High-purity quartz identification method, device, medium and equipment

Abstract

The invention discloses a high-purity quartz recognition method and device, a medium and equipment, and relates to the technical field of image recognition. The method comprises the following steps: acquiring an image data set of rock ore; taking the image data set as an input, taking the range information of the quartz with the purity greater than the preset purity as an output, and training a CNN network to obtain a classification model; the range information and images in the image data set are fused to obtain images fused with the range information, the images fused with the range information serve as input of a U-Net network, the segmented quartz images serve as output of the U-Net network, the U-Net network is trained, and a high-purity quartz segmentation model is obtained; cascading the classification model and the high-purity quartz segmentation model to obtain a high-purity quartz recognition model; and inputting an image of the rock ore to be identified, which is acquired in real time, into the high-purity quartz identification model to obtain an identification result. According to the scheme, the high-purity quartz in the rock ore can be accurately identified.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition, and particularly relates to a high-purity quartz recognition method, device, medium, and equipment. Background Art

[0002] As an important industrial mineral material, high-purity quartz is widely used in high-tech fields such as high-tech electronics and solar energy, and plays an important role in the development of industry and economy. Therefore, it has important research value and application potential to accurately and efficiently evaluate the quality of quartz in rock ores. High-purity quartz refers to rock ores in which the silica content in quartz is higher than 99.99% and the characteristics of fluid inclusions meet specific requirements.

[0003] Traditional methods for high-purity quartz recognition and evaluation include microscopic observation method, temperature and pressure estimation method, etc. The microscopic observation method uses a polarized light microscope to measure the properties such as quartz content, particle size, transparency, and the content, size, and distribution characteristics of its fluid inclusions, and uses related equipment to measure information such as the initial melting temperature and internal substance composition; the temperature and pressure estimation method is to measure the volume and pressure of the liquid in the inclusion and calculate the temperature and pressure at the time of mineral crystallization in combination with its physical properties. Traditional methods for high-purity quartz recognition and evaluation are time-consuming, and the analysis results are highly subjective, with high requirements for the personal experience of researchers.

[0004] In recent years, deep learning technology has made great progress in the field of image recognition, but its application in high-purity quartz recognition and evaluation is still rare. At present, the application of this technology still has some limitations and defects, which to a certain extent limit the accuracy of high-purity quartz recognition and evaluation. For high-purity quartz recognition, the existing recognition methods based on deep learning technology rely more on a single traditional neural network, and it is difficult to comprehensively capture the long-range dependence relationships and features in the image in the face of its complex morphology, texture, and boundaries; for the quantitative estimation of fluid inclusions, it mainly relies on manual analysis, traditional microscopes, or basic image processing, statistical analysis, etc. methods, which are lower than expected in terms of accuracy and efficiency. At the same time, the existing technology lacks the capture of global and local information of high-purity quartz images, which affects the accuracy of recognition and evaluation.

[0005] In summary, the existing technology has insufficient accuracy and low efficiency in high-purity quartz recognition. Summary of the Invention

[0006] Based on this, it is necessary to provide a high-purity quartz recognition method, device, medium, and equipment for the technical problems of insufficient accuracy and low efficiency in high-purity quartz recognition in the existing technology.

[0007] The present invention adopts the following technical solutions: In a first aspect, the present invention provides a method for identifying high-purity quartz, the method comprising: Obtaining an image dataset of rock ores; Using the image dataset as an input and the range information of quartz with a purity greater than a preset purity as an output, training a CNN network to obtain a classification model; fusing the range information with the images in the image dataset to obtain an image fused with the range information, using the image fused with the range information as an input of a U-Net network, and using the segmented quartz image as an output of the U-Net network, training the U-Net network to obtain a high-purity quartz segmentation model; cascading the classification model and the high-purity quartz segmentation model to obtain a high-purity quartz identification model; Inputting the image of the rock ore to be identified collected in real time into the high-purity quartz identification model to obtain the identification result of quartz in the rock ore to be identified.

