A mineral microwave sorting method and system based on visible light and infrared light dual-mode cooperative imaging

Abstract

The application discloses a kind of ore microwave sorting method and system based on visible light and infrared light bimodulus collaborative imaging, belong to mineral processing technical field.Firstly, ore feeding and single layerization are transported, and different minerals are made to produce differential thermal response by microwave irradiation;Then visible light image and infrared thermal image are synchronously collected by common-aperture bimodulus imaging device;The outline of ore is extracted from the visible light image and a binary target mask is generated;The mask is used to spatially constrain the infrared thermal image, and only the temperature information of the ore region is retained;Based on the temperature information after constraint, the temperature characteristics of each ore particle are extracted, and the ore waste properties are distinguished;Finally, according to the discrimination result, position tracking and sorting are performed.This method uses the high-resolution boundary of visible light image to compensate for the lack of infrared image edge blur and particle adhesion, and suppresses background thermal interference through the mask, significantly improving the recognition accuracy and stability of ore sorting under complex conditions.

Description

Technical Field

[0001] This application belongs to the field of mineral processing technology, specifically relating to a microwave sorting method and system for ores based on dual-mode collaborative imaging of visible light and infrared light. Background Technology

[0002] With the continuous expansion of mineral resource mining and the sustained decline in ore grade, pre-selection and waste disposal before energy-intensive processes such as crushing and grinding has become an important means to improve resource utilization and reduce mineral processing costs. In recent years, non-contact ore sorting technology based on microwave heating and infrared thermal imaging has attracted widespread attention. This technology utilizes the differences in dielectric properties of different mineral components to generate differentiated thermal responses under microwave irradiation. Infrared thermal imaging is then used to collect the temperature distribution on the ore surface, thereby determining the ore grade or waste properties and enabling early waste disposal.

[0003] However, existing ore sorting methods based on single infrared thermal imaging have inherent limitations in practical applications. First, the edges of ore particles in infrared thermal images exhibit lower temperatures due to their thin geometric thickness and high heat dissipation rate, resulting in non-uniform attenuation of edge radiation intensity and difficulty in forming clear and sharp contour boundaries, affecting the accurate positioning of the actual spatial area occupied by the ore. Second, in scenarios involving continuous transport of multiple particles, adjacent ore particles come into contact or adhere to each other, appearing as connected thermal zones in infrared images, making it difficult to distinguish individual particle boundaries and thus causing deviations in target identification and positioning. Furthermore, some ore particles have weak microwave energy absorption capabilities, resulting in limited temperature rise under the same heating conditions, leading to indistinct temperature characteristics in their infrared thermal images. When the temperature of these particles is close to that of the background areas such as the conveyor belt and radiant plates, the thermal contrast between the target and the background decreases, resulting in a low signal-to-noise ratio in the infrared image, making it difficult to distinguish the ore target from the background.

[0004] The aforementioned problems limit the sorting accuracy and operational stability of infrared thermal imaging technology under complex working conditions.

[0005] To address the aforementioned shortcomings, there is an urgent need in the field for a sorting method that can accurately acquire the geometric boundaries of ore, effectively separate adhering particles, and achieve high-precision positioning of the target to be sorted. This is to improve the reliability and engineering practicality of microwave-heated infrared detection routes. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for microwave sorting of ores based on dual-mode collaborative imaging of visible light and infrared light, which can effectively solve the problems in the background art.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A microwave sorting method for ores based on dual-mode collaborative imaging of visible and infrared light includes the following specific steps: The ore to be sorted is fed and conveyed in a single layer to form a single-layer discrete particle flow. The monolayer discrete particle flow is heated by microwave irradiation using a microwave heating device, so that different mineral components produce differentiated thermal responses based on differences in dielectric properties. Visible light and infrared thermal images of heated ore bulk material were simultaneously acquired using a common aperture dual-mode imaging device. The visible light image is processed to extract the contour boundary of the ore particles, and a binary target mask is generated based on the contour boundary. The binary target mask is used to characterize the actual spatial area occupied by the ore particles in the visible light image. The binary target mask is fused with the infrared thermal image. The infrared thermal image is spatially constrained by the binary target mask, and only the temperature information in the infrared thermal image corresponding to the area occupied by the real space is retained. Based on the temperature information retained after constraints, the temperature characteristics of each ore particle are extracted, and the mineral waste attributes of each ore particle are determined according to the temperature characteristics. Based on the results of the mineral waste property identification, the location tracking and sorting of each ore particle are controlled.

[0008] A microwave ore sorting system based on dual-mode collaborative imaging of visible and infrared light includes: The ore feeding and single-layer conveying subsystem is used to process the ore to be sorted into a single-layer discrete particle flow and convey it. A microwave heating device is installed on the conveying path of the ore feeding and single-layer conveying subsystem to irradiate and heat the single-layer discrete particle flow with microwaves, so that different mineral components produce differentiated thermal responses. A common aperture dual-mode imaging device is set in the downstream detection area of ​​the microwave heating device to simultaneously acquire visible light images and infrared thermal images of the heated ore bulk. The common aperture dual-mode imaging device includes a visible light imaging unit, an infrared thermal imaging unit, a common aperture front-end optical system, a beam splitting element, and a synchronous triggering unit. The image acquisition, processing and sorting control system is connected to the common aperture dual-mode imaging device and the ore feeding and single-layer conveying subsystem, respectively, and is used to receive the visible light image and the infrared thermal image, and perform image processing, boundary extraction, mask generation, image fusion, temperature feature extraction and mineral waste attribute discrimination; A pneumatic discharge device is located downstream of the common aperture dual-mode imaging device and connected to the image acquisition, processing and sorting control system. It is used to perform point-by-point blowing separation of each ore particle based on the mineral waste property discrimination result.

[0009] Furthermore, the image acquisition module is used to simultaneously receive visible light images and infrared thermal images from the common aperture dual-mode imaging device, and to perform format conversion and timestamp synchronization; The image processing module is used to perform filtering, edge detection, morphological closing operation and connected component analysis on the visible light image, extract the contour boundary of the ore aggregate, and use the watershed algorithm to separate the adhering particles to generate a binary target mask. The image fusion module is used to perform pixel-by-pixel masking operations on the infrared thermal image using the binary target mask as a spatial constraint condition, and generate a fused image that retains only the temperature information of the ore bulk region. The feature extraction module is used to extract the temperature features of each ore particle from the fused image, the temperature features including statistical features, distribution features, hot spot features and regional features; The classification and discrimination module is used to classify the mineral waste attributes of each ore particle based on the temperature characteristics, using a temperature threshold method or a machine learning classifier, and output a discrimination label. The motion control module is used to calculate the time it takes for each ore particle to reach the sorting position based on the identification label and tracking position of the ore particles, combined with the conveyor belt running speed, the distance between the detection position and the sorting actuator, and the response time of the actuator, and to generate sorting control commands to send to the pneumatic discharge device.

[0010] In summary, this application includes at least one of the following beneficial technical effects: 1. This invention addresses the problems of significant edge thermal attenuation, blurred contours, and difficulty in accurately identifying boundaries in infrared thermal images of mineral aggregates after microwave heating. It fully utilizes the advantages of visible light images, such as high spatial resolution, clear boundary information, and strong detail expression, to accurately extract the true geometric boundaries of mineral aggregates. This provides a reliable spatial reference for the subsequent constraint and fusion of infrared temperature information, effectively overcoming the inherent limitations of single infrared thermal imaging in boundary recognition.

[0011] 2. This invention utilizes the characteristics of visible light images, which are not affected by thermal diffusion effects and have significant differences in material appearance, to perform adhesion separation processing on the contours of ore particles in the visible light domain, thereby achieving independent particle division, avoiding recognition errors caused by thermal fusion, and significantly improving the recognition accuracy and operational stability in scenarios of continuous multi-particle transport.

[0012] 3. This invention limits the extraction range of infrared temperature features by using a target mask, so that the conveyor belt, chute, environmental radiation area and other non-ore background areas do not participate in temperature statistics and mineral waste discrimination, effectively suppressing the influence of background thermal interference on the identification results, realizing accurate positioning and quantitative analysis of target ore particles, and significantly improving the accuracy and reliability of identification.

