental change will have a significant impact on the operation of power equipment. First, the poor heat dissipation conditions caused by thin air can easily cause overheating problems of the equipment. Second, the electrical insulation strength in high-altitude areas will also be reduced, increasing the risk of equipment failure. In order to cope with the challenges brought by high-altitude environment, reducing the power operation of power equipment has become an effective preventive measure. By reducing the power, the operating load of the equipment can be reduced, thereby reducing the risk of overheating of the equipment. At the same time, reducing the power also helps to improve the electrical insulation strength and reduce the possibility of equipment failure. Therefore, in high-altitude areas, the X-ray power of the X-ray transmission (XRT) equipment is usually reduced to collect ore images.

[0003] However, the ore images collected by the XRT equipment in the low-power state of X-rays have the following problems: the higher the power of the X-rays of the XRT equipment, the stronger the penetration, which can significantly improve the signal-to-noise ratio of weak absorption areas, and the image is naturally clearer, which is conducive to identifying the ore points in the ore image, thereby better performing ore waste identification; the penetration ability of the ore image collected by the low-power X-rays is not enough, the image does not show the ore points, and there are a large number of noise points, which seriously affect the precipitation of the ore points in the image. Since the collected ore image cannot see the ore points, subsequent ore point extraction and sorting cannot be performed, which greatly reduces the accuracy of ore waste sorting in high-altitude environments.

[0004] Therefore, how to recover the ore points of the ore image collected by the XRT equipment in the low-power state of X-rays to realize the ore waste sorting in high-altitude environments has become a problem to be solved. SUMMARY

[0005] The application provides a low-power ore image ore point recovery method, an electronic device and a storage medium, which aims to recover the ore points of the ore image collected by the XRT equipment in the low-power state of X-rays to realize the ore waste sorting in high-altitude environments.

[0006] In a first aspect, the embodiments of the application provide a low-power ore image ore point recovery method, which comprises:

[0007] based on the sample metal wedge, a first metal wedge image and a second metal wedge image are collected under a first data collection condition and a second data collection condition respectively, the first data collection condition includes a first X-ray power and a first conveying speed, the second data collection condition includes a second X-ray power and a second conveying speed, the first X-ray power is greater than the second X-ray power, and the first conveying speed is less than the second conveying speed;

[0008] based on the sample ore, a first ore image set and a second ore image set are collected under a first data collection condition and a second data collection condition respectively;

[0009] based on the first metal wedge image, the second metal wedge image, the first ore image set and the second ore image set, a training data set is constructed, and a convolutional neural network model is trained based on the training data set to obtain a mineral point recovery model;

[0010] an image of the ore to be recovered is collected under the second data collection condition, and the image of the ore to be recovered is input into the mineral point recovery model to obtain a mineral point recovery image.

[0011] Further, the first metal wedge image and the second metal wedge image are collected under the first data collection condition and the second data collection condition, comprising:

[0012] at a fixed position of the conveying device, the sample metal wedge is placed, the conveying device is controlled to run at a first conveying speed, and a first metal wedge image is collected under a first X-ray power condition; the conveying device is controlled to run at a second conveying speed, and a second metal wedge image is collected under a second X-ray power condition.

[0013] Further, the first metal wedge image and the second metal wedge image are collected under the first data collection condition and the second data collection condition, comprising:

[0014] at a certain position of the conveying device, a first sample metal wedge is placed, a plurality of first metal wedge images of the first sample metal wedge are obtained under a first collection condition; a plurality of second metal wedge images of the first sample metal wedge are obtained under a second collection condition;

[0015] at the same position of the conveying device, a second sample metal wedge is placed, a plurality of first metal wedge images of the second sample metal wedge are obtained under the first collection condition; a plurality of second metal wedge images of the second sample metal wedge are obtained under the second collection condition;

[0016] The first metal wedge image is obtained based on a plurality of first metal wedge images of the first sample metal wedge and a plurality of first metal wedge images of the second sample metal wedge; and the second metal wedge image is obtained based on a plurality of second metal wedge images of the first sample metal wedge and a plurality of second metal wedge images of the second sample metal wedge.

[0017] The density of the first sample metal wedge and the second sample metal wedge is greater than a set threshold.

[0018] Further, the training data set is constructed based on the first metal wedge image, the second metal wedge image, the first ore image set and the second ore image set, including:

[0019] The pixel fitting curve is determined based on the first metal wedge image and the second metal wedge image.