[0008] Further, after obtaining the identification result of quartz in the rock ore to be identified, it further comprises: Extracting the data features in the image dataset, fusing the data features in the image dataset with the features output by the encoder layer of the high-purity quartz identification model to obtain the fused features, the data features in the image dataset including quartz content, particle size and transparency, and the features output by the encoder layer including the texture and boundary of quartz; Using the fused features as an input and the index of fluid inclusions as an output, training a Transformer network to obtain a fluid inclusion index evaluation model, the index including the content, size and distribution mode of fluid inclusions; Using the fluid inclusion index evaluation model to determine the fluid inclusion index of quartz in the rock ore to be identified, evaluating the identification result of quartz in the rock ore to be identified according to the fluid inclusion index, and optimizing the parameters of the high-purity quartz identification model according to the evaluation result.

[0009] Further, fusing the data features in the image dataset with the features output by the encoder layer of the high-purity quartz identification model to obtain the fused features specifically comprises: Respectively assigning corresponding weight coefficients to each type of feature in the data features in the image dataset and the features output by the encoder layer; Weighted summing each type of feature based on the weight coefficients corresponding to each type of feature to obtain the fused features.

[0010] Further, after obtaining the identification result of quartz in the rock ore to be identified, it further comprises: Generate an evaluation report corresponding to the recognition result according to a preset report template, and send the evaluation report to the user, so that the user can view the picture of the recognition result and the process data of the high-purity quartz recognition model processing the rock ore to be recognized through the evaluation report.

[0011] In a second aspect, the present invention provides a high-purity quartz recognition device, including: An acquisition module, configured to acquire an image data set of rock ore; A construction module, configured to use the image data set as an input, use the range information of quartz with a purity greater than a preset purity as an output, train a CNN network to obtain a classification model; fuse the range information with the images in the image data set to obtain an image fused with the range information, use the image fused with the range information as an input of a U-Net network, use the segmented quartz image as an output of the U-Net network, train the U-Net network to obtain a high-purity quartz segmentation model; cascade the classification model and the high-purity quartz segmentation model to obtain a high-purity quartz recognition model; A recognition module, configured to input an image of a rock ore to be recognized collected in real time into the high-purity quartz recognition model to obtain a recognition result of quartz in the rock ore to be recognized.

[0012] The present invention provides a computer-readable storage medium, and the storage medium stores a computer program, and when the computer program is executed by a processor, the high-purity quartz recognition method is implemented.

[0013] The present invention provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, and when the processor executes the program, the high-purity quartz recognition method is implemented.

[0014] At least one technical solution adopted by the present invention can achieve the following beneficial effects: The present invention first uses an image data set as input and the range information of quartz with a purity greater than a preset purity as output to train a CNN network to obtain a classification model; then fuses the range information with the images in the image data set to obtain an image fused with the range information, and uses the image fused with the range information as the output of the U-Net network to train the U-Net network to obtain a high-purity quartz segmentation model; finally, cascades the classification model and the high-purity quartz segmentation model to obtain a high-purity quartz recognition model. It can first extract the high-level features of quartz in the ore to be identified through multiple convolutional layers of the CNN network, and classify the quartz in the ore to be identified as a whole through the extracted high-level features (judging which regions in the ore to be identified correspond to high-purity quartz and which regions correspond to low-purity quartz), and then use the U-Net network to extract the features of the high-purity quartz region and the low-purity quartz region in the ore image to be identified at different scales according to the ore image fused with the range information, capture the detailed information in the high-purity quartz region and the low-purity quartz region, and accurately locate the boundary between the high-purity quartz region and the low-purity quartz region, realizing the range information of quartz with a purity greater than the preset purity as the "prior knowledge" of the U-Net network, enabling it to preferentially process the high-probability regions where quartz with a purity greater than the preset purity is located in the image, reducing ineffective calculations (such as background regions), thereby being able to improve the inference speed of the model and accurately segment the regions corresponding to high-purity quartz in the quartz image of the ore to be identified, so as to realize the efficient and accurate identification of high-purity quartz. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention. In the drawings: Figure 1 is a flowchart of a method for identifying high-purity quartz provided by the present invention; Figure 2 is another flowchart of a method for identifying high-purity quartz provided by the present invention; Figure 3 is a flowchart for quantitative estimation of fluid inclusions provided by the present invention; Figure 4 is a schematic diagram of a device for identifying high-purity quartz provided by the present invention; Figure 5 is a schematic diagram of a computer device for implementing a method for identifying high-purity quartz provided by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0016] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with specific embodiments of the present invention and the corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0017] Currently, the operating platform mentioned in the present invention can be devices such as desktop computers and laptop computers that can execute the solutions of the present invention. For the convenience of description, only the operating platform will be described as the execution subject below. The following will detail the technical solutions provided by each embodiment of the present invention with reference to the drawings.