[0013] 4. This invention integrates the real geometric boundary accurately extracted in the visible light domain with the thermal response features in the infrared domain to achieve early pre-selection and waste disposal before high-energy-consuming processes such as crushing and grinding. This effectively reduces the amount of subsequent processing, lowers unit energy consumption, and improves the overall economic benefits of the mine, providing reliable technical support for the efficient utilization of low-grade and complex-component ore resources. Attached Figure Description

[0014] Figure 1 A schematic diagram of the overall scheme for a microwave ore sorting method based on dual-mode collaborative imaging of visible and infrared light; Figure 2 A schematic diagram illustrating the core principle of collaborative fusion of visible light and infrared dual-mode images; Figure 3 Flowchart for synchronous acquisition and calibration of a common aperture dual-mode imaging device; Figure 4 This is a flowchart of the image processing from boundary extraction and target mask generation to mask-constrained infrared temperature feature extraction. Figure 5 This is a comparison chart showing the performance of dual-mode collaborative imaging and single infrared imaging in terms of boundary recognition accuracy and multi-particle scenarios. Figure 6 This is a schematic diagram of the actual operation results. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of this invention clearer, the following description is provided in conjunction with the appendix. Figure 1 To be continued Figure 6 The present invention will be further described in detail below with reference to specific embodiments.

[0016] This invention provides a microwave sorting method for ores based on dual-mode collaborative imaging with visible and infrared light. The method involves placing bulk ore in an industrial microwave field for differentiated heating. A dual-mode imaging device with a common aperture for visible and infrared light is used to simultaneously acquire the geometric contour information and infrared temperature distribution information of the bulk ore. Image processing technology is used to accurately extract the true spatial boundaries of the bulk ore from the visible light image and generate a target mask. The target mask is then used to spatially constrain the infrared thermal image to extract the true temperature characteristics of the ore region. Finally, based on these temperature characteristics, the automatic identification and separation of mineral waste attributes are achieved.

[0017] The system architecture of this invention comprises five core subsystems: an ore silo and conveying system, a microwave heating device, a common-aperture dual-mode imaging device, an image acquisition, processing, and sorting control system, and a pneumatic discharge device. The ore silo and conveying system is responsible for crushing the ore to be sorted to a suitable particle size and then conveying it to the conveyor belt via a silo and a vibrating feeder, achieving continuous and uniform material supply. The microwave heating device uses a 2.45 GHz industrial magnetron microwave source to irradiate and heat the ore bulk on the conveyor belt, causing different mineral components to produce differentiated thermal responses due to differences in dielectric properties. The common-aperture dual-mode imaging device simultaneously acquires visible light reflection images and infrared thermal radiation images of the ore bulk through the same front-end incident aperture, ensuring spatial consistency of the dual-mode images. The image acquisition, processing, and sorting control system completes core processing steps such as image preprocessing, boundary extraction, mask generation, pixel-level fusion, temperature feature extraction, and waste material identification. The pneumatic discharge device drives a pneumatic blowing mechanism according to sorting control commands to achieve targeted blowing separation of ore and waste rock.

[0018] The ore feeding and single-layer conveying subsystem includes components such as an ore silo, a vibrating feeder, a heat-resistant conveyor belt, and its drive motor. The ore silo stores the ore to be sorted; its capacity is determined by the production line's processing capacity and is typically designed to hold the amount of ore required for at least 30 minutes of continuous production. The vibrating feeder drives the trough to generate periodic vibrations via an electromagnetic vibrator, achieving uniform ore distribution and flow control. The vibration frequency range of the vibrating feeder is typically set between 50 Hz and 60 Hz, and the amplitude can be adjusted from 0.5 mm to 3 mm according to the ore's flowability. The heat-resistant conveyor belt is made of heat-resistant, low-microwave-absorption, or low-infrared-reflection materials. Its width is determined by the ore throughput and particle size, typically ranging from 800 mm to 2000 mm, and its operating speed can be adjusted from 0.5 m / s to 3 m / s according to the required sorting accuracy.

[0019] In this application, step S1 is ore feeding and single-layer conveying. The core purpose is to process the ore to be sorted into a single-layer discrete particle flow with suitable particle size and stably feed it into the subsequent microwave heating area so that each particle can independently and fully absorb microwave energy. Specifically, it is carried out according to the following steps.

[0020] Step S101: The ore to be sorted is first crushed to reduce the particle size to a suitable range for sorting. The crushing operation uses a jaw crusher or a cone crusher to crush the raw ore to a particle size of less than 80mm. Then, a screening operation is performed to separate a specific particle size range suitable for microwave sorting from the crushed products. The specific particle size range is usually 10mm to 50mm.

[0021] In step S102, the crushed and screened ore particles enter the ore silo through a chute and are uniformly fed to the heat-resistant conveyor belt by a vibrating feeder.

[0022] The vibrating feeder drives the trough to generate periodic vibration through an electromagnetic vibrator. The vibration frequency range is set from 50Hz to 60Hz, and the amplitude is adjusted from 0.5mm to 3mm according to the flowability of the ore.

[0023] The heat-resistant conveyor belt is made of heat-resistant, low microwave absorption and low infrared reflection materials. The width is determined according to the ore processing volume and particle size, usually from 800mm to 2000mm. The running speed can be adjusted from 0.5m / s to 3m / s according to the sorting accuracy requirements.

[0024] In step S103, the feeding rate of the vibrating feeder is coordinated with the conveyor belt speed through a closed-loop control system. The controller collects the current operating speed of the conveyor belt from the speed sensor in real time, and dynamically adjusts the drive voltage of the vibrating feeder according to the matching relationship between material flow rate and conveying capacity, thereby achieving adaptive control of the feeding rate.

[0025] The specific control method of coordinated regulation is as follows: the controller calculates the theoretically required material flow rate based on the current running speed of the conveyor belt, compares the theoretically required material flow rate with the actual output of the vibrating feeder, calculates the feeding rate deviation value through a PID algorithm, and outputs a corresponding control voltage adjustment signal to the power drive circuit of the vibrating feeder. The adjustment response time of the feeding rate is less than 500ms to meet the real-time control requirements under dynamic working conditions.

[0026] Step S104: Through the aforementioned coordinated control, the ore particles are laid out in a single layer on the heat-resistant conveyor belt along the conveying direction. During the conveying process, the ore particles naturally spread out on the surface of the conveyor belt by gravity, maintaining a loose contact between the particles without significant accumulation.

[0027] Determining the conveyor belt speed requires considering the microwave heating power, target temperature rise requirements, and the residence time of the ore in the microwave field. Let the length of the microwave heating zone be... The unit is Microwave heating power is The unit is The required temperature rise for the ore is The unit is Heating time Determined through the heat balance equation, the calculation formula is as follows: in The total mass of ore entering the heating zone, in units of ; Specific heat capacity of ore, unit: ; The effective power coefficient of microwaves is dimensionless. Based on the relationship between heating time and delivery speed... The corresponding conveyor belt running speed can be calculated. The unit is .

[0028] Step S105: The microwave heating device is used to irradiate and heat the bulk ore on the conveyor belt. The microwave heating device includes a microwave generator, a microwave transmission waveguide, a microwave cavity, an observation window, and a protective shielding structure.

[0029] The microwave generator uses a magnetron as the microwave source, with a fixed output frequency of 2.45 GHz. The rated power range can be selected from 5 kW to 100 kW depending on the characteristics of the mineral and the processing volume requirements. The microwave energy generated by the magnetron is transmitted to the microwave cavity through a rectangular waveguide or coaxial transmission line.

[0030] The microwave cavity uses a metal shielding material for its inner wall, preferably 304 stainless steel or aluminum alloy, with a wall thickness of not less than 3mm, to ensure uniform distribution of microwave energy within the cavity and prevent microwave leakage. The structural form of the microwave cavity is determined based on the ore processing volume and heating uniformity requirements, and can be a single-mode cavity, multi-mode cavity, or traveling wave cavity.

[0031] For large-scale industrial production, multi-mode cavities are used to obtain a larger heating area and a more uniform energy distribution. One specific implementation of a multi-mode cavity involves placing one or more motor-driven rotating stirrers inside a rectangular cavity. By dynamically changing the cavity boundary conditions, the electromagnetic field distribution pattern is disrupted, thereby achieving uniform energy distribution over time.

[0032] In summary, step S1 provides a complete description of the ore feeding, single-layer conveying, and microwave heating device. Step S1 controls the feed particle size through crushing and screening, achieves single-layer particle spreading through closed-loop coordination of feed rate and conveying speed, and correlates conveying speed and microwave heating parameters through a heat balance equation. It also provides the specific structure of the microwave heating device, laying a complete process and equipment foundation for the differentiated thermal response of the ore under the microwave field based on the difference in dielectric properties in subsequent step S2, as well as for the subsequent image acquisition by the dual-mode imaging system.