[0020] After adding noise to each first ore image in the first ore image set and adding X-ray transmission interference based on the pixel fitting curve, the response value matching is performed with any one or more second ore images in the second ore image set, to obtain a matched ore image set.

[0021] The training data set is constructed based on the first ore image set and the matched ore image set.

[0022] Further, the ore point recovery model includes a model input module, a global feature extraction module, a local feature extraction module, an image recovery module and a model output module connected in sequence.

[0023] The model input module is used for inputting the ore image to be recovered.

[0024] The global feature extraction module is used for extracting the global feature of the ore image to be recovered.

[0025] The local feature extraction module is used for extracting the local feature of the ore image to be recovered based on the global feature.

[0026] The image recovery module is used for performing feature fusion on the local feature to obtain a high-resolution recovery image.

[0027] The model output module is used for outputting the ore point recovery image based on the high-resolution recovery image.

[0028] Further, the global feature extraction module includes a head convolution layer, a first deep separable convolution layer, a channel attention layer, a normalization layer and a first feedforward neural network layer connected in sequence, and a feature splicing layer.

[0029] The head convolution layer is connected with the model input module.

[0030] The feature splicing layer is connected with the head convolutional layer and the first feedforward neural network layer, and is configured to splice outputs of the head convolutional layer and the first feedforward neural network layer to obtain global features.

[0031] Further, the local feature extraction module includes a second deep separable convolutional layer and a plurality of multi-resolution feature extraction units connected with the second deep separable convolutional layer and arranged side by side.

[0032] The second deep separable convolutional layer is connected with the global feature extraction module, and is configured to extract high-resolution features of the image to be restored based on the global features.

[0033] The multi-resolution extraction unit is configured to extract each resolution feature of the image to be restored based on the high-resolution features.

[0034] Further, the image restoration module includes a feature fusion layer, an attention mechanism layer, a bottom-up fusion layer, and a deconvolution or interpolation layer connected in sequence.

[0035] The feature fusion layer and the attention mechanism layer are configured to perform feature fusion on each resolution feature, and obtain a multi-resolution feature set through the attention mechanism.

[0036] The bottom-up fusion layer and the deconvolution / interpolation layer are configured to perform bottom-up fusion on the multi-resolution feature set, and then perform a reverse operation to obtain a high-resolution restored image.

[0037] In a second aspect, an electronic device is provided, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor implements the method of the first aspect or any of the implementation manners thereof when executing the computer program.

[0038] In a third aspect, a computer-readable storage medium is provided, which stores a computer program. The computer program is executable by a processor to implement the method of the first aspect or any of the implementation manners thereof.

[0039] In a fourth aspect, a computer program product is provided. When the computer program product is executed on an electronic device, the electronic device performs the method of the first aspect or any of the implementation manners thereof.

[0040] Compared with the prior art, the embodiment of the present application has the beneficial effects that: the first metal wedge image and the first ore image set are collected under the data collection condition of high X-ray power and low conveying speed, the second metal wedge image and the second ore image set are collected under the data collection condition of low X-ray power and high conveying speed, and a training data set is constructed, a convolutional neural network model is trained based on the training data set, and a mineral point recovery model is obtained, so that the mineral point recovery model can perform mineral point recovery on the input ore image collected under the working condition of low X-ray power and high conveying speed, and accurately recover the mineral points that can only be displayed under the working condition of high X-ray power and low conveying speed, thereby solving the problem that the low-power ore image cannot display the mineral points, and the mineral waste cannot be sorted in a high-altitude environment.

[0041] It can be understood that the electronic device, computer readable storage medium and computer program product provided by the embodiment of the present application have the same beneficial effects as the above-mentioned low-power ore image mineral point recovery method, which will not be described here. BRIEF DESCRIPTION OF DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiments or prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0043] Figure 1 A flowchart of a low-power ore image mineral point recovery method provided by an embodiment of the present application is shown in the figure.

[0044] Figure 2 A schematic diagram of an ore sorting device provided by an embodiment of the present application is shown in the figure.

[0045] Figure 3 A flowchart of an implementation of S130 in a low-power ore image mineral point recovery method provided by an embodiment of the present application is shown in the figure.

[0046] Figure 4 A schematic diagram of the change of a perception loss and mean square error weighted combination loss value of a mineral point recovery model provided by an embodiment of the present application with the number of iterations is shown in the figure.