[0018] Reference Figure 1 , a high-purity quartz identification method in the present invention specifically includes the following steps: S10: Obtain an image dataset of rock ores.

[0019] In this embodiment, referring to Figure 2 , the methods for obtaining the image dataset of rock ores include but are not limited to: obtaining by taking multi-angle photographs of slices of rock ores, or obtaining by collecting existing image data of rock ores. Among them, the existing image data of rock ores includes but is not limited to historical annotation data and open-source network data. Historical annotation data refers to previously annotated image data of rock ores (labels are high-purity quartz or non-high-purity quartz).

[0020] Preferably, after obtaining the image dataset of rock ores, preprocess the image dataset of rock ores (i.e., feature engineering) to obtain a preprocessed image dataset. The preprocessing methods include but are not limited to: cleaning, scaling, normalizing, standardizing, flipping, cropping, denoising, and histogram equalization of the images in the image dataset.

[0021] S20: Use the image dataset as the input, and use the range information of quartz with a purity greater than the preset purity as the output. Train a convolutional neural network (CNN) to obtain a classification model; fuse the range information with the images in the image dataset to obtain an image fused with the range information. Use the image fused with the range information as the input of the U-Net network, and use the segmented quartz image as the output of the U-Net network. Train the U-Net network to obtain a high-purity quartz segmentation model; cascade the classification model and the high-purity quartz segmentation model to obtain a high-purity quartz identification model.

[0022] In this embodiment, the preset purity refers to the content of silicon dioxide in quartz set in advance, and the preset purity can be set to a value greater than or equal to 99.99% and less than 1. Quartz with a purity greater than the preset purity refers to high-purity quartz, that is, quartz with a silicon dioxide content higher than 99.99%; quartz not greater than the preset purity refers to low-purity quartz, that is, quartz with a silicon dioxide content lower than 99.99%.

[0023] Specifically, the CNN classification model learns the features of the rock and ore images and outputs the probability that each region in the image belongs to high-purity quartz (i.e., quartz with a purity greater than the preset purity), and then screens out the high-probability regions as the range where "high-purity quartz may exist".

[0024] Optionally, the range information can be represented by a probability mask or a binary mask: Probability mask: Generate a matrix with the same size as the input image, and the value of each pixel represents the probability that the position belongs to high-purity quartz (e.g., 0.8 represents an 80% probability).

[0025] Binary mask: Set a probability threshold (e.g., 0.9), mark the high-probability regions as 1 (belonging to high-purity quartz), and mark the low-probability regions as 0 to form a binary region mask.

[0026] It should be noted that the range information does not change the pixel values of the original image, but exists in the form of additional matrix or coordinate information, which is used to indicate the regions that the U-Net segmentation model needs to focus on processing.

[0027] In this embodiment, the specific ways to fuse the range information with the images in the image dataset include but are not limited to: (1) Channel stacking fusion: Use the range information (such as the probability mask) as an additional channel to be concatenated with the RGB channels of the original image to form a new input tensor. For example: the original image has 3 channels (RGB), and the range information has 1 channel (probability value). After fusion, the tensor input to the U-Net has 4 channels (RGB + probability). The encoder of the U-Net can simultaneously extract the visual features of the original image (such as texture, color) and the probability features of the range information (such as the coordinates of high-probability regions) to achieve feature-level fusion.

