Automatic ore fine sorting method and system based on XRT and hyperspectrum

Abstract

The invention discloses an automatic fine ore sorting method and system based on XRT (X ray tomography) and hyperspectrum, which are used for realizing comprehensive fusion of ore surface and internal information by combining X-ray tomography and hyperspectral imaging technologies so as to improve the recognition precision and the sorting efficiency. XRT is used for extracting the internal structure and density characteristics of the ore, hyperspectral imaging is used for identifying the mineral composition characteristics of the ore surface, and information complementation is achieved. The system adopts a multi-modal convolutional neural network to perform independent feature extraction on XRT and hyperspectral data, and the extracted features are fused through a deep residual network. And the fused features are processed by a cross-modal attention mechanism to ensure that key features of different modals are fully integrated and optimized, and finally high-precision ore classification and sorting are realized. According to the method, the limitation of a single data source on feature extraction is overcome, the ore recognition and sorting precision is greatly improved, and the refined and intelligent development of ore sorting is promoted.

Description

Technical Field

[0001] The present invention relates to the field of intelligent ore sorting, and specifically to an automated fine ore sorting method and system based on XRT and hyperspectral. Background Art

[0002] Single-modality ore detection technology, such as only using visible light imaging methods or relying on X-ray technology to obtain internal information of the ore. The disadvantage of these technologies is that the data features of a single mode are limited and cannot fully and accurately describe the internal structure and surface composition of the ore. For example, single visible light imaging is easily disturbed by factors such as ore surface reflection and lighting conditions, resulting in unclear boundaries. Although simple X-ray tomography can present the internal density characteristics of the ore, it is difficult to reflect its chemical composition and surface composition characteristics, especially when there are multiple mineral components in the ore, a single data source is difficult to achieve fine sorting. In addition, these prior art methods also have problems such as information redundancy, single features, and susceptibility to interference from external environmental factors, which limits the accuracy of ore identification. Summary of the invention

[0003] The present invention provides an automated fine ore sorting method