[0047] Figure 5 A schematic diagram of the change of a signal-to-noise ratio of a mineral point recovery model output image provided by an embodiment of the present application with the number of iterations is shown in the figure.

[0048] Figure 6A schematic diagram of changes in structural similarity between a restored ore point model output image and a high-power ore image with the number of iterations is provided for an embodiment of the present application.

[0049] Figure 7 A schematic diagram of an architecture of an ore restoration model is provided for an embodiment of the present application.

[0050] Figure 8 A schematic diagram of a channel attention layer in an ore restoration model is provided for an embodiment of the present application.

[0051] Figure 9 A schematic diagram of a to-be-restored ore image, a restored ore point detail image, a false color image, and a restored ore point identification image is provided for an embodiment of the present application.

[0052] Figure 10 A comparison schematic diagram of a to-be-restored ore image and a restored ore point image is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0053] In the following description, for purposes of explanation and not limitation, specific details are set forth, such as a particular system architecture, techniques, etc., in order to provide a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application can be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present application with unnecessary detail.

[0054] It should be understood that the term “includes” when used in the specification and the appended claims herein, specifies the presence of stated features, integers, steps, operations, elements, and / or components but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0055] It should also be understood that the term “and / or” when used in the specification and the appended claims herein, means any one or more of the associated listed items or a combination thereof.

[0056] As used in the description of the application and the appended claims herein, the term “if’ can be interpreted as meaning “when” or “upon” or “in response to a determination” or “in response to a detection” depending on the context. Similarly, the phrase “if determined” or “if detected [the described condition or event]” can be interpreted as meaning “upon a determination” or “in response to a determination” or “upon a detection of [the described condition or event]” or “in response to a detection of [the described condition or event]” depending on the context.

[0057] In addition, in the description of the present application and the appended claims, the terms "first", "second", "third", etc. are used only to distinguish descriptions and cannot be understood as indicating or implying relative importance.

[0058] Reference to "one embodiment" or "some embodiments" or "one implementation" or "some implementations" or "one example" or "some examples" described in the present application means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the present application. Therefore, the phrases "in one embodiment", "in some embodiments", "in other some embodiments", "in yet some embodiments", etc. appearing in various places in the specification are not necessarily all referring to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically stated. The terms "comprise", "include", "have" and their conjugates mean "including but not limited to", unless otherwise specifically stated.

[0059] For the convenience of understanding, the technical solutions of the present application will be described in detail below with reference to the accompanying drawings.

[0060] Figure 1 A low-power ore image ore point recovery method is provided for an embodiment of the present application. As shown in Figure 1 For the convenience of description, only parts related to the present embodiment are shown, and the method provided by the present embodiment includes the following steps:

[0061] S110, based on the sample metal wedge, respectively collecting a first metal wedge image and a second metal wedge image under a first data acquisition condition and a second data acquisition condition, the first data acquisition condition including a first X-ray power and a first conveying speed, the second data acquisition condition including a second X-ray power and a second conveying speed, the first X-ray power being greater than the second X-ray power, and the first conveying speed being less than the second conveying speed.

[0062] In a possible implementation, an ore sorting device is used to collect a first metal wedge image and a second metal wedge image of a sample metal wedge under a first data acquisition condition and a second data acquisition condition. As shown in Figure 2 The ore sorting device includes a feeding device 1, a belt 2, an XRT device (including an X-ray source 3 and a detection card 4), an air nozzle 5, and a bin 6.

[0063] Specifically, a metal wedge with a length greater than a preset length threshold and a slope height within a preset height range is selected as a sample metal wedge, so that the first metal wedge image and the second metal wedge image collected based on the sample metal wedge can cover all pixel values.

[0064] Preferably, the preset length threshold is 1 meter, or when half of the detection surface of the belt 2 is greater than 1 meter, the preset length threshold is half of the detection surface of the belt 2; and the preset height range is 1-30 millimeters.

[0065] In a specific implementation, the sample metal wedge is fed onto the belt 2 by the feeding device 1, the X-ray power of the X-ray source 3 is set to a first X-ray power, the rotating speed of the belt 2 is set to a first conveying speed, when the belt 2 starts to rotate, the sample metal wedge is line scanned by a line scanning camera (not shown in the figure) until the sample metal wedge falls into the hopper 6, and a first metal wedge image is collected; similarly, the sample metal wedge is fed onto the belt 2 again by the feeding device 1, the X-ray power of the X-ray source 3 is set to a second X-ray power, the rotating speed of the belt 2 is set to a second conveying speed, when the belt 2 starts to rotate, the sample metal wedge is line scanned by the line scanning camera until the sample metal wedge falls into the hopper 6, and a second metal wedge image is collected.