[0028] (2) Spatial position fusion: Convert the range information into region coordinates or masks, corresponding to the spatial positions of the original image, and mask the non-target regions through the mask. For example: Generate a binary mask (1 represents the possible region of high-purity quartz, 0 represents others), multiply it with the original image pixel by pixel, so that the U-Net only processes the regions where the mask is 1 and ignores the background noise.

[0029] In this embodiment, the range information of quartz with a purity higher than the preset purity output by the classification model is fused with the rock and ore image, and then the image fused with the range information is used as the input of the U-Net network, and the segmented quartz image is used as the output of the U-Net network. By training the U-Net network, it is possible to use the U-Net network to extract the features of the images of high-purity quartz and low-purity quartz at different scales respectively, better capture the detailed information in the images of high-purity quartz and low-purity quartz, so that the trained high-purity quartz segmentation model can more accurately locate the boundary between high-purity quartz and low-purity quartz, thereby improving the segmentation accuracy of high-purity quartz.

[0030] Moreover, the U-Net network can fully exploit the context information of the images of quartz with different purities through a small amount of input data. By establishing skip connections between the encoder and the decoder and combining low-level detailed information with high-level semantic information, it can achieve accurate segmentation of high-purity quartz images with limited data, improving the generalization ability of the high-purity quartz segmentation model to segment different quartz images.

[0031] S30: Input the image of the rock and ore to be identified collected in real time into the high-purity quartz identification model to obtain the identification result of the quartz in the rock and ore to be identified.

[0032] In this embodiment, the rock and ore to be identified is sliced, and high-definition photos are taken of the rock and ore to be identified from multiple angles. The obtained image of the rock and ore to be identified is input into the high-purity quartz identification model to obtain the identification result of the quartz in the rock and ore to be identified, and thus it can be identified whether the quartz in the rock and ore to be identified is high-purity quartz.

[0033] In this embodiment, the classification model and the high-purity quartz segmentation model are fused in a cascaded manner of "first classifying and locating the range, and then segmenting and finely annotating". The range information (probability mask) of high-purity quartz output by the classification model is fused with the original image and then input into the U-Net to achieve "global guidance + local refinement" collaborative recognition. This design not only uses the CNN to quickly screen the target area, but also captures the fine boundary through the U-Net, solving the problem that a single model is difficult to handle long-range dependencies and pixel-level details simultaneously, and finally improving the efficiency and accuracy of high-purity quartz recognition.

[0034] Based on Figure 1A high-purity quartz identification method is shown as follows. First, an image dataset is used as input, and the range information of quartz with a purity greater than a preset purity is used as output. The CNN network is trained to obtain a classification model. Then, the range information is fused with the images in the image dataset to obtain an image fused with the range information. The image fused with the range information is used as the input of the U-Net network, and the segmented quartz image is used as the output of the U-Net network. The U-Net network is trained to obtain a high-purity quartz segmentation model. Finally, the classification model and the high-purity quartz segmentation model are cascaded to obtain a high-purity quartz identification model. It can first extract the high-level features (i.e., the range information of quartz with a purity greater than the preset purity) of the quartz image in the rock or ore to be identified through multiple convolutional layers of the CNN network, and classify the quartz in the rock or ore to be identified as a whole through the extracted high-level features (judging which regions in the rock or ore to be identified correspond to high-purity quartz and which regions correspond to low-purity quartz). Then, the U-Net network is used to extract the features of the high-purity quartz region and the low-purity quartz region in the rock or ore image to be identified at different scales according to the rock or ore image fused with the range information, capture the detailed information in the high-purity quartz region and the low-purity quartz region, accurately locate the boundary between the high-purity quartz region and the low-purity quartz region, and use the high-level features as the "prior knowledge" of the U-Net network to give priority to processing the high-probability regions where quartz with a purity greater than the preset purity is located in the image, reducing invalid calculations (such as background regions), thereby improving the inference speed of the model and accurately segmenting the regions corresponding to the high-purity quartz in the rock or ore image to be identified, so as to realize the efficient and accurate identification of high-purity quartz.

