Multi-feature fusion ore selection identification method and ore sorting machine

Through the ore separation recognition method of multi-feature fusion, combined with surface detection images and transmission projection images, the features are extracted and fused, and the problems of low recognition accuracy and low efficiency in the prior art are solved, achieving efficient ore recognition.

CN118570555BActive Publication Date: 2024-11-19HUNAN JUMPER TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411002330.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-25
Publication Date
2024-11-19
Estimated Expiration
2044-07-25

AI Technical Summary

Technical Problem

When collecting multiple image data, existing ore sorters will have low recognition accuracy if only one image is considered. If multiple images are considered, the data volume will be large and the recognition efficiency will be low. It is difficult to comprehensively utilize the advantages of various sensors to improve recognition accuracy. and efficiency.

Method used

Using the ore separation recognition method of multi-feature fusion, the ore separation recognition model is constructed and trained, and the surface detection image and transmission projection image are combined, the surface features and transmission characteristics are extracted, and the two are fused through the full connection layer to generate the total features for classification. Improve the accuracy and efficiency of ore identification.

Benefits of technology

实现了充分利用不同类型图像信息,突出每类特征的共性和个性,提高了矿石识别的准确性和效率,降低了识别难度。

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118570555B_ABST
    Figure CN118570555B_ABST
Patent Text Reader

Abstract

The present invention relates to a multi-feature fused ore selection identification method and an ore sorting machine thereof, wherein the input layer includes two input ends, which are respectively used to receive the surface detection image and the transmission projection image of the ore; then, the surface features and the transmission features are respectively extracted; the surface features and the transmission features are then fused to learn the complex relationship between the surface features and the transmission features, and the higher-dimensional features are extracted as the total features; finally, the ore type is identified according to the total features, and the ore selection identification result is output. The method is a multi-feature fused ore selection identification method, which is based on the principle of homogeneity priority, and can make full use of different types of image information to achieve multi-directional fusion; it can also fuse homogeneous features, highlight the commonality and individuality of each type of features, and comprehensively improve the accuracy and efficiency of ore identification.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of image recognition, and in particular to a multi-feature fusion ore separation recognition method and an ore separator thereof. Background Art

[0002] Ore sorting machines in the prior art may collect multiple image data, such as color images, near-infrared images, XRT images, dual-energy XRT images, ultraviolet fluorescence images, black-and-white grayscale images, etc. If only one of them is considered, the data may be single and the recognition accuracy may be low. For example, color images only reflect the visible light band of the sample surface; near-infrared images only reflect the near-infrared band of the sample surface; and XRT images only reflect the plane projection of the sample after being transmitted by X-rays. If multiple types are considered, the data volume may be large and the recognition efficiency may be low.

[0003] Therefore, how to find the commonalities and characteristics of various types of images, such as some images are surface images, and some images are transmission projection images, can not only comprehensively consider multiple types of sensors, give full play to their respective advantages, and improve recognition accuracy; it can also reduce the difficulty of recognition and improve recognition efficiency to a certain extent. Summary of the invention

[0004] In order to solve at least one of the above technical problems, the present invention provides a multi-feature fusion ore selection identification method, comprising:

[0005] Build and train a mine identification model, including:

[0006] An input layer, comprising: a surface detection image input terminal and a transmission projection image input terminal, for inputting at least one surface detection image and at least one transmission projection image respectively;

[0007] The feature extraction layer includes: a surface feature extraction unit and a transmission feature extraction unit; the surface feature extraction unit is connected to the surface detection image input end and is used to fuse each surface detection image to extract the surface feature; the transmission feature extraction unit is connected to the transmission projection image input end and is used to fuse each transmission projection image to extract the transmission feature;

[0008] The fully connected layer is connected to the surface feature extraction unit and the transmission feature extraction unit, and is used to fuse the surface features and the transmission features, extract the total features for classification, and output the mineral selection classification results;

[0009] At least one surface detection image and at least one transmission projection image of the ore to be identified are acquired and input into the ore selection identification model to obtain the ore selection identification result.