[0033] For step S2, following step S1, after the ore to be sorted enters the microwave heating area in a single-layer discrete state, the different mineral components are irradiated by microwave to produce differentiated thermal responses, and the following steps are followed.

[0034] In step S201, after the ore bulk enters the microwave heating zone, microwave energy penetrates the material and interacts with the mineral components. The physical mechanism of microwave heating is as follows: as a high-frequency electromagnetic wave, microwaves exert periodic oscillations on the polar molecules and charged ions in the material through their electric field components. Polar molecules repeatedly polarize and change direction with the electric field, while charged ions undergo directional migration under the influence of the electric field. The polarization and ion migration processes require overcoming intermolecular forces and ion migration resistance, thereby converting electromagnetic energy into thermal energy.

[0035] The heating power density of microwaves on materials is determined by the dielectric properties of the materials, and the calculation formula is as follows: in, This is the microwave heating power density, expressed in W / m³. The vacuum permittivity is 8.854 × 10⁻⁶. - ¹²F / m; is the relative permittivity of the material, which is dimensionless; The tangent of the dielectric loss angle is dimensionless. This represents the effective value of the electric field strength, expressed in V / m.

[0036] In step S202, different mineral components have significantly different dielectric parameters due to differences in chemical composition, crystal structure, water content, and metal mineral content.

[0037] Quartz, feldspar, and other silicate minerals typically have a relative permittivity of 4 to 6 and a dielectric loss tangent of less than 0.05, exhibiting weak absorption of microwave energy. Pyrite, chalcopyrite, magnetite, and other metal sulfides and oxides, with relative permittivity exceeding 10 and a dielectric loss tangent greater than 0.1, demonstrate strong absorption of microwave energy. Hydrous minerals exhibit high dielectric loss due to the strong polarity of water molecules. When ores with different dielectric parameters receive the same electromagnetic energy in the same microwave field, the different dielectric loss factors result in varying volumetric heating effects, leading to differences in temperature distribution.

[0038] Step S203: The heating process of the ore particles in the microwave field follows the heat conduction equation. Treating the ore particles as a homogeneous medium, the partial differential equation for the internal temperature distribution over time is: in, This represents the density of the ore, expressed in kg / m³. Specific heat capacity of ore, expressed in J / (kg·K); Temperature, in Kelvin (K). Time, in seconds; Thermal conductivity, expressed in W / (m·K); The microwave heating power density is defined in step S201, with units of W / m³.

[0039] Because the size of ore particles is on the order of millimeters to centimeters, the relaxation time of heat conduction is much shorter than the microwave heating time. Therefore, it can be approximated that the internal temperature distribution of the ore particles is uniform. The temperature rise of the ore particles when they leave the microwave heating zone depends mainly on the dielectric properties of the particles, the microwave power density, and the heating time. For a single ore particle, the temperature rise... It can be simplified to: in, Heating time, in seconds.

[0040] In step S204, the differentiated thermal response of the ore bulk after microwave irradiation manifests in three forms.

[0041] The first form is the difference in temperature amplitude: the temperature rise of ore particles with high dielectric loss can reach 30°C to 80°C, while the temperature rise of particles with low dielectric loss is only 5°C to 20°C.

[0042] The second type is the difference in temperature distribution characteristics: ore particles containing metal mineral inclusions or fissure water may form local hot spots due to concentrated absorption of microwave energy; homogeneous ore particles have a more uniform temperature distribution.

[0043] The third form is the difference in heating rate: under the same heating conditions, the temperature rise rate of high-loss ore is higher than that of low-loss ore, and different types of ore can be further distinguished by the temperature change rate.

[0044] The calibration of parameters such as microwave power, heating time, and conveyor speed needs to be optimized through experiments for specific mineral types and particle sizes. Typical value ranges are: microwave power density 0.5W / cm² to 5W / cm², heating time 2s to 10s, and conveyor belt speed 0.5m / s to 2m / s.

[0045] Step S2 involves using microwave irradiation to induce differentiated thermal responses in different mineral components. After the ore aggregate leaves the microwave-heated area, its surface temperature distribution carries thermal characteristic information that distinguishes ore from waste rock. To accurately extract this thermal characteristic information and simultaneously acquire the spatial geometric boundaries of the ore aggregate, this application employs a common-aperture dual-mode imaging device to simultaneously acquire visible light and infrared thermal images within the detection area after the ore aggregate leaves the microwave-heated area.

[0046] The common-aperture dual-mode imaging device is the core sensing component of this application. It simultaneously acquires visible light and infrared thermal images of the ore aggregate through the same front-end incident aperture, ensuring spatial consistency and temporal synchronization of the dual-mode images. The common-aperture dual-mode imaging device includes five functional modules: a visible light imaging unit, an infrared thermal imaging unit, a common-aperture front-end optical system, a beam splitter, and a synchronization triggering unit.

[0047] The detector of the visible light imaging unit adopts an area array CMOS sensor or an area array CCD sensor, with a resolution set from 1920×1080 pixels to 4096×3072 pixels, a pixel size from 3μm to 6μm, and a frame rate of no less than 60fps to meet real-time processing requirements.

[0048] The visible light imaging unit has a spectral response range of 380nm to 780nm, imaging the reflected light from the ore surface to acquire information on the ore's geometry, surface texture, color, and spatial location. Visible light images have high spatial resolution, clearly presenting the edge contours and surface details of ore particles.

[0049] The detectors in the infrared thermal imaging unit employ either cooled mercury cadmium telluride detectors or uncooled vanadium oxide microbolometers. Cooled mercury cadmium telluride detectors have a sensitivity of 20 mK to 30 mK and operate in the 3 μm to 5 μm or 8 μm to 14 μm wavelength range; uncooled vanadium oxide microbolometers typically have a sensitivity of 30 mK to 50 mK.

[0050] The resolution of the infrared detector is selected according to application requirements, with typical specifications ranging from 640×512 pixels to 1280×1024 pixels and a frame rate of no less than 30fps. The spectral response range of the infrared thermal imaging unit is 3μm to 14μm, corresponding to the peak band of thermal radiation of objects at room temperature, and can accurately obtain temperature distribution information on the surface of minerals.

[0051] A common-aperture front-end optical system receives optical signals from the target scene through the same incident aperture and guides them to different detectors. The common-aperture front-end optical system employs a coaxial reflective design, with the primary reflector being a parabolic or spherical mirror with an aperture ranging from 100mm to 300mm. The focal length is determined based on the working distance and field-of-view requirements.

[0052] A beam-splitter is placed in the main optical path to separate the incident light into visible and infrared light according to wavelength range. The beam-splitter can be a beam-splitter prism or a beam-splitter filter. The beam-splitter prism utilizes the difference in refractive index of materials for different wavelengths to achieve angular separation; the beam-splitter filter employs a multilayer dielectric film structure to achieve high transmission and high reflection for visible and infrared light, respectively. The material of the beam-splitter simultaneously satisfies the transmission characteristics of both the visible and infrared bands; typical materials include germanium, silicon, and zinc selenide. These infrared materials are then coated with an anti-reflection film for use in the visible light band.

[0053] The synchronous triggering unit ensures the temporal consistency of the dual-mode images. In high-speed dynamic scenarios of ore sorting, visible light images and infrared thermal images are acquired at the same time or within a very short time interval to avoid spatial misalignment between the dual-mode images caused by ore movement.

[0054] The synchronous triggering unit employs a hardware synchronization method. The controller simultaneously sends trigger pulse signals to both the visible light imaging unit and the infrared thermal imaging unit. The two detectors synchronously begin exposure and readout at the rising or falling edge of the trigger pulse signal. The timing jitter of the synchronous triggering unit is controlled within 100μs, ensuring that the acquisition time difference of the dual-mode images meets the accuracy requirements of pixel-level fusion. For a scenario where the conveyor belt runs at a speed of 1m / s, the spatial displacement corresponding to a 100μs time difference is only 0.1mm, which is less than the spatial resolution corresponding to a detector pixel.

[0055] Step S3: Synchronous acquisition of visible light and infrared thermal images. After the ore aggregate completes microwave irradiation and leaves the microwave heating area, it enters the detection area of ​​the common aperture visible light-infrared dual-mode imaging device. Before this, the ore aggregate undergoes a brief cooling process of 3 ms to 8 ms, and enters the field of view of the imaging device only after the surface temperature field of the ore aggregate tends to stabilize.