[0035] When applying a high-purity quartz identification method provided by the present invention, it is not necessary to execute according to Figure 1 the order of the steps shown. The specific execution order of each step can be determined according to needs, and the present invention does not limit this.

[0036] In addition, in one or more embodiments of the present invention, after obtaining the identification result of the quartz in the rock or ore to be identified, it further includes: Extracting the data features in the image dataset, fusing the data features in the image dataset with the features output by the encoder layer of the high-purity quartz identification model to obtain the fused features. The data features in the image dataset include quartz content, particle size, and transparency, and the features output by the encoder layer include the texture and boundary of quartz.

[0037] In this embodiment, referring to Figure 3 , the fluid inclusion quantitative estimation process includes: Data collection, that is, obtaining the image dataset of the rock or ore, and the data sources include but are not limited to historical annotation data and open-source network data.

[0038] Feature engineering, that is, preprocessing the collected data so that the data is suitable for the training of the Transformer network. The specific methods of data preprocessing have been listed above and will not be elaborated here.

[0039] Model training: Train the Transformer network, and use the high-purity quartz recognition model for feature-level migration during the training process to achieve model fusion, obtain the fluid inclusion index evaluation model, and then use the obtained model for quantitative estimation of fluid inclusions.

[0040] Specifically, select and extract the feature information in the quartz image dataset according to expert knowledge and data analysis to obtain data features that can reflect the characteristics of quartz. Fuse the features output by the encoder of the high-purity quartz recognition model (such as the texture of quartz, the boundary of quartz, etc.) with the data features in the image dataset (quartz content, particle size, and transparency) to obtain the features of quartz at different scales, realize the migration of the features of quartz images at different scales, and thus improve the model's learning ability for different types of features.

[0041] Use the fused features as input and the index of fluid inclusions as output to train the Transformer network to obtain the fluid inclusion index evaluation model. The index includes the content, size, and distribution mode of fluid inclusions.

[0042] In this embodiment, the Transformer network constructs a dependency relationship through the self-attention mechanism. Its basic structure includes the self-attention mechanism, multi-head attention, position encoding, encoder, and decoder. The core idea of the self-attention mechanism is to calculate the relationship between each position in the sequence and construct a vector representation, so as to connect different regions to achieve global modeling; the multi-head attention mechanism is to parallelize multiple attention mechanisms to learn image features from multiple perspectives; the position encoding is used to mark the sequence position information; the encoder is used to process the input data features and extract global features; the decoder generates the target according to the features output by the encoder for pixel-level segmentation.

[0043] Use the fluid inclusion index evaluation model to determine the fluid inclusion index of quartz in the rock or ore to be identified, evaluate the recognition result of quartz in the rock or ore to be identified according to the fluid inclusion index, and optimize the parameters of the high-purity quartz recognition model according to the evaluation result.

[0044] In this embodiment, the contents of impurities such as metal elements and organic substances contained in high-purity quartz, such as aluminum, iron, calcium, and magnesium, directly affect its application performance. These impurities usually exist in the form of fluid inclusions, which are the most abundant and main types of impurities in high-purity quartz. Their types and contents directly affect the purity of quartz. The presence of these impurity elements will reduce the purity of quartz and affect the product quality of high-purity quartz. For example, when there are too many inclusions, especially fine-grained gas-liquid two-phase inclusions and mineral inclusions, in high-purity quartz, the impurity elements in the inclusions will cause the purity of high-purity quartz to decrease. The formation of fluid inclusions usually occurs during the formation of minerals, where gases, liquids, etc. are trapped in mineral crystals. Studying fluid inclusions helps to understand the temperature, pressure, and chemical composition information during their formation, so as to analyze the genesis of high-purity quartz deposits and assist in the discovery of new deposits.

[0045] Specifically, if the recognition result of quartz in the rock or ore to be recognized is high-purity quartz, but the fluid inclusion index of quartz in the rock or ore to be recognized is greater than the maximum value of the fluid inclusion index in high-purity quartz, it indicates that the recognition accuracy of the high-purity quartz recognition model is not high, and the parameters of the high-purity quartz recognition model need to be corrected until the recognition result of the high-purity quartz recognition model matches the result output by the fluid inclusion evaluation model.