[0010] Furthermore, the input layer further comprises: a surface statistical image input terminal and a transmission statistical image input terminal, respectively used to input at least one surface statistical image and at least one transmission statistical image; a feature extraction layer, comprising a surface statistical feature extraction unit and a transmission statistical feature extraction unit;

[0011] The input of the surface statistical feature extraction unit is connected to the input end of the surface statistical image, and the output is connected to the fully connected layer, which is used to fuse the surface statistical images, extract the surface statistical features, and access the fully connected layer;

[0012] The input of the transmission statistical feature extraction unit is connected to the transmission statistical image input terminal, and the output is connected to the fully connected layer, which is used to fuse the transmission statistical images, extract the transmission statistical features, and access the fully connected layer.

[0013] Furthermore, a plurality of surface detection images or a plurality of transmission projection images are combined to calculate statistical images of the respective combined images as inputs to the surface statistical image input terminal and the transmission statistical image input terminal.

[0014] Further, constructing and training a mine selection identification model includes: constructing and training a mine selection identification model group, including several of the mine selection identification models;

[0015] Before identification, one or more ore selection identification models are selected from the ore selection identification model group as the current identification model; the acquired surface detection image and transmission projection image are input into the current identification model to obtain the ore selection identification result.

[0016] Furthermore, each ore selection identification model in the ore selection identification model group corresponds to a different surface detection image and transmission projection image; before identification, according to the currently acquired surface detection image and transmission projection image, a ore selection identification model is selected in the ore selection identification model group as the current identification model to obtain the ore selection identification result;

[0017] Or, each mineral selection identification model in the mineral selection identification model group corresponds to the same surface detection image and transmission projection image; before identification, multiple mineral selection identification models are selected as current identification models in the mineral selection identification model group according to the currently acquired surface detection image and transmission projection image, and multiple mineral selection identification results are obtained; based on the multiple mineral selection identification results, a final mineral selection identification result is obtained.

[0018] Furthermore, the steps of constructing and training the ore selection identification model group include:

[0019] Construct n surface detection images P respectively i and m transmission projection images Q jis the input, and the mining identification result is the N×M mining identification models under construction as the output; wherein n, m, N, M, i, j are all integers, and 0<n≤N, 0<m≤M, 0<i≤n; 0<j≤m; n represents the number of surface detection images in the mining identification model under construction; m represents the number of transmission projection images in the mining identification model under construction; N represents the total number of surface detection images; M represents the total number of transmission projection images; i represents the i-th surface detection image; j represents the j-th transmission projection image;

[0020] For each ore selection identification model under construction, select Surface detection image sample library and The transmission projection image sample library is used for training to obtain the trained intermediate ore selection recognition model;

[0021] According to the parameters of the trained intermediate ore selection identification model, the parameters of the ore selection identification model under construction are determined to obtain N×M trained ore selection identification models to form an ore selection identification model group.

[0022] Furthermore, the input layer includes two convolutional layers, one for inputting the surface detection image and the other for inputting the transmission projection image; each convolutional layer converts the number of channels of the image to 64 and uses a 3x3 convolution kernel for feature extraction.

[0023] Furthermore, in the feature extraction layer, each feature extraction unit includes several convolutional layers, several batch normalization layers, activation functions and pooling layers connected in sequence.

[0024] Further, a fully connected layer, including input neurons, output neurons and a weight matrix, is followed by a classification function; the number of output neurons is equal to the number of ore categories.

[0025] On the other hand, the present invention also provides an ore sorting machine, which adopts any of the above-mentioned ore sorting identification methods.

[0026] The present invention provides a multi-feature fusion ore selection and identification method and an ore sorting machine thereof, which is a multi-feature fusion ore selection and identification method, especially dividing it into surface features and transmission features. Based on the principle of homogeneity priority, the surface features of all surface detection images are first fused, the transmission features of all transmission projection images are fused, and then the two are fused to obtain the total feature. It can fully utilize different types of image information to achieve multi-directional fusion; it can also fuse homogeneous features, highlight the commonality and individuality of each type of feature, and comprehensively improve the accuracy and efficiency of ore identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 The present invention is a flowchart of an embodiment of the method for identifying mineral processing according to the present invention. DETAILED DESCRIPTION

[0028] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0029] It should be noted that if the embodiments of the present invention involve directional indications, such as up, down, left, right, front, back, etc., then the directional indication is only used to explain the relative position relationship, movement status, etc. between the components in a certain specific posture. If the specific posture changes, the directional indication will also change accordingly. In addition, if the embodiments of the present invention involve descriptions of "first, second", "S1, S2", "step one, step two", etc., then such descriptions are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of indicated technical features or indicating the execution order of the method, etc. Those skilled in the art can understand that anything that does not violate the gist of the invention under the technical concept of the invention should be included in the protection scope of the present invention.