[0066] Preferably, the first X-ray power is 2000 watts (w), the first conveying speed is 0.2 meters per second (m / s), the second X-ray power is 500 w, and the second conveying speed is 3 m / s.

[0067] More specifically, the sample metal wedge can be one or more metal wedges, which can be a pure copper wedge, a pure iron wedge, a pure titanium wedge, etc.; preferably, the sample metal wedge can include two wedges, and the difference between the metal densities of the two wedges is greater than a set threshold; for example, the sample metal wedge includes a large-area pure titanium wedge and a large-area pure iron wedge.

[0068] Preferably, S110 includes:

[0069] The sample metal wedge is placed at a fixed position of the conveying device, such as a certain position of the belt, the conveying device is controlled to operate at a first conveying speed, a first metal wedge image is collected under a first X-ray power condition, and a second metal wedge image is collected under a second X-ray power condition.

[0070] Specifically, the sample metal wedges can be placed in sequence at the same position, a plurality of first metal wedge images of each sample metal wedge are obtained under a first collection condition, preferably 2-3 images to remove interference after denoising, and a plurality of second metal wedge images of each sample metal wedge are obtained under a second collection condition; specifically, each sample metal wedge can be a pure copper wedge or several different pure metal wedges; each type of pure metal wedge can be photographed one image, or 2-3 images to form an image set, and then denoised to remove interference. Importantly, the sample metal wedge should be placed at a fixed position of the conveying device to ensure data alignment.

[0071] More preferably, S110 can optionally include:

[0072] S111: placing the first sample metal wedge at a certain position of the conveying device, such as a certain position of the belt, obtaining a plurality of first metal wedge images of the first sample metal wedge under the first acquisition condition, preferably 2-3 images, to remove interference by denoising; and obtaining a plurality of second metal wedge images of the first sample metal wedge under the second acquisition condition;

[0073] S112: placing the second sample metal wedge at the same position of the conveying device, such as the same position on the belt, obtaining a plurality of first metal wedge images of the second sample metal wedge under the first acquisition condition, preferably 2-3 images, to remove interference by denoising; and obtaining a plurality of second metal wedge images of the second sample metal wedge under the second acquisition condition;

[0074] S113: obtaining a first metal wedge image based on the plurality of first metal wedge images of the first sample metal wedge and the plurality of first metal wedge images of the second sample metal wedge; and obtaining a second metal wedge image based on the plurality of second metal wedge images of the first sample metal wedge and the plurality of second metal wedge images of the second sample metal wedge;

[0075] The density of the first sample metal wedge and the second sample metal wedge differs by more than a set threshold value.

[0076] In the above embodiment, the preferred embodiment of the present application for collecting the first metal wedge image and the second metal wedge image is given, which includes at least four core invention points; 1. The structure of the sample metal is not a simple and conventional selection, but an important discovery in the invention process; only the wedge structure can ensure that the collected image can present almost all pixel values for subsequent correction of the training set. 2. The first data acquisition condition simulates the conventional environment of high power and low speed, and the second data acquisition condition simulates the harsh environment of low power and high speed (such as reducing power to improve safety performance in high-altitude areas), to obtain different data under two environments for subsequent recovery of ore spots in low-power ore images. 3. In the preferred scheme, each sample metal wedge is placed at the same position of the conveying device for data alignment, which can ensure that the collected data is at the same position each time to form matched data. 4. The material of the sample metal in the preferred scheme includes two metals with large density difference, which can further cover all pixel values to further improve the subsequent correction accuracy.

[0077] S120, based on the sample ore, collecting a first ore image set and a second ore image set under the first data acquisition condition and the second data acquisition condition, respectively.

[0078] Specifically, a large number of ores are selected as sample ores, the more the number of sample ores, the richer the training data set constructed subsequently, and the more accurate the ore point recovery model obtained by using the training data set for model training.

[0079] Preferably, the number of sample ores is at least 10,000, and the sample ores are copper ores.