[0046] In the solution shown in this embodiment, first, the data features in the image dataset of the rock or ore are fused with the features output by the encoder layer of the high-purity quartz recognition model to obtain the fused features, which can obtain the features of quartz at different scales, realize the migration of the features of the quartz image at different scales, and thus improve the model's learning ability for different types of features. Then, taking the fused features as the input and the fluid inclusion index as the output, the Transformer network is trained to obtain the fluid inclusion index evaluation model, which can migrate the features extracted from the high-purity quartz recognition model into the Transformer network to help the model better understand the relationship between high-purity quartz and fluid inclusions, and thus improve the quantitative estimation result of fluid inclusions. Finally, using the fluid inclusion index evaluation model to determine the fluid inclusion index of quartz in the rock or ore to be recognized, evaluating the recognition result of quartz in the rock or ore to be recognized according to the fluid inclusion index of quartz in the rock or ore to be recognized, and optimizing the parameters of the high-purity quartz recognition model according to the evaluation result can verify the recognition accuracy of the high-purity quartz recognition model and further improve the recognition accuracy of high-purity quartz.

[0047] In addition, in one or more embodiments of the present invention, fusing the data features in the image dataset with the features output by the encoder layer of the high-purity quartz recognition model to obtain the fused features specifically includes: Correspondingly weighted coefficients are assigned to each type of feature in the data features of the image dataset and the features output by the encoder layer respectively.

[0048] Based on the weighted coefficients corresponding to each type of feature respectively, each type of feature is weighted and summed to obtain the fused feature.

[0049] In the solution of this embodiment, by combining the influence degrees of different types of features on the evaluation of the fluid inclusion index, different weighted coefficients are set for each separate type of feature, and the fluid inclusion index in quartz can be evaluated more precisely.

[0050] In addition, in one or more embodiments of the present invention, after obtaining the recognition result of quartz in the rock or ore to be recognized, it further includes: Generating an evaluation report corresponding to the recognition result according to a preset report template, and sending the evaluation report to the user, so that the user can view the picture of the recognition result and the process data of the high-purity quartz recognition model processing the rock or ore to be recognized through the evaluation report.

[0051] In this embodiment, after generating the recognition result of quartz in the rock or ore to be recognized, an evaluation report is generated according to the pre-embedded report template, so that the user can view the picture of the recognition result and the process data of the high-purity quartz recognition model processing the rock or ore to be recognized through the evaluation report, which can improve the work efficiency of the user.

[0052] The above is a method for recognizing high-purity quartz provided by one or more embodiments of the present invention. Based on the same idea, the present invention also provides a corresponding device for recognizing high-purity quartz, as Figure 4 shown, including: An acquisition module, configured to acquire an image dataset of rock or ore.

[0053] A construction module, configured to use the image dataset as an input, use the range information of quartz with a purity higher than a preset purity as an output, train a CNN network to obtain a classification model; fuse the range information with the images in the image dataset to obtain an image fused with the range information, use the image fused with the range information as an input of a U-Net network, use the segmented quartz image as an output of the U-Net network, train the U-Net network to obtain a high-purity quartz segmentation model; cascade the classification model and the high-purity quartz segmentation model to obtain a high-purity quartz recognition model.

[0054] A recognition module, configured to input the image of the rock or ore to be recognized collected in real time into the high-purity quartz recognition model to obtain the recognition result of quartz in the rock or ore to be recognized.

[0055] For the specific limitations of a high-purity quartz identification device, reference may be made to the limitations of a high-purity quartz identification method in the foregoing text, which will not be elaborated herein. Each module in the high-purity quartz identification device may be implemented in whole or in part by software, hardware, or a combination thereof. Each module may be embedded in or independent of a processor in a computer device in the form of hardware, or may be stored in a memory in the computer device in the form of software, so as to facilitate the processor to call and execute the operations corresponding to each of the above modules.