[0030] like Figure 1 As shown, the present invention provides a multi-feature fusion ore selection identification method, comprising:

[0031] S1: Build and train a mineral processing identification model, including:

[0032] The input layer A comprises: a surface detection image input terminal A1 and a transmission projection image input terminal A2, which are used to input at least one surface detection image and at least one transmission projection image respectively;

[0033] The feature extraction layer B includes a surface feature extraction unit B1 and a transmission feature extraction unit B2; the surface feature extraction unit B1 is connected to the surface detection image input terminal A1 and is used to fuse each surface detection image to extract the surface feature; the transmission feature extraction unit B2 is connected to the transmission projection image input terminal A2 and is used to fuse each transmission projection image to extract the transmission feature;

[0034] The fully connected layer C is connected to the surface feature extraction unit B1 and the transmission feature extraction unit B2, and is used to fuse the surface features and the transmission features, extract the total features for classification, and output the ore selection classification results;

[0035] S2: Acquire at least one surface detection image and at least one transmission projection image of the ore to be identified, and input them into the ore selection identification model to obtain the ore selection identification result.

[0036] In this embodiment, a multi-feature fusion ore selection identification method of the present invention is provided, and its input layer includes two input terminals, a surface detection image input terminal and a transmission projection image input terminal, which are used to receive the surface detection image and the transmission projection image of the ore respectively. Next, the surface feature extraction layer is responsible for extracting the surface features from the input surface detection image, which may include but is not limited to surface attributes such as texture, color, and shape; the transmission feature extraction layer is responsible for extracting the transmission features from the input transmission projection image, which may include but is not limited to transmission characteristics such as density and internal structure. Then, the fully connected layer fuses the surface features and the transmission features, and may include multiple neurons to learn the complex relationship between the surface features and the transmission features, so as to extract higher-dimensional features as the total features. Finally, the ore type is identified according to the total features, and the ore selection identification results are output.

[0037] This method is a multi-feature fusion ore selection and identification method, especially dividing it into surface features and transmission features. Based on the principle of homogeneity priority, the surface features of all surface detection images are fused, the transmission features of all transmission projection images are fused, and then the two are fused to obtain the total features. It can fully utilize different types of image information to achieve multi-directional fusion; it can also integrate homogeneous features, highlight the commonality and individuality of each type of feature, and comprehensively improve the accuracy and efficiency of ore identification.

[0038] Preferably, the input layer A is the starting point of the entire ore selection identification model and is responsible for receiving and processing input data. Preferably, the input layer may optionally include but is not limited to two convolution layers, one for inputting the surface detection image and one for inputting the transmission projection image; each convolution layer converts the number of channels of the image to 64 and uses a 3x3 convolution kernel for feature extraction. The specific number of channels and the size of the convolution kernel can be arbitrarily set according to the accuracy and size of the image, and are not limited thereto.

[0039] Specifically, the surface detection image may include, but is not limited to, color images, near-infrared images, ultraviolet fluorescence images, black-and-white grayscale images, etc., and may be RGB color images or grayscale images; the transmission projection image may include, but is not limited to, XRT images, dual-energy XRT images, etc. For surface feature input, a convolution layer may be designed as the starting point of the surface feature input. For transmission feature input, a convolution layer may also be designed as the starting point of the transmission feature input.

[0040] Preferably, the input layer may further include:

[0041] The image adjustment unit is used to adjust the surface detection image and the transmission projection image to a uniform size; specifically, since the sizes of the surface detection image and the transmission projection image collected by various acquisition devices, such as sensors, are different, it is necessary to adjust the images first.

[0042] The normalization unit is used to scale the surface detection image and the transmission projection image to a preset range; specifically, it is optional but not limited to scaling the pixel value of each image to a smaller range, such as 0 to 1 or -1 to 1, to speed up the training process and improve the stability of the model.

[0043] The data enhancement unit is used to increase the diversity of data through operations such as rotation, scaling, and flipping, thereby improving the generalization ability of the model.

[0044] The feature extraction layer B includes: a surface feature extraction unit B1 and a transmission feature extraction unit B2, which are used to extract surface features and transmission features respectively. The specific structural form can be arbitrarily set according to the size, accuracy, etc. of each image; preferably, these two feature extraction units can optionally include several convolutional layers, several batch normalization layers, activation functions and pooling layers connected in sequence.