[0080] In specific implementation, the X-ray power of the X-ray source 3 is set to a first X-ray power, and the rotating speed of the belt 2 is set to a first conveying speed. When the belt 2 starts to rotate, the sample ores are continuously fed to the belt 2 through the feeding device 1, and the sample ores on the belt 2 are line-scanned by the line-scan camera until all the sample ores fall into the hopper 6, a plurality of first ore images are collected and a first ore image set is constructed. Similarly, the X-ray power of the X-ray source 3 is set to a second X-ray power, and the rotating speed of the belt 2 is set to a second conveying speed. When the belt 2 starts to rotate, the sample ores are continuously fed to the belt 2 again through the feeding device 1, and the sample ores on the belt 2 are line-scanned by the line-scan camera until all the sample ores fall into the hopper 6, a plurality of second ore images are collected and a second ore image set is constructed.

[0081] In S130, a training data set is constructed based on the first metal wedge block image, the second metal wedge block image, the first ore image set and the second ore image set, and a convolutional neural network model is trained based on the training data set to obtain an ore point recovery model.

[0082] In a possible implementation, the training data set is constructed based on the first metal wedge block image, the second metal wedge block image, the first ore image set and the second ore image set, which can optionally include the following steps:

[0083] Based on the first metal wedge block image and the second metal wedge block image, a pixel fitting curve is determined.

[0084] After adding noise to each first ore image in the first ore image set and adding X-ray transmission interference based on the pixel fitting curve, the response value of the first ore image is matched with that of any one or more second ore images in the second ore image set to obtain a matched ore image set.

[0085] The training data set is constructed based on the first ore image set and the matched ore image set.

[0086] In specific implementation, as Figure 3As shown, based on the first metal wedge image and the second metal wedge image, the pixel relationship corresponding to each transmission is calculated by the two-point correction method, most pixel values can be covered, all corresponding pixels are fitted to determine the pixel fitting curve; the first ore image set is processed, the pixel fitting curve is used to increase the X-ray transmission interference, and the pine and / or Gaussian distribution is used to increase the mixed noise points, data reconstruction is performed, and the response value is matched with any one or more second ore images in the second ore image set to obtain a matched ore image set matched with the second ore image set, and the first ore image set and the matched ore image set are constructed as a training data set.

[0087] Notably, the key of the present application lies in steps S110-S120, how to design two sampling conditions to obtain the first metal wedge image, the second metal wedge image, the first ore image set and the second ore image set, and how to construct the training data set accordingly; in the preferred scheme, the way of data reconstruction and response value matching by the pixel fitting curve is given to construct the training data set. As for the specific structure of the convolutional neural network model and the training method, any method of the prior art can be used, and is not limited to the following examples.

[0088] Further, the designed convolutional neural network model can be trained and tested using the training data set, and the loss function is preferably designed as a combination of perceptual loss and mean square error loss to achieve complementarity, improve the ore point recovery capability, and reduce image blur, and then an optimal model is obtained by continuously adjusting the parameters as an ore point recovery model. The trained ore point recovery model is converted from the PyTorch model file (Python Pickle Torch Model, PTH) format to the Open Neural Network Exchange (ONNX) format, and then converted to the Offline Model (OM) format by the Huawei Ascend Tensor Compiler (ATC) tool, and finally deployed on the Huawei AI accelerator card.

[0089] As an example, Figure 4 to Figure 6 The figures respectively show the changes of the perceptual loss and the weighted combination loss value of the mean square error of the model, the signal-to-noise ratio of the model output image, and the structural similarity with the iteration number during the model training process. As can be seen from the figures, the perceptual loss and the weighted combination loss value of the mean square error of the model tend to 0 after 120 iterations; the signal-to-noise ratio of the model output image reaches a maximum value of 42 after 120 iterations; and the structural similarity between the low-power ore image output by the model and the high-power ore image stabilizes at 0.950 after 120 iterations.

[0090] In a possible implementation, as shown in Figure 7 the ore spot restoration model comprises, in sequence, a model input module, a global feature extraction module, a local feature extraction module, an image restoration module, and a model output module;

[0091] The model input module is configured to input the ore image to be restored.

[0092] The global feature extraction module is configured to extract global features of the ore image to be restored.

[0093] The local feature extraction module is configured to extract local features of the ore image to be restored based on the global features.

[0094] The image restoration module is configured to perform feature fusion on the local features to obtain a high-resolution restored image.

[0095] The model output module is configured to output an ore spot restoration image based on the high-resolution restored image.