[0056] The common aperture dual-mode imaging device is installed at a distance of 500 mm to 1500 mm from the microwave heating area. For a conveyor belt with a width of 1200 mm, the working distance of the common aperture dual-mode imaging device is 800 mm to 1500 mm, corresponding to a field of view of 40° to 60°.

[0057] Step S301: The synchronous triggering unit generates a trigger pulse. After detecting that the ore particles have entered the imaging area, the synchronous triggering unit simultaneously sends trigger pulses to both the visible light imaging unit and the infrared thermal imaging unit. The trigger signal is generated in the following two ways: Method 1 is the external trigger mode. The photoelectric sensor detects the position of the ore and outputs a trigger signal.

[0058] Method two is the continuous acquisition mode. The camera continuously acquires images at a fixed frame rate, and the controller predicts the triggering timing based on the ore position at the software level. The prediction method is as follows: the controller estimates whether the ore has entered the imaging area based on the conveyor belt speed and the known distance of the ore from the microwave heating area outlet to the imaging area entrance, combined with the position of the ore in the previous visible light image, thereby determining the triggering timing.

[0059] Step S302: Simultaneous acquisition and quality verification of dual-mode images. After the trigger pulse arrives, the visible light imaging unit begins exposure and reads a frame of visible light image, while the infrared thermal imaging unit begins integration and reads a frame of infrared thermal image. After receiving the two image data streams, the controller timestamps and verifies the quality of the visible light and infrared thermal images.

[0060] The quality verification includes checking whether the visible light image and infrared thermal image are complete without missing frames, whether the pixel data is within the valid range, and whether the acquisition time difference between the visible light image and the infrared thermal image meets the synchronization accuracy requirements.

[0061] The synchronization accuracy requirement is defined as follows: the spatial displacement corresponding to the time difference is less than the spatial resolution corresponding to the detector pixel. For a scenario where the conveyor belt runs at a speed of 1 m / s, the spatial displacement corresponding to a 100 μs time difference is 0.1 mm, which is less than the spatial resolution corresponding to the detector pixel.

[0062] Step S303: Explanation of the purpose of the acquired image information. The acquired visible light image is used to obtain the geometric contours, spatial positions, and image differences between the ore particles and the background. The visible light image has a grayscale level of 8 to 16 bits, and its dynamic range is adaptively adjusted according to the scene brightness. Ore particles appear as target areas with specific textures and colors in the visible light image, exhibiting grayscale or color differences compared to the conveyor belt background. The features of the visible light image are unaffected by the microwave heating process and can accurately reflect the spatial occupancy of the ore particles.

[0063] The acquired infrared thermal images are used to obtain the surface temperature distribution and thermal response characteristics of the ore bulk. Each pixel value of the infrared thermal image corresponds to the infrared radiation intensity at that location in the target scene, and after emissivity correction and temperature calibration, it is converted into a surface temperature value. The temperature resolution of the infrared thermal images is 0.03°C to 0.1°C, which is sufficient to detect subtle temperature differences between ore particles.

[0064] Step S304: Spatial calibration of the common aperture dual-mode imaging device. After the common aperture dual-mode imaging device is installed, it is necessary to establish the mapping relationship between the visible light image coordinate system and the infrared thermal image coordinate system. Spatial calibration is performed according to the following specific steps.

[0065] The first step involves placing a calibration plate with specific geometric features within the field of view of the common-aperture dual-mode imaging device. The calibration plate contains both visible light and infrared feature points. The visible light feature points are represented by checkerboard corner points. The infrared feature points are represented by blackbody radiation regions of known temperatures, specifically achieved by embedding small blackbody radiation sources at corresponding positions on the calibration plate.

[0066] The second step involves controlling the common-aperture dual-mode imaging device to simultaneously acquire visible light and infrared thermal images of the calibration plate. Image processing algorithms are then used to identify the pixel coordinates of feature points on the calibration plate within the visible light and infrared thermal images. Specifically, for the visible light image, Harris corner detection or checkerboard corner sub-pixel localization algorithms are employed; for the infrared thermal image, grayscale centroid method or threshold segmentation method is used to locate the center of the blackbody region.

[0067] The third step involves calculating the spatial transformation matrix between the visible light image and the infrared thermal image based on the known positional relationships of the feature points in physical space. The spatial transformation matrix can be calculated using either a perspective transformation model or a polynomial transformation model. The perspective transformation model requires at least four corresponding feature point pairs, while the polynomial transformation model requires more point pairs to improve accuracy.

[0068] The fourth step is to store the calibrated spatial transformation matrix in the image acquisition, processing, and sorting control system for later use in image fusion processing. The calibration accuracy is controlled within one pixel.

[0069] In summary, step S3 completes the synchronous acquisition and spatial calibration of the dual-mode images. Through time synchronization and spatial registration, a pixel-level correspondence is established between the visible light image and the infrared thermal image, providing a reliable foundation for subsequent fusion processing based on visible light boundary extraction and constraint of infrared temperature information.

[0070] The next step is step S4 in this embodiment: ore particle boundary extraction based on visible light images. In the acquired visible light images, there are grayscale or color differences between the ore particles and the conveyor belt background. By performing four image processing operations on the visible light image—filtering, edge detection, morphological closing operation, and connected component analysis—the contour boundaries of each ore particle are extracted, and this is carried out according to the following steps.

[0071] Step S401, filtering process, the filtering process suppresses noise interference in the visible light image. Image noise sources include sensor noise, circuit noise, random texture noise on the mineral surface, and local reflection interference.

[0072] The filtering algorithm uses Gaussian filtering or median filtering.

[0073] Gaussian filtering is a linear smoothing filter; its kernel function is a two-dimensional Gaussian distribution. The filtering equation for Gaussian filtering is: in, For the filtered pixel in coordinates The grayscale value at that location; and These are the x and y coordinates of pixels, in pixels. The standard deviation of the Gaussian kernel controls the smoothness level, and the unit is pixels; Pi is a constant.

[0074] Gaussian filtering is effective in suppressing Gaussian noise while preserving the edge features of the image, making it suitable for scenarios with moderate noise levels.

[0075] Median filtering is a non-linear filter. The principle of median filtering is to sort the gray values ​​of all pixels within the filtering window by size and take the median as the output pixel value. Median filtering has a good suppression effect on salt-and-pepper noise and impulse noise, while having a relatively small blurring effect on image edges, making it suitable for scenes with random bright or dark spots.

[0076] The size of the filter kernel is determined based on the noise level and particle size. For ores with a particle size of 10 mm to 50 mm, the filter kernel size is taken as 3×3 to 7×7 pixels, with a standard deviation of [missing value]. Take 1 to 2 pixels.

[0077] Step S402, edge detection: Edge detection extracts the preliminary contours of the ore particles from the filtered image. The edge detection algorithm employs at least one of the Canny algorithm, Sobel algorithm, or Prewitt algorithm.

[0078] The Canny algorithm is a multi-stage edge detection algorithm. Its processing flow includes: Gaussian smoothing filtering for noise reduction, calculating image gradient magnitude and direction, non-maximum suppression for edge thinning, and dual-threshold detection of connected edges. The dual thresholds in the Canny algorithm include a high threshold... and low threshold , and These are all dimensionless gradient magnitude thresholds. Gradient magnitude greater than... The pixels are marked as strong edge pixels, and the gradient magnitude is between and Pixels between these points are marked as weak edge pixels, with gradient magnitudes less than 1. The pixels are suppressed.

[0079] Weak edge pixels are only retained as valid edges when they are connected to strong edge pixels. The Canny algorithm has high edge detection accuracy and strong noise resistance, and is suitable for scenes with complex textures and irregular edges, such as loose minerals.

[0080] The Sobel algorithm uses convolutional kernels in two directions to calculate the gradients of an image in the horizontal and vertical directions, respectively. The gradient magnitude is obtained by taking the square root of the sum of the squares of the gradients in both directions. The Sobel algorithm is computationally efficient and suitable for scenarios with high real-time requirements.

[0081] The Prewitt algorithm is similar in principle to the Sobel algorithm, but the convolution kernel coefficients of the Prewitt algorithm are different from those of the Sobel algorithm, and the Prewitt algorithm is slightly less sensitive to noise than the Sobel algorithm.

[0082] Step S403, morphological closing operation: Morphological closing operation fills the contour gaps in the edge detection results. Due to the roughness and irregular shape of the ore surface, and the edge blurring that may be caused by filtering, the contours obtained by edge detection usually have discontinuities and non-continuous regions.