[0045] In this embodiment, a preferred embodiment of the feature extraction layer is given, which may include several convolution layers for extracting features at different levels; each convolution layer is followed by an activation function to introduce nonlinearity; the activation function may be, but is not limited to, a ReLU function; and batch normalization is used between the convolution layer and the activation function to accelerate the training process and improve the stability of the model; finally, several pooling layers are used to reduce the dimension of the features while increasing the unique invariance to the image. It is worth noting that the specific structure of the feature extraction layer is only for illustrative purposes and is not limited thereto. It can be improved based on deep learning models such as VGG, ResNet, etc. to adaptively extract surface features and transmission features.

[0046] The fully connected layer C, located after the convolution layer and the pooling layer, is used to integrate surface features and transmission features to extract higher-level overall features and perform classification; preferably, the fully connected layer includes input neurons, output neurons and a weight matrix, followed by a classification function; the number of output neurons is equal to the number of ore categories. More preferably, if the ore types include concentrates and tailings, the classification function is preferably a binary classification function, such as a Sigmoid activation function, to determine whether the ore is a concentrate or tailings based on probability; if the ore types include concentrates, middlings and tailings, the classification function is preferably a multi-classification function, such as a Softmax function, to output the probabilities of each category separately.

[0047] More preferably, in the ore selection identification model:

[0048] The input layer A may also optionally include a surface statistical image input terminal A3 and a transmission statistical image input terminal A4, which are respectively used to input at least one surface statistical image and at least one transmission statistical image; the feature extraction layer B may optionally include a surface statistical feature extraction unit B3 and a transmission statistical feature extraction unit B4; the input of the surface statistical feature extraction unit B3 is connected to the surface statistical image input terminal A3, and the output is connected to the fully connected layer C, which is used to fuse the surface statistical images, extract surface statistical features, and access the fully connected layer; the input of the transmission statistical feature extraction unit B4 is connected to the transmission statistical image input terminal A4, and the output is connected to the fully connected layer C, which is used to fuse the transmission statistical images, extract transmission statistical features, and access the fully connected layer.

[0049] Specifically, the specific structure of the surface statistical image input terminal A3 and the transmission statistical image input terminal A4 can be designed by referring to the structure of the surface detection image input terminal A1 and the transmission projection image input terminal A2, and the specific structure of the surface statistical feature extraction unit B3 and the transmission statistical feature extraction unit B4 can be designed by referring to the structure of the surface feature extraction unit B1 and the transmission feature extraction unit B2, which will not be repeated here.

[0050] Specifically, the statistical graph may be, but is not limited to, a frequency histogram, a bar graph, a line graph, a fan graph, a frequency distribution histogram, a compound line graph, etc., to obtain statistical graphs of various initial graphs. It is worth noting that for each surface detection image or transmission projection image, the type of its statistical graph may be the same or different.

[0051] Preferably, in order to reduce the number of input ports in the input layer, multiple surface detection images or multiple transmission projection images can be optionally merged to calculate statistical images of each merged image as input to the surface statistical image input terminal A3 and the transmission statistical image input terminal A4.

[0052] More preferably, since there are many types of surface detection images and transmission projection images, as exemplified above, surface detection images may include: color images, near-infrared images, ultraviolet fluorescence images, black and white grayscale images, etc.; transmission projection images may include XRT images, dual-energy XRT images, etc.; and the acquisition equipment, i.e., the sensing equipment, included in each ore sorting machine is also different, and may only be able to collect any one or several of the above-mentioned surface detection images and transmission projection images. Therefore, for each ore sorting machine, the basic structure of its ore sorting identification model can be constructed according to step S1, and the surface features and transmission features are extracted respectively. However, the number and fine structure of channels of each input layer and feature extraction layer in the model, as well as various parameters will vary due to the number and specific type of surface detection graphics and transmission projection images.

[0053] Therefore, preferably, in step S1, a mineral processing identification model is constructed and trained, which may optionally include: constructing and training a mineral processing identification model group, including several of the mineral processing identification models; in step S2, one or more mineral processing identification models are selected from the mineral processing identification model group as the current identification model; and the acquired surface detection image and transmission projection image are input into the current identification model to obtain the mineral processing identification result.