[0096] Preferably, the global feature extraction module comprises, in sequence, a head convolutional layer, a first deep separable convolutional layer, a channel attention layer, a normalization layer, a first feedforward neural network layer, and a feature concatenation layer.

[0097] The head convolutional layer is connected to the model input module.

[0098] The feature concatenation layer is connected to the head convolutional layer and the first feedforward neural network layer, and is configured to concatenate the outputs of the head convolutional layer and the first feedforward neural network layer to obtain the global features.

[0099] Specifically, the head convolutional layer uses a 3x3 convolutional kernel to map the input image from a low-dimensional space to a high-dimensional feature space while keeping the original resolution of the image unchanged, thereby increasing the feature representation capability of the original image. The first deep separable convolutional layer is used to extract features. The channel attention layer is used to weight all channels to highlight the ore spot features. The normalization layer and the first feedforward neural network layer are used to increase the nonlinear capability of the feature set and the expression capability of complex features, thereby compensating for the linear limitations of the attention mechanism. The feature concatenation layer is used to perform feature fusion on the outputs of the head convolutional layer and the first feedforward neural network layer, thereby enhancing the high-resolution feature representation capability and obtaining the global features.

[0100] As an example, as shown in Figure 8As shown, the structure of the channel attention layer is as follows: after inputting the features via the input layer, global average pooling and global maximum pooling are performed on the features via the globally average-pooled layer and the globally maximum-pooled layer arranged side by side, respectively, to generate two groups of pooled features, which are spliced and then pooled and fused via a first convolutional layer with a convolution kernel of 3x3, are reduced in dimension via a fully connected layer, have the number of channels changed via a second convolutional layer with a convolution kernel of 1x1, pass through an activation function sigmoid, and finally are multiplied by the input features to obtain an output feature set, which is output by the output layer.

[0101] Preferably, the local feature extraction module comprises a second deep separable convolutional layer and a plurality of multi-resolution feature extraction units connected to the second deep separable convolutional layer and arranged side by side.

[0102] The second deep separable convolutional layer is connected to the global feature extraction module and is configured to extract high-resolution features of the image to be restored based on the global features.

[0103] The multi-resolution feature extraction unit is configured to extract features of each resolution of the image to be restored based on the high-resolution features. Preferably, different resolution features can be extracted via a down-sampling layer and an up-sampling layer based on the high-resolution features.

[0104] Preferably, the image restoration module comprises a feature fusion layer, an attention mechanism layer, a bottom-up fusion layer, and a deconvolution / interpolation layer connected in sequence.

[0105] The feature fusion layer and the attention mechanism layer are configured to perform feature fusion on the features of each resolution, and obtain a multi-resolution feature set via the attention mechanism.

[0106] The bottom-up fusion layer and the deconvolution / interpolation layer are configured to perform bottom-up fusion on the multi-resolution feature set, and then perform a reverse operation to obtain a high-resolution restored image.

[0107] Preferably, the model output module comprises a third deep separable convolutional layer, a dynamic sparse attention layer, a second feedforward neural network layer, a residual connection layer, and a Tail convolutional layer connected in sequence, and the third deep separable convolutional layer is connected to the image restoration module.

[0108] Specifically, the third deep separable convolutional layer and the dynamic sparse attention layer can effectively reduce the computational complexity, and then the second feedforward neural network layer and the residual connection layer are used to integrate the global and local features and ensure the stability of the features. Finally, the Tail convolutional layer is used to output the restored image of the ore.

[0109] The design of the global feature extraction module and the local feature extraction module in the ore spot recovery model provided in the application is respectively used to extract the global feature and the local feature of the ore image to be recovered, and then a high-resolution recovered image is obtained through the image recovery module, so that the ore spot recovery model can extract rich feature information of the ore image to be recovered, improve the ore spot recovery capability, and further improve the accuracy of the ore spot recovery image.

[0110] In S140, the ore image to be recovered is collected under the second data collection condition, and the ore image to be recovered is input into the ore spot recovery model to obtain an ore spot recovery image.

[0111] In an application scenario, after the XRT device is deployed to a high-altitude area, the on-site X-ray power is set to 500w, the conveying speed is set to 3m / s, and the ore image to be recovered is collected. The ore image to be recovered is input into the trained ore spot recovery model for inference to obtain an ore recovery image with high ore spot, high signal-to-noise ratio and high structural similarity, which is comparable to the ore recovery image collected under the laboratory environment with X-ray power of 2000W and conveying speed of 0.2m / s.