[0083] Morphological closing operations consist of dilation and erosion operations combined in the order of dilation followed by erosion. The dilation operation is defined as replacing the gray value of the current pixel with the maximum gray value within the area covered by the structuring element. The purpose of the dilation operation is to expand the boundary of high gray value areas and fill in small dark areas.

[0084] The erosion operation is defined as replacing the gray value of the current pixel with the smallest gray value within the area covered by the structuring element. The effect of erosion is to shrink the boundaries of high gray value regions and eliminate small bright areas. Morphological closing operations, through a process of first dilation and then erosion, fill small gaps in the contour without significantly altering the shape and area of ​​the original contour.

[0085] The shape of the structuring element can be circular, square, or cross-shaped, and its radius or size is determined based on the gap width. For filling the gaps in the outline of ore aggregates, the radius of the structuring element is 2 to 5 pixels.

[0086] Step S404, Connected Component Analysis: Connected component analysis removes small noise areas in the edge image and marks each individual ore particle.

[0087] Connected component analysis employs either seed-fill-based or scan-line-based algorithms. The algorithm first traverses all pixels in the image, grouping adjacent pixels with the same grayscale value into the same connected component. Then, it calculates the area, centroid position, and bounding rectangle of each connected component. Based on a preset minimum area threshold, noise regions smaller than this threshold are removed. The setting of the minimum area threshold requires consideration of both the smallest particle size of the ore and the image resolution.

[0088] For example, for an imaging device with a resolution of 1920×1080 pixels and a field of view of 1200 mm, the spatial size corresponding to a single pixel is approximately 0.625 mm. Assuming the minimum ore grain size is 10 mm, the minimum ore area threshold is set to 150 pixel² to 200 pixel².

[0089] After the connected region analysis is completed, each ore particle corresponds to an independent connected region, and the pixel set of each connected region is the preliminary outline of the ore particle.

[0090] Step S4 completes the entire process of extracting the contour boundaries of the ore particles from the visible light image. The resulting independent contour of each ore particle will serve as the basis for separating the adhering particles in step S5 and generating the target mask in step S6.

[0091] For step S5, the separation of adhering particles: The preliminary outline of the ore particles obtained in step S4 will merge into a connected region when multiple ore particles come into contact or adhere to each other on the conveyor belt. To obtain the independent outline of each particle, the adhering region needs to be segmented, which is carried out according to the following steps.

[0092] Step S501: Select a segmentation algorithm and perform a distance transformation. Use the watershed algorithm to separate the adhering particles. The watershed algorithm treats the image grayscale value as the terrain height and simulates the process of water being injected into the terrain surface: the water flow starts from a local low point and rises. When the floods from two different injection areas meet, a watershed is formed to separate the two areas.

[0093] When the watershed algorithm is used to separate cohesive ore particles, it first performs a distance transformation on the binary contour image obtained in step S4. The distance transformation calculates the distance from each background pixel to the nearest particle boundary. Specifically, for each non-particle pixel (background pixel) in the binary contour image, the minimum Euclidean distance between that pixel and all particle boundary pixels is calculated, and this minimum distance value is assigned to the background pixel. The result of the distance transformation forms a distance map, where pixels far from the boundaries have larger distance values, and pixels located in the particle center region have the largest distance values.

[0094] The specific implementation of the distance transformation algorithm can adopt the multi-channel scanning method or the Euclidean distance transformation method.

[0095] Multi-channel scanning calculates the chessboard distance or city distance from each pixel to the nearest zero pixel through two scans. The chessboard distance is defined as the maximum difference between two pixels in the horizontal and vertical coordinate directions, while the city distance is defined as the sum of these differences. Multi-channel scanning is computationally efficient but has limited distance measurement accuracy.

[0096] The Euclidean distance transformation method achieves accurate distance measurement by calculating the Euclidean distance between pixels. The distance calculation formula is as follows: ,in and These are the coordinates of two pixels, in pixels. This method is highly accurate but computationally intensive.

[0097] Step S502: Seed point selection and flood filling segmentation. Thresholding is performed on the distance map obtained by distance transformation, and local maxima are selected as seed points. Each seed point corresponds to the center position of an ore particle.

[0098] The flooding process begins at a seed point and extends outwards to the surrounding area. During flooding, floods from different seed points meet at ridges with equal distance values; these ridges serve as dividing lines. By overlaying these dividing lines onto the original contour image, the contiguous particle region can be segmented into multiple independent particle contours.

[0099] For separating adherent particles, an improved labeled watershed algorithm is employed. This improved algorithm sets foreground and background markers on the distance transformation map to control the flooding process. Foreground markers are placed at local maxima on the distance map, i.e., within the particle interior, while background markers are placed at the boundaries between particle gaps. By specifying foreground and background markers, the watershed algorithm generates dividing lines only between these markers, thus avoiding over-segmentation.

[0100] Step S503: Segmentation effect evaluation and post-processing optimization. After segmentation, the segmentation effect needs to be evaluated. The evaluation indicators of separation error include undersegmentation rate and oversegmentation rate. Undersegmentation refers to two or more particles being incorrectly merged into one region, while oversegmentation refers to a single particle being incorrectly segmented into multiple regions.

[0101] The number of individual particle profiles after segmentation should match the actual number of particles. The ratio of the number of individual particle profiles after segmentation to the actual number of particles should be controlled within the allowable error range of 0.95 to 1.05.

[0102] For cases where the separation effect is unsatisfactory, post-processing optimization of the segmentation results can be performed by adjusting the parameters of the distance transform or by employing morphological operations. Adjusting the parameters of the distance transform includes changing the threshold of the distance transform algorithm or adjusting the precision of the Euclidean distance calculation. Morphological post-processing includes merging over-segmented regions and re-segmenting under-segmented regions.

[0103] In summary, steps S501 to S503 together complete the separation of the adhered particles. After separation, the previously adhered ore particles obtain their own independent closed contours, each contour corresponding to the real spatial boundary of an ore particle, providing accurate input for generating the target mask in step S6.

[0104] Regarding step S6 in this embodiment, based on the closed contour of the ore aggregate obtained in step S4 and the independent particle contours separated in step S5, the target ore region in the visible light image is filled to generate a binary target mask. The binary target mask is used to characterize the actual spatial area occupied by the ore aggregate in the visible light image, wherein the pixel value of the ore aggregate region is set to 255, corresponding to white; and the pixel value of the background region is set to 0, corresponding to black.

[0105] Step S601: Generate a binary target mask by converting the particle contour image obtained in step S5 into a binary image, wherein the inside of the contour line is filled with the foreground color and the outside of the contour line is filled with the background color.

[0106] The filling algorithm uses either scan line filling or seed filling.

[0107] The scan-line fill method starts scanning line by line from the top of the image. Filling begins when the scan line enters the foreground area and stops when it leaves the foreground area, until the entire image has been scanned.

[0108] The seed filling method starts from a seed point and expands outward in a 4-connected or 8-connected manner, marking all pixels connected to the seed point as the foreground color.

[0109] Binary mask The mathematical expression is: in, Indicates pixel coordinates The mask value at the specified location, expressed in gray levels; and These are the x and y coordinates of a pixel, respectively, in pixels.

[0110] The resolution of the binary target mask is consistent with the original visible light image, ensuring a one-to-one correspondence between mask pixels and image pixels. In the mask image, a pixel value of 255 indicates that the location belongs to the ore aggregate, while a pixel value of 0 indicates that the location belongs to the background. The generated target mask is used to distinguish the ore aggregate region from the conveyor belt, environmental background, and other non-target regions, retaining only the target region information corresponding to the true geometric boundaries of the ore aggregate.

[0111] Step S602, Coordinate Mapping of the Target Mask: Before using the target mask for subsequent infrared thermal image fusion, coordinate mapping of the target mask is required. The target mask is generated based on the pixel coordinate system of the visible light image, while the infrared thermal image has an independent pixel coordinate system. The spatial resolution, field of view center, and pixel size of the two may differ.

[0112] The coordinate mapping method uses nearest neighbor interpolation or bilinear interpolation.

[0113] Nearest neighbor interpolation maps a pixel in the target mask to the infrared image coordinate system and takes the nearest integer coordinate pixel value as the mapping result. This method is computationally simple but may introduce spatial quantization errors.

[0114] Bilinear interpolation uses a weighted average of the gray values ​​of the four nearest pixels around the mapped point, which provides higher accuracy but doubles the computational cost.

[0115] The mapping error is controlled within 0.5 pixels to meet the accuracy requirements of pixel-level fusion.