[0054] Specifically, in one embodiment, each ore selection identification model in the ore selection identification model group can be selected to correspond to different surface detection images and transmission projection images. For example, for different surface detection images and transmission projection images, any combination thereof can be selected, such as any surface detection image and any transmission projection image, any multiple surface detection images and any multiple transmission projection images, to construct a corresponding ore selection identification model, and map and store them. In the current ore sorting machine, when the current surface detection image and the current transmission projection image are obtained through the sensor it has, the corresponding current identification model can be selected, and the ore selection identification can be carried out in a targeted manner, thereby improving the flexibility and pertinence of the model and further improving the identification efficiency.

[0055] More specifically, in another embodiment, each ore selection identification model in the ore selection identification model group may correspond to the same surface detection image and transmission projection image, respectively. For example, for the same surface detection image and transmission projection image, such as a combination of a color image, a near-infrared image, and an XRT image, two ore selection identification models are constructed, numbered as ore selection identification model 1 and ore selection identification model 2. In the current ore sorting machine, when these three images are obtained, ore selection identification model 1 and ore selection identification model 2 are input respectively to obtain different sorting results, and the two sorting results may be combined to obtain the final sorting result to further improve the sorting accuracy. Optionally, the final sorting result is determined by a voting mechanism, or the final sorting result is determined according to the sum of the output probabilities. For example, taking the separation of concentrates and tailings as an example, assuming that the probability of concentrates and tailings output by ore separation identification model 1 is A and B, and the probability of concentrates and tailings output by ore separation identification model 2 is C and D, then according to the probability sum - A+C and B+D, the probability of them being concentrates and tailings is finally determined to obtain the final separation result.

[0056] More preferably, since there are many types of surface detection images and transmission projection images, it is not only time-consuming and storage-intensive to build and train the ore selection identification model one by one, but also when encountering more advanced equipment, it is possible that a certain image obtained by it is not considered during the previous model construction and training. Therefore, in order to reduce the efficiency of model construction and training, reduce the storage capacity of the equipment, and also improve the scope of application of the method, it is optional to merge the models with the same number of input terminals. For example, taking the case where the surface detection images include 3 and the transmission projection images include 2 as an example, in its input layer, the surface detection image input terminal includes 3 ports, and the transmission projection image input terminal includes 2 ports, that is, a total of 5 ports, then no matter which three types of images the surface detection images are and which two types of images the transmission projection images are, as long as the number of ports is 3+2, a common ore selection identification model is shared, and the shared ore selection identification model can be trained with any 3 surface detection images and any 2 transmission projection images. Preferably, step S1, which constructs the training process, optionally includes:

[0057] S11: construct n surface detection images P respectively i and m transmission projection images Q j is the input, and the mining identification result is the N×M mining identification models under construction as the output; wherein n, m, N, M, i, j are all integers, and 0<n≤N, 0<m≤M, 0<i≤n; 0<j≤m; n represents the number of surface detection images in the mining identification model under construction; m represents the number of transmission projection images in the mining identification model under construction; N represents the total number of surface detection images; M represents the total number of transmission projection images; i represents the i-th surface detection image; j represents the j-th transmission projection image;

[0058] S12: For each mining identification model under construction, select C N n surface detection image sample library and C M m The transmission projection image sample library is used for training to obtain the trained intermediate ore selection recognition model;

[0059] S13: Determine the parameters of the ore selection identification model under construction according to the parameters of the trained intermediate ore selection identification model to obtain N×M trained ore selection identification models to form an ore selection identification model group.

[0060] For example, it is assumed that the conventional surface detection images include three, namely, a color image, a near-infrared image, and an ultraviolet fluorescence image; and the transmission projection images include two, namely, an XRT image and a dual-energy XRT image.

[0061] Then, step S11: 6 under-construction mineral processing identification models can be optionally constructed respectively, corresponding to taking 1 surface detection image and 1 transmission projection image as input; 1 surface detection image and 2 transmission projection images as input; taking 2 surface detection images and 1 transmission projection image as input; taking 2 surface detection images and 2 transmission projection images as input; taking 3 surface detection images and 1 transmission projection image as input; taking 3 surface detection images and 2 transmission projection images as input.

[0062] Step S12: Optionally, the ore type may be marked for each surface detection image and transmission projection image sample to construct a corresponding sample library, namely, a color image sample library, a near infrared image sample library, an ultraviolet fluorescence image sample library, an XRT image sample library, and a dual-energy XRT image sample library, for a total of 5 sample libraries. In the sample library, a corresponding number of sample libraries are selected for training.