[0112] As an example, Figure 9 is a schematic diagram of an ore image to be recovered, an ore spot recovery detail image, a pseudo-color image and an ore spot identification image, wherein Figure 9 a is the ore image to be recovered, Figure 9 b is the ore spot recovery detail image, Figure 9 c is the pseudo-color image, Figure 9 d is the ore spot identification image. Figure 10 is a comparison schematic diagram of an ore image to be recovered and an ore spot recovery image obtained by inputting the ore image to be recovered into the trained ore spot recovery model, wherein Figure 10 a is the ore image to be recovered, Figure 10 b is the ore spot recovery image.

[0113] Further, after obtaining the ore spot recovery image, the ore spot is extracted through the image processing beneficiation algorithm to generate a binary ore spot image, and the ore waste can be sorted based on the binary ore spot image.

[0114] The technical scheme provided by the embodiment is used to respectively collect a first metal wedge image and a first ore image set under the data collection condition of high X-ray power and low conveying speed, collect a second metal wedge image and a second ore image set under the data collection condition of low X-ray power and high conveying speed, and construct a training data set. A convolutional neural network model is trained based on the training data set, and a mineral point recovery model is obtained. The mineral point recovery model can perform mineral point recovery on an input ore image collected under the condition of low X-ray power and high conveying speed, and accurately recover mineral points that can only be shown under the condition of high X-ray power and low conveying speed. The problem that mineral points cannot be shown in a low-power ore image and the problem that sorting of ore and waste cannot be performed in a high-altitude environment are solved.

[0115] In summary, the technical scheme provided by the application uses a laboratory device as a carrier to design a mineral point recovery method for a low-power ore image, and the purpose is to extract mineral points that cannot be seen in an ore image collected under the condition of low X-ray power and successfully apply it in a high-altitude area. In a laboratory environment, data sets of metal wedge and ore images under different powers and different conveying speeds are collected, a set of matched ore image sets is obtained through curve fitting and noise point matching, and is used as a training data set. A mineral point recovery model is designed to recover mineral points that cannot be extracted by a low-power ore image, and the final effect is very good, and the mineral point extraction result is obvious.

[0116] The technical scheme provided by the application has the following technical effects: (1) the device can be applied to sorting of ore and waste in a 5000-meter high-altitude area when the X-ray power is reduced by 4 times, and the cost can be greatly saved; (2) the recognition accuracy of ore is improved, and the ore can be applied to high-precision ore; (3) the low-power ore image after mineral point recovery is no longer black, the definition is greatly improved, a large number of mineral points can be displayed, and the accuracy of sorting of ore and waste is improved to more than 90%; (4) the image reconstruction speed is as high as 50 frames per second (fps), and the time consumption requirement of the device is met.

[0117] On the other hand, the application further provides a computer storage medium storing executable program code; the executable program code is used to execute any of the mineral point recovery methods for a low-power ore image.

[0118] On the other hand, the application further provides an electronic device including a memory and a processor; the memory stores program code executable by the processor; the program code is used to execute any of the mineral point recovery methods for a low-power ore image.

[0119] For example, the program code can be segmented into one or more modules / units, one or more modules / units are stored in the memory and executed by the processor to complete the present application. One or more modules / units can be a series of computer program instructions capable of completing a specific function, which is used to describe the execution process of the program code in the electronic device.

[0120] The electronic device can be a desktop computer, a notebook computer, a palm computer, a cloud server and the like. The electronic device can include, but is not limited to, a processor, a memory. Those skilled in the art can understand that the electronic device can also include an input / output device, a network access device, a bus and the like.

[0121] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0122] The memory can be an internal storage unit of the electronic device, such as a hard disk or a memory. The memory can also be an external storage device of the electronic device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card and the like. Further, the memory can include both the internal storage unit and the external storage device of the electronic device. The memory is used to store program codes and other programs and data required by the electronic device. The memory can also be used to temporarily store data that has been output or will be output.

[0123] The above computer storage medium and electronic device are created based on the above method, and the technical effects and advantages thereof will not be repeated here. The technical features of the above embodiments can be combined arbitrarily. In order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the present application.