[0116] In summary, step S6 completes the generation of the binary mask from the visible light contour and the mapping of the mask to the infrared image coordinate system. The generated binary target mask after coordinate mapping will serve as the spatial basis for constraining the extraction of temperature information from the infrared thermal image in step S7.

[0117] Step S7: The target mask is fused with the infrared thermal image. The visible light imaging unit and the infrared thermal imaging unit share the same front-end incident aperture and a unified optical reference, thus ensuring spatial consistency between the visible light image and the infrared thermal image. The ore fragment region in the binary target mask generated in step S6 is fused with the corresponding temperature information in the infrared thermal image, specifically as follows.

[0118] Step S701: Perform mask constraint fusion, using the binary target mask as a spatial constraint, and perform pixel-by-pixel masking operations on the infrared thermal image. The masking operations employ logical multiplication or pixel-by-pixel multiplication.

[0119] Let the infrared thermal image be The binary target mask is Then merge the images The calculation formula is: in: Represents the fused image in pixel coordinates The pixel value at that location, expressed as a temperature value or grayscale level; Indicates the pixel coordinates of the infrared thermal image The original pixel value at that location, expressed as a temperature value or grayscale level; This indicates the binary target mask at pixel coordinates. The mask value at the location is in gray levels and can be either 0 or 255. and , , are the x and y coordinates of the pixel, respectively, in pixels; 255 is the standard gray value of the foreground region in the binary target mask, dimensionless.

[0120] When the pixel value of the binary target mask When equal to 255, the merged image pixel value Equal to the original pixel value of the infrared thermal image ;when When equal to 0, It equals 0.

[0121] Step S702: Obtain the fusion result. After the above masking operation, the fused image is obtained. Only temperature information corresponding to the actual geometric region of the ore bulk in the infrared thermal image is retained. Conveyor belts, chutes, environmental background, radiation interference areas, and other non-target areas outside the target region in the fused image are masked, and the pixel values ​​of these areas are set to 0. Temperature data of the ore bulk region is fully preserved, while temperature data of the background region is set to zero.

[0122] Step S7 completes the spatial constraint fusion of the visible light target mask with the infrared thermal image. Through mask constraint, the interference of background thermal radiation on the subsequent extraction of ore temperature features is eliminated, providing clean input data for the accurate extraction of the temperature features of each ore particle in step S8.

[0123] Step S8: Temperature feature extraction based on the fused image. The fused image obtained in step S7... In this process, temperature information is only retained within the ore particle area defined by the binary target mask, while the pixel values ​​of the background area are set to zero. Step S8 extracts temperature features for each individual ore particle for subsequent waste mineral identification. The temperature features are divided into four categories: statistical features, distribution features, hot spot features, and regional features, each containing several specific indicators. All features are calculated independently for each individual ore particle; the background area is not included in any statistics. The specific implementation is carried out according to the following steps.

[0124] Step S801: Extract statistical features, including average temperature, maximum temperature, minimum temperature, temperature variance, and temperature standard deviation.

[0125] Average temperature The calculation formula is: in, This represents the set of pixels corresponding to the mineral particles in the fused image, in pixels. for The total number of pixels, dimensionless; To merge images at pixel coordinates The pixel value at that location, in degrees Celsius or Kelvin; and These are the x and y coordinates of a pixel, respectively, in pixels. The average temperature reflects the overall thermal state of the ore particles.

[0126] highest temperature Defined as This refers to the maximum temperature value of all pixels within the ore region, reflecting the peak thermal response of the ore particles.

[0127] minimum temperature Defined as This refers to the minimum temperature value of all pixels within the ore region, reflecting the edge cooling effect or low-temperature area of ​​the ore particles.

[0128] Temperature variance The calculation formula is: in, The above average temperature is expressed in degrees Celsius or Kelvin. Temperature variance reflects the uniformity of temperature distribution on the surface of ore particles.

[0129] Temperature standard deviation This is the square root of the temperature variance, i.e. The dimensions of the temperature standard deviation are the same as those of temperature.

[0130] Step S802: Extract distribution features, including the temperature gradient. The temperature gradient is defined as the partial derivative of temperature with respect to spatial coordinates. For discrete images, pixels... Temperature gradient magnitude at The central difference approximation is used for calculation: in, , , , These are the temperature values ​​of adjacent pixels, in degrees Celsius or Kelvin. The unit is Celsius or Kelvin per pixel.

[0131] For each ore particle, calculate the following statistics for its surface temperature gradient: average temperature gradient. Maximum temperature gradient and temperature gradient variance Temperature gradients reflect the steepness of the surface temperature of ore particles; hot spot regions typically exhibit larger temperature gradients.

[0132] Step S803: Extract hot spot features. A hot spot refers to a localized area on the surface of an ore particle where the temperature is higher than the surrounding area. First, a temperature threshold is set. The calculation formula is: in, The average temperature calculated in step S801, in degrees Celsius or Kelvin; The standard deviation of temperature is expressed in degrees Celsius or Kelvin; the coefficient 2 is dimensionless.

[0133] Temperature value greater than The set of pixels constitutes the hot spot region, denoted as . .

[0134] Hot spot area The calculation formula is: in, Hot spot region The number of pixels in the image, dimensionless; The spatial area corresponding to a single pixel is measured in square millimeters or square centimeters and is determined by the spatial resolution of the imaging device.

[0135] hot spot location Defined as the centroid coordinates of the hot spot region, the calculation formula is: in, and These are pixel coordinates, in pixels. The hotspot location reflects the concentrated absorption location of microwave energy within the mineral particles.

[0136] Step S804: Extract regional features, including the temperature of the central region, the temperature of the edge region, and the temperature difference between the center and the edge.

[0137] The central region is defined as follows: The smallest circumcircle of the ore particle's outline is taken, and the area inside this smallest circumcircle is taken as the central region. The ratio of the central region's area to the total area of ​​the ore particle is set as follows: , The value range is from 0.3 to 0.5. Dimensionless. In practice, first calculate the center and radius of the smallest circumcircle of the ore particle profile, then multiply by the same center and radius. The central region is obtained as a circle.

[0138] The edge region is defined as the annular region remaining after subtracting the central region from the ore particle outline.

[0139] Central area temperature This is the arithmetic mean of the temperature values ​​of all pixels within the central region. The edge region temperature... It is the arithmetic mean of the temperature values ​​of all pixels within the edge region.

[0140] Center-edge temperature difference The calculation formula is: in, and The units are all in degrees Celsius or Kelvin. The center-edge temperature difference reflects the degree of uneven heat conduction within the ore particles.

[0141] Step S805: Store the temperature feature vector. The temperature characteristics of each ore particle are stored in the form of a feature vector. Represented as: Feature vector The length of all feature vectors is determined based on the actual feature category selected. The set is used as the input for step S9, the waste mining discrimination.

[0142] In summary, step S8 completes the extraction of multi-dimensional temperature features for each ore particle. The extracted features cover temperature amplitude, distribution uniformity, local hot spots, and center-edge differences, which can characterize the thermal response characteristics of ore particles after microwave heating, providing a quantitative basis for subsequent mineral waste identification based on thresholds or machine learning.

[0143] For step S9, based on the temperature feature vector extracted in step S8, the mineral waste attribute is determined for each ore particle. The determination method uses either a temperature threshold method or a machine learning classifier. The two methods can be selected according to the actual application scenario. The machine learning classifier needs to be pre-trained and implemented according to the following steps.

[0144] Step S901: Temperature threshold method discrimination. The basic principle of the temperature threshold method is: based on laboratory calibration results, determine the difference threshold of various temperature characteristics between ore and waste rock. When a certain temperature characteristic of a certain ore particle exceeds the preset threshold, the ore particle is determined to be of the waste rock attribute.

[0145] Setting the threshold requires experimental data analysis. The specific procedure is as follows: collect samples of the target mineral and waste rock, obtain the temperature characteristics of the samples under the same microwave heating conditions, statistically analyze the distribution range of various temperature characteristics of the ore samples and waste rock samples, and determine the critical value that can effectively distinguish between the two as the discrimination threshold.

[0146] For example, if the dielectric loss of the target ore is greater than that of the waste rock, then the temperature rise of the target ore after microwave heating will be greater than that of the waste rock. In this case, an average temperature threshold is set. For the ore particles to be identified, calculate their average temperature. ,like If it is, it is determined to be the target ore; otherwise, it is determined to be waste rock.

[0147] The temperature threshold method is simple to calculate and is suitable for scenarios where there are significant differences in temperature characteristics between ore and waste rock.