[0063] Taking step S11, the ore dressing identification model under construction is: taking 1 surface detection image and 1 transmission projection image as input as an example, the color image sample library and the XRT image sample library are used for training, the near-infrared image sample library and the XRT image sample library are used for training, the ultraviolet fluorescence image sample library and the XRT image sample library are used for training, the color image sample library and the dual-energy XRT image sample library are used for training, the near-infrared image sample library and the dual-energy XRT image sample library are used for training, and the ultraviolet fluorescence image sample library and the dual-energy XRT image sample library are used for training, and 6 trained intermediate ore dressing identification models are obtained; similarly, taking 2 surface detection images and 2 transmission projection images as input as an example, 3 trained intermediate ore dressing identification models can be obtained, and taking 3 surface detection images and 2 transmission projection images as input as an example, only 1 trained intermediate ore dressing identification model can be obtained.

[0064] In step S13, preferably, the six trained intermediate ore selection identification models obtained in step S12 are taken as an example, and the parameters of the ore selection identification model with one surface detection image and one transmission projection image as input can be determined according to the parameters of the six trained intermediate ore selection identification models to obtain the corresponding ore selection identification model. By analogy, the parameters of the ore selection identification model with n surface detection images Pi and m transmission projection images Q can be obtained. j The input is N×M ore selection identification models with the ore selection identification result as the output, forming a ore selection identification model group.

[0065] More preferably, in step S13, the step of determining the parameters of the ore selection identification model under construction according to the parameters of the trained intermediate ore selection identification model is preferably:

[0066] Calculate the average value of the parameters of the intermediate ore selection identification model to determine the parameters of the ore selection identification model under construction; or set weights for the parameters of the intermediate ore selection identification model and determine the parameters of the ore selection identification model under construction based on the weights.

[0067] Similarly, taking step S12 as an example, for the six intermediate mineral processing identification models obtained by taking a surface detection image and a transmission projection image as input, the average value of the optional calculation parameter is the final model parameter value to obtain the final identification model of this type; or for the intermediate mineral processing identification model obtained by training with certain important images, such as near-infrared images and dual-energy XRT images, a higher weight is set, the parameters of each model are matched with weights, the final model parameter value is determined, and the final identification model of this type is obtained.

[0068] On the other hand, the present invention also provides an ore sorting machine, which adopts any of the above-mentioned multi-feature fusion ore sorting identification methods. Specifically, the specific structure of the ore sorting machine can adopt any form of the prior art, which will not be repeated here. The key of the present invention is to provide its ore sorting principle, and does not impose any limitation on its structure.

[0069] On the other hand, the present invention also provides a computer storage medium storing executable program code; the executable program code is used to execute any of the above-mentioned multi-feature fusion mineral processing identification methods.

[0070] On the other hand, the present invention also provides a terminal device, including a memory and a processor; the memory stores program codes that can be executed by the processor; the program codes are used to execute any of the above-mentioned multi-feature fusion mineral selection identification methods.

[0071] Exemplarily, the program code may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of completing specific functions, which are used to describe the execution process of the program code in the terminal device.

[0072] The terminal device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory. Those skilled in the art will appreciate that the terminal device may also include an input / output device, a network access device, a bus, etc.

[0073] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0074] The memory may be an internal storage unit of the terminal device, such as a hard disk or a memory. The memory may also be an external storage device of the terminal device, such as a plug-in hard disk, a smart memory card (SmartMedia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the terminal device. Further, the memory may also include both an internal storage unit of the terminal device and an external storage device. The memory is used to store the program code and other programs and data required by the terminal device. The memory may also be used to temporarily store data that has been output or is to be output.