[0056] The present invention also provides a computer-readable storage medium storing a computer program, which can be used to execute the Figure 1 provided high-purity quartz identification method.

[0057] The present invention also provides Figure 5 a schematic structural diagram of the computer device shown, as Figure 5 shown. At the hardware level, the computer device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory. Of course, other hardware required for other services may also be included. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to implement the Figure 1 provided high-purity quartz identification method.

[0058] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it may include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium used in the various embodiments provided by the present invention may include at least one of non-volatile and volatile memories. The non-volatile memory may include a read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical memory, etc. The volatile memory may include a random access memory (RAM) or an external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0059] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered to be within the scope recorded by the present invention.

Claims

1. A method for identifying high-purity quartz, characterized in that, Including: Obtain an image dataset of rock and ore; Taking the image dataset as input and the range information of quartz with a purity greater than a preset purity as output, training a CNN network to obtain a classification model; fusing the range information with the images in the image dataset to obtain an image fused with the range information, taking the image fused with the range information as the input of the U-Net network, and taking the segmented quartz image as the output of the U-Net network, training the U-Net network to obtain a high-purity quartz segmentation model; cascading the classification model and the high-purity quartz segmentation model to obtain a high-purity quartz recognition model; Input the image of the rock and ore to be recognized collected in real time into the high-purity quartz recognition model to obtain the recognition result of quartz in the rock and ore to be recognized.

2. The high-purity quartz identification method according to claim 1, wherein After obtaining the recognition result of quartz in the rock and ore to be recognized, it further includes: Extract the data features in the image dataset, fuse the data features in the image dataset with the features output by the encoder layer of the high-purity quartz recognition model to obtain the fused features. The data features in the image dataset include quartz content, particle size and transparency, and the features output by the encoder layer include the texture and boundary of quartz; Taking the fused features as input and the index of fluid inclusions as output, training a Transformer network to obtain a fluid inclusion index evaluation model. The index includes the content, size and distribution mode of fluid inclusions; Use the fluid inclusion index evaluation model to determine the fluid inclusion index of quartz in the rock and ore to be recognized, evaluate the recognition result of quartz in the rock and ore to be recognized according to the fluid inclusion index, and optimize the parameters of the high-purity quartz recognition model according to the evaluation result.

3. The method for identifying high-purity quartz according to claim 2, characterized in that, Fusing the data features in the image dataset with the features output by the encoder layer of the high-purity quartz recognition model to obtain the fused features, specifically including: Respectively assign corresponding weight coefficients to each type of feature in the data features in the image dataset and the features output by the encoder layer; Based on the weight coefficients corresponding to each type of feature, perform weighted summation on each type of feature to obtain the fused features.

4. The method for identifying high-purity quartz according to claim 1, wherein After obtaining the recognition result of quartz in the rock and ore to be recognized, it further includes: Generate an evaluation report corresponding to the recognition result according to a preset report template, and send the evaluation report to the user, so that the user can view the picture of the recognition result and the process data of the high-purity quartz recognition model processing the rock and ore to be recognized through the evaluation report.

5. A high-purity quartz identification device, characterized in that, Including: An acquisition module for acquiring an image dataset of rock and ore; A building module, which takes the image dataset as input, takes the range information of quartz with a purity greater than a preset purity as output, trains a CNN network to obtain a classification model; fuses the range information with the images in the image dataset to obtain an image fused with the range information, takes the image fused with the range information as the input of a U-Net network, takes the segmented quartz image as the output of the U-Net network, trains the U-Net network to obtain a high-purity quartz segmentation model; cascades the classification model and the high-purity quartz segmentation model to obtain a high-purity quartz recognition model; A recognition module, which inputs the image of the rock and ore to be recognized collected in real time into the high-purity quartz recognition model to obtain the recognition result of quartz in the rock and ore to be recognized.

6. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, and when the computer program is executed by a processor, it implements a high-purity quartz recognition method according to any one of claims 1 to 4.

7. A computer device, characterized in that, It includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements a high-purity quartz recognition method according to any one of claims 1 to 4.

Images