[0148] Step S902: Machine learning classifier discrimination. The machine learning classifier employs at least one algorithm selected from support vector machines, random forests, or convolutional neural networks. The input to the classifier is the temperature feature vector extracted in step S8. The output is a classification label for the ore particles, and the label value can be either the target ore or waste rock.

[0149] The training process for a convolutional neural network classifier is as follows: The training dataset was constructed as follows: a large number of ore particle samples were collected, and each sample was microwave-heated. Then, the temperature feature vector was extracted according to step S8. At the same time, the true label of the sample should be marked manually or based on known grade analysis results. , Indicates the target ore. Represents waste rock. The temperature feature vector... As input features, labels As a supervisory signal, a set of training samples is formed. The above process is repeated to collect a sufficient number of samples to form a training dataset.

[0150] The structure of a convolutional neural network is designed as follows: the input layer receives temperature feature vectors. It contains several fully connected layers and activation function layers. The output layer uses the Softmax or Sigmoid function to output the probability value of belonging to the target ore. .

[0151] The loss function uses binary cross-entropy loss. The calculation formula is as follows: in, The number of samples taken in one training session is dimensionless. For the first The true label of each sample, taking the value 0 or 1, is dimensionless; The first output of the convolutional neural network The predicted probability that a sample belongs to the target ore, with a value ranging from 0 to 1, and is dimensionless.

[0152] The training objective is to adjust the weight parameters of the convolutional neural network using the backpropagation algorithm, so that the loss function... Minimize the value of . After training, the convolutional neural network can perform training based on the input temperature feature vector. Automatically output classification results.

[0153] The training processes for Support Vector Machines (SVMs) and Random Forest classifiers are similar, both requiring labeled training datasets. SVMs achieve classification by finding the optimal hyperplane, while Random Forests achieve classification by ensembling multiple decision trees. During training, the classifier learns the mapping relationship from the temperature feature space to the mine waste label space.

[0154] Step S903: Output and process the discrimination results. The results of the mineral waste discrimination are output in the form of labels for each ore particle in each frame of the image. The labels include three types: target ore, waste rock, and pending. The classification accuracy meets the predetermined performance requirements, usually not less than 95%.

[0155] Samples with low classification confidence are marked as pending. The handling of pending samples involves manual review or secondary discrimination using other auxiliary methods. Classification confidence is calculated as follows: for the temperature threshold method, confidence is low when the temperature feature value is close to the threshold; for convolutional neural networks, confidence is low when the output predicted probability... The confidence level is low when it is close to 0.5. The specific confidence threshold can be set according to the actual application.

[0156] Step S9 determines the mineral waste attributes of each ore particle. The determination result assigns a target ore, waste rock, or pending label to each ore particle. These labels will be used for target location tracking and sorting execution control in step S10 to achieve precise blowing and removal of waste rock or recovery of target ore.

[0157] Finally, step S10 is the target position tracking and sorting execution control step. Combining the conveyor belt speed, the distance between the detection position and the sorting actuator, and the actuator's response time, the time it takes for each ore particle to reach the sorting position is calculated. When a particle identified as the target ore or waste rock reaches the blowing area, the image acquisition, processing, and sorting control system sends a control command to drive the pneumatic discharge device to achieve targeted blowing of the corresponding particle, as detailed below.

[0158] Step S1001, target position tracking. Target position tracking uses Kalman filtering or the Hungarian algorithm to match the position of ore particles with the detection frame.

[0159] Kalman filtering is a recursive estimation algorithm based on the state space of a linear system, suitable for smoothing and predicting the trajectory of moving targets. The basic steps of the Kalman filtering algorithm are: state prediction and state update. State prediction predicts the current state estimate based on the state estimate from the previous time step and the target motion model; state update updates the state estimate based on the current observations. The target motion model assumes that the ore particles move at a uniform linear velocity on the conveyor belt, and the state vector includes the position and velocity of the ore particles.

[0160] The Hungarian algorithm is a combinatorial optimization algorithm for solving maximum weighted matching in bipartite graphs. In target location tracking, it is used to optimally match the detected ore particle positions with the tracked trajectories, solving the data association problem. Specifically, it constructs a cost matrix between the detected position and the existing trajectory, where the cost is the distance between the positions. Then, it applies the Hungarian algorithm to find the optimal match, assigning the detected position to the nearest trajectory or a newly created trajectory.

[0161] Step S1002: Arrival Time Calculation and Spatial Coordinate Prediction. The arrival time of the target at the sorting position is calculated based on the conveyor belt speed. Let the distance between the detection position and the sorting actuator be... The unit is meters; the conveyor belt speed is... The unit is meters per second. This refers to the time required for an ore particle to move from the detection position to the sorting position. The calculation formula is: in, The measurement is the straight-line distance between the detection location and the sorting actuator, expressed in meters. The speed of the conveyor belt is expressed in meters per second. The unit is seconds.

[0162] Based on the position and direction of movement of the ore particles at the detection location, combined with Predict the spatial coordinates of the ore particle when it reaches the sorting actuator. The prediction assumes that the ore particle is moving at a uniform linear speed on the conveyor belt, with the direction of motion along the conveyor belt's forward direction.

[0163] The response time of the sorting execution control includes the sensor response time, controller processing time, and pneumatic actuator action time. The pneumatic actuator response time meets the preset response performance requirements, typically between 10 and 50 milliseconds. The positioning accuracy meets the preset positioning performance requirements, typically within ±5 millimeter.

[0164] Step S1003: Sorting execution control. The pneumatic discharge device drives the pneumatic blowing mechanism according to the sorting control command. The pneumatic blowing mechanism includes multiple blowing units arranged along the conveying direction, each blowing unit corresponding to a spatial position. The controller in the image acquisition, processing, and sorting control system triggers the corresponding blowing unit to operate based on the ore particle arrival position predicted in step S1002.

[0165] The blowing unit is driven by a solenoid valve controlling the movement of a cylinder piston, which ejects compressed air at high speed from the nozzle, forming a high-speed air jet. The speed of the air jet is typically 50 to 150 meters per second, and the blowing duration is 20 to 100 milliseconds. The high-speed air jet acts on the ore particles, using the impact force of the airflow to change the direction of the ore particles, blowing the target ore into the collection tank or blowing waste rock into the waste tank.

[0166] Step S1004, Description of the Image Acquisition, Processing, and Sorting Control System Architecture: The hardware platform of the image acquisition, processing, and sorting control system adopts a high-performance industrial computer or an embedded GPU computing platform. The system software architecture includes an image acquisition module, an image processing module, a feature extraction module, a classification and discrimination module, and a motion control module.

[0167] The image acquisition module is responsible for receiving image data from visible light and infrared cameras, performing format conversion and timestamp synchronization. The image processing module implements functions such as filtering, edge detection, morphological processing, mask generation, and image fusion.

[0168] The feature extraction module extracts the temperature feature vectors of each ore particle from the fused image. The classification and discrimination module determines the mineral waste attributes based on the temperature feature vectors. The motion control module generates sorting control commands based on the ore particle positions and the discrimination results, and sends them to the pneumatic discharge device.

[0169] In summary, step S10 completes the entire process from target location tracking to sorting execution control. Through time prediction and spatial positioning, targeted injection of each ore particle is achieved, separating the target ore from waste rock and completing the pre-selection and waste disposal operation.

[0170] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention. Therefore, the embodiments should be regarded as exemplary and non-limiting in all respects.

[0171] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A microwave sorting method for ores based on dual-mode collaborative imaging of visible and infrared light, characterized in that, Includes the following steps: The ore to be sorted is fed and conveyed in a single layer to form a single-layer discrete particle flow. The monolayer discrete particle flow is heated by microwave irradiation using a microwave heating device, so that different mineral components produce differentiated thermal responses based on differences in dielectric properties. Visible light and infrared thermal images of heated ore bulk material were simultaneously acquired using a common aperture dual-mode imaging device. The visible light image is processed to extract the contour boundary of the ore particles, and a binary target mask is generated based on the contour boundary. The binary target mask is used to characterize the actual spatial area occupied by the ore particles in the visible light image. The binary target mask is fused with the infrared thermal image. The infrared thermal image is spatially constrained by the binary target mask, and only the temperature information in the infrared thermal image corresponding to the area occupied by the real space is retained. Based on the temperature information retained after constraints, the temperature characteristics of each ore particle are extracted, and the mineral waste attributes of each ore particle are determined according to the temperature characteristics. Based on the results of the mineral waste property identification, the location tracking and sorting of each ore particle are controlled.