[0075] The above-mentioned ore sorting machine, computer storage medium and terminal equipment are created based on the above-mentioned data generation method. Their technical functions and beneficial effects will not be repeated here. The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the various technical features in the above-mentioned embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0076] The above-mentioned embodiments only express several implementation methods of the present invention, and the description is relatively specific and detailed, but it cannot be understood as limiting the scope of the invention patent. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. A multi-feature fusion ore selection identification method, characterized in that: include: Build and train a mine identification model, including: An input layer, comprising: a surface detection image input terminal and a transmission projection image input terminal, for inputting at least one surface detection image and at least one transmission projection image respectively; The feature extraction layer includes: a surface feature extraction unit and a transmission feature extraction unit; the surface feature extraction unit is connected to the surface detection image input end and is used to fuse each surface detection image to extract the surface feature; the transmission feature extraction unit is connected to the transmission projection image input end and is used to fuse each transmission projection image to extract the transmission feature; The fully connected layer is connected to the surface feature extraction unit and the transmission feature extraction unit, and is used to fuse the surface features and the transmission features, extract the total features for classification, and output the mineral selection classification results; Acquire at least one surface detection image and at least one transmission projection image of the ore to be identified, and input them into the ore selection identification model to obtain the ore selection identification result; Constructing and training a mine selection identification model, including: constructing and training a mine selection identification model group, including several of the mine selection identification models; Before identification, one or more ore selection identification models are selected from the ore selection identification model group as the current identification model; the acquired surface detection image and transmission projection image are input into the current identification model to obtain the ore selection identification result; The steps to build and train the ore selection identification model group include: Construct n surface detection images P respectively i and m transmission projection images Q j is the input, and the mining identification result is the N×M mining identification models under construction as the output; wherein n, m, N, M, i, j are all integers, and 0<n≤N, 0<m≤M, 0<i≤n; 0<j≤m; n represents the number of surface detection images in the mining identification model under construction; m represents the number of transmission projection images in the mining identification model under construction; N represents the total number of surface detection images; M represents the total number of transmission projection images; i represents the i-th surface detection image; j represents the j-th transmission projection image; For each ore selection identification model under construction, select Surface detection image sample library and The transmission projection image sample library is used for training to obtain the trained intermediate ore selection recognition model; According to the parameters of the trained intermediate ore selection identification model, the parameters of the ore selection identification model under construction are determined to obtain N×M trained ore selection identification models to form an ore selection identification model group.

2. The ore selection identification method according to claim 1, characterized in that: The input layer further includes: a surface statistical image input terminal and a transmission statistical image input terminal, respectively used to input at least one surface statistical image and at least one transmission statistical image; a feature extraction layer including a surface statistical feature extraction unit and a transmission statistical feature extraction unit; The input of the surface statistical feature extraction unit is connected to the input end of the surface statistical image, and the output is connected to the fully connected layer, which is used to fuse the surface statistical images, extract the surface statistical features, and access the fully connected layer; The input of the transmission statistical feature extraction unit is connected to the transmission statistical image input terminal, and the output is connected to the fully connected layer, which is used to fuse the transmission statistical images, extract the transmission statistical features, and access the fully connected layer.

3. The ore selection identification method according to claim 2, characterized in that: A plurality of surface detection images or a plurality of transmission projection images are combined to calculate statistical images of the respective combined images as inputs to the surface statistical image input terminal and the transmission statistical image input terminal.

4. The ore selection identification method according to claim 1, characterized in that: Each ore selection identification model in the ore selection identification model group corresponds to a different surface detection image and transmission projection image; before identification, according to the currently acquired surface detection image and transmission projection image, a ore selection identification model is selected in the ore selection identification model group as the current identification model to obtain the ore selection identification result; Or, each mineral selection identification model in the mineral selection identification model group corresponds to the same surface detection image and transmission projection image; before identification, multiple mineral selection identification models are selected as current identification models in the mineral selection identification model group according to the currently acquired surface detection image and transmission projection image, and multiple mineral selection identification results are obtained; based on the multiple mineral selection identification results, a final mineral selection identification result is obtained.

5. The ore selection identification method according to any one of claims 1 to 4, characterized in that: The input layer includes two convolutional layers, one for inputting the surface detection image and the other for inputting the transmission projection image; each convolutional layer converts the number of channels of the image to 64 and uses a 3x3 convolution kernel for feature extraction.

6. The ore selection identification method according to claim 5, characterized in that: In the feature extraction layer, each feature extraction unit includes several convolutional layers, several batch normalization layers, activation functions and pooling layers connected in sequence.

7. The ore selection identification method according to claim 6, characterized in that: A fully connected layer, including input neurons, output neurons, and a weight matrix, followed by a classification function; the number of output neurons is equal to the number of ore categories.

8. An ore sorting machine, characterized in that: The ore dressing identification method according to any one of claims 1 to 7 is adopted.