[0124] The above embodiments only express several implementation manners of the present application, and the description is relatively specific and detailed, but it should not be understood as a limitation on the patent scope of the present application. It should be noted that, for ordinary skilled persons in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A method for mineral point recovery from low-power ore images, characterized in that, include: Based on the sample metal wedge, images of the first metal wedge and the second metal wedge are acquired under the first data acquisition condition and the second data acquisition condition, respectively. The first data acquisition condition includes the first X-ray power and the first transport speed, and the second data acquisition condition includes the second X-ray power and the second transport speed. The first X-ray power is greater than the second X-ray power, and the first transport speed is less than the second transport speed. Based on the sample ore, a first ore image set and a second ore image set were acquired under the first data acquisition condition and the second data acquisition condition, respectively. A training dataset is constructed based on images of a first metal wedge, a second metal wedge, a set of first ore images, and a set of second ore images. This includes determining pixel fitting curves based on the first and second metal wedge images; adding noise to each first ore image in the first ore image set and adding X-ray transmission interference based on the pixel fitting curves, then matching the response values ​​of each image with any one or more second ore images in the second ore image set to obtain a set of matched ore images; and constructing a training dataset based on the first ore image set and the set of matched ore images. A convolutional neural network model is trained based on a training dataset to obtain a mineral deposit restoration model. This model includes a model input module, a global feature extraction module, a local feature extraction module, an image restoration module, and a model output module, all connected in sequence. The model input module is used to input the image of the ore to be restored. The global feature extraction module extracts global features from the image of the ore to be restored. The local feature extraction module extracts local features from the image of the ore to be restored based on the global features. The image restoration module fuses the local features to obtain a high-resolution restored image. The model output module outputs the restored mineral deposit image based on the high-resolution restored image. Under the second data acquisition conditions, images of the ore to be recovered are acquired and input into the ore recovery model to obtain the ore recovery image.

2. The method according to claim 1, characterized in that, Acquire images of the first and second metal wedges under the first and second data acquisition conditions, including: A sample metal wedge is placed at a fixed position on the conveying device. The conveying device is controlled to run at a first conveying speed and under a first X-ray power condition to acquire an image of the first metal wedge. The conveying device is then controlled to run at a second conveying speed and under a second X-ray power condition to acquire an image of the second metal wedge.

3. The method according to claim 1, characterized in that, Acquire images of the first and second metal wedges under the first and second data acquisition conditions, including: At a certain position of the conveying device, a first sample metal wedge is placed. Under a first acquisition condition, several images of the first sample metal wedge are obtained; under a second acquisition condition, several images of the first sample metal wedge are obtained. At the same position of the conveying device, a second sample metal wedge is placed. Under the first acquisition condition, several first metal wedge images of the second sample metal wedge are obtained; under the second acquisition condition, several second metal wedge images of the second sample metal wedge are obtained. A first metal wedge image is obtained based on several first metal wedge images of the first sample metal wedge and several first metal wedge images of the second sample metal wedge; a second metal wedge image is obtained based on several second metal wedge images of the first sample metal wedge and several second metal wedge images of the second sample metal wedge. The density difference between the first sample metal wedge and the second sample metal wedge is greater than the set threshold.

4. The method according to claim 1, characterized in that, The global feature extraction module includes a head convolutional layer, a first depthwise separable convolutional layer, a channel attention layer, a normalization layer, and a first feedforward neural network layer connected in sequence. and feature splicing layer; The head convolutional layer is connected to the model input module; The feature concatenation layer is connected to the head convolutional layer and the first feedforward neural network layer, and is used to concatenate the outputs of the head convolutional layer and the first feedforward neural network layer to obtain global features.

5. The method according to claim 1, characterized in that, The local feature extraction module includes a second depthwise separable convolutional layer and several multi-resolution feature extraction units that are connected to the second depthwise separable convolutional layer and arranged in parallel. The second depthwise separable convolutional layer is connected to the global feature extraction module and is used to extract high-resolution features of the image to be restored based on global features. The multi-resolution extraction unit is used to extract features at various resolutions from the image to be restored based on high-resolution features.

6. The method according to any one of claims 1 to 5, characterized in that, The image restoration module includes a feature fusion layer, an attention mechanism layer, a bottom-up fusion layer, and a deconvolution or interpolation layer connected in sequence. The feature fusion layer and attention mechanism layer are used to fuse features at different resolutions and then obtain a multi-resolution feature set through the attention mechanism. Bottom-up fusion layers and deconvolution / interpolation layers are used to fuse multi-resolution feature sets from the bottom up, and then perform the reverse operation to obtain a high-resolution restored image.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1 to 6.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 6.

Images