2. The microwave sorting method for ores based on dual-mode collaborative imaging of visible and infrared light according to claim 1, characterized in that, The common aperture dual-mode imaging device includes a visible light imaging unit, an infrared thermal imaging unit, a common aperture front-end optical system, a beam splitting element, and a synchronous triggering unit. The specific steps for synchronous data acquisition include: After detecting that the ore particles have entered the imaging area, the synchronous triggering unit simultaneously sends trigger pulses to the visible light imaging unit and the infrared thermal imaging unit, with the time jitter controlled within 100 microseconds. After receiving the trigger pulse, the visible light imaging unit and the infrared thermal imaging unit synchronously start exposure and readout, respectively acquiring a visible light image and an infrared thermal image; The visible light image and the infrared thermal image are timestamped and their quality is verified to ensure that the spatial displacement corresponding to the acquisition time difference is less than the spatial resolution corresponding to the detector pixel.

3. The microwave sorting method for ores based on dual-mode collaborative imaging of visible and infrared light according to claim 1, characterized in that, The specific steps for extracting the contour boundaries of ore particles from a visible light image include: The visible light image is filtered to suppress noise, and the filtering process uses Gaussian filtering or median filtering. Edge detection is performed on the filtered image to extract the preliminary contours of the ore particles. The edge detection uses the Canny algorithm, Sobel algorithm, or Prewitt algorithm. Morphological closing operations are performed on the edge-detected image to fill the contour gaps. The morphological closing operation is composed of dilation and erosion operations combined in the order of dilation followed by erosion. Connectivity analysis is performed on the image after morphological closing operation to remove small noise areas and mark each independent ore particle. The pixel set of each independent connected region constitutes the preliminary outline of the ore particle.

4. The microwave sorting method for ores based on dual-mode collaborative imaging of visible and infrared light according to claim 3, characterized in that, After labeling the individual ore particles, the process also includes a step to separate the aggregated particles: When the outlines of multiple ore particles merge into a connected region, the watershed algorithm is used to segment the contiguous region. The watershed algorithm processing includes: performing distance transformation on the binary contour image, calculating the distance from each background pixel to the nearest particle boundary, and generating a distance map; The distance map is thresholded, and local maxima are selected as seed points. Each seed point corresponds to the center position of a ore particle. Starting from the seed point, flooding is carried out towards the surrounding area, forming a dividing line where floods from different seed points meet, thus dividing the adhered particle area into multiple independent particle profiles.

5. The microwave sorting method for ores based on dual-mode collaborative imaging of visible and infrared light according to claim 1, characterized in that, The specific steps for generating a binary target mask based on the contour boundary include: The extracted or separated particle contour image is converted into a binary image. The inside of the contour line is filled with the foreground color and the outside of the contour line is filled with the background color using the scan line filling method or seed filling method, generating a binary target mask with the same resolution as the original visible light image, where the foreground region corresponds to the ore bulk region. The binary target mask is mapped to the pixel coordinate system of the infrared thermal image using nearest neighbor interpolation or bilinear interpolation methods, with the mapping error controlled within 0.5 pixels.

6. The microwave sorting method for ores based on dual-mode collaborative imaging of visible and infrared light according to claim 1, characterized in that, The specific steps for fusing the binary target mask with the infrared thermal image include: Using the binary target mask as a spatial constraint, pixel-by-pixel masking operations are performed on the infrared thermal image. The masking operations are performed using logical multiplication or pixel-by-pixel multiplication. Let the infrared thermal image be The binary target mask is Then merge the images The calculation formula is: in, Represents the fused image in pixel coordinates Pixel value at that location, Indicates the pixel coordinates of the infrared thermal image The original pixel value at that location, This indicates the binary target mask at pixel coordinates. Mask value at that location, and These are the x and y coordinates of the pixel, respectively. This represents the standard grayscale value of the foreground region in a binary target mask. when When it equals 255, equal ;when When equal to 0, Equal to 0, so that the fused image retains only the temperature information in the infrared thermal image that corresponds to the actual geometric region of the ore bulk.

7. The microwave sorting method for ores based on dual-mode collaborative imaging of visible and infrared light according to claim 1, characterized in that, The specific steps for extracting the temperature characteristics of each ore particle include: Extract statistical features, including average temperature, maximum temperature, minimum temperature, temperature variance, and temperature standard deviation; Extract distribution features, including temperature gradients, and calculate the average temperature gradient, maximum temperature gradient, and temperature gradient variance; Extract hot spot features, which include hot spot area and hot spot location, wherein a hot spot is defined as a set of pixels whose temperature value is greater than the sum of the average temperature and twice the temperature standard deviation. Extract regional features, which include the temperature of the central region, the temperature of the edge region, and the temperature difference between the center and the edge. The central region is the region inside the smallest outer circle of the ore particle profile, and the edge region is the annular region of the ore particle profile minus the central region. The temperature characteristics of each ore particle are stored as a feature vector and used as input for ore waste discrimination.

8. The microwave sorting method for ores based on dual-mode collaborative imaging of visible and infrared light according to claim 1, characterized in that, The property identification of mine waste adopts the temperature threshold method or a machine learning classifier, wherein the machine learning classifier adopts at least one algorithm among support vector machine, random forest or convolutional neural network. When using a machine learning classifier, a training dataset is pre-constructed. Each sample in the training dataset includes a temperature feature vector and a corresponding mine waste label. The classifier is trained with the objective of minimizing the binary cross-entropy loss function, which is calculated using the following formula: in, This represents the value of the loss function. The number of samples taken in one training session. For the first The true label of each sample The output of the machine learning classifier is the first The predicted probability that a sample belongs to the target ore; The results of the mineral waste identification are output in the form of a label for each ore particle. The label includes target ore, waste rock, and undetermined samples. Undetermined samples are manually reviewed or identified a second time using auxiliary methods.

9. A microwave ore sorting system based on dual-mode cooperative imaging of visible and infrared light, used to execute the microwave ore sorting method based on dual-mode cooperative imaging of visible and infrared light as described in any one of claims 1 to 8, characterized in that, include: The ore feeding and single-layer conveying subsystem is used to process the ore to be sorted into a single-layer discrete particle flow and convey it. A microwave heating device is installed on the conveying path of the ore feeding and single-layer conveying subsystem to irradiate and heat the single-layer discrete particle flow with microwaves, so that different mineral components produce differentiated thermal responses. A common aperture dual-mode imaging device is set in the downstream detection area of ​​the microwave heating device to simultaneously acquire visible light images and infrared thermal images of the heated ore bulk. The common aperture dual-mode imaging device includes a visible light imaging unit, an infrared thermal imaging unit, a common aperture front-end optical system, a beam splitting element, and a synchronous triggering unit. The image acquisition, processing and sorting control system is connected to the common aperture dual-mode imaging device and the ore feeding and single-layer conveying subsystem, respectively, and is used to receive the visible light image and the infrared thermal image, and perform image processing, boundary extraction, mask generation, image fusion, temperature feature extraction and mineral waste attribute discrimination; A pneumatic discharge device is located downstream of the common aperture dual-mode imaging device and connected to the image acquisition, processing and sorting control system. It is used to perform point-by-point blowing separation of each ore particle based on the mineral waste property discrimination result.

10. An image acquisition, processing, and sorting control system for microwave sorting of ores based on dual-mode collaborative imaging of visible and infrared light, characterized in that, include: The image acquisition module is used to simultaneously receive visible light images and infrared thermal images from the common aperture dual-mode imaging device, and to perform format conversion and timestamp synchronization. The image processing module is used to perform filtering, edge detection, morphological closing operation and connected component analysis on the visible light image, extract the contour boundary of the ore aggregate, and use the watershed algorithm to separate the adhering particles to generate a binary target mask. The image fusion module is used to perform pixel-by-pixel masking operations on the infrared thermal image using the binary target mask as a spatial constraint condition, and generate a fused image that retains only the temperature information of the ore bulk region. The feature extraction module is used to extract the temperature features of each ore particle from the fused image, the temperature features including statistical features, distribution features, hot spot features and regional features; The classification and discrimination module is used to classify the mineral waste attributes of each ore particle based on the temperature characteristics, using a temperature threshold method or a machine learning classifier, and output a discrimination label. The motion control module is used to calculate the time it takes for each ore particle to reach the sorting position based on the identification label and tracking position of the ore particles, combined with the conveyor belt running speed, the distance between the detection position and the sorting actuator, and the response time of the actuator, and to generate sorting control commands to send to the pneumatic discharge device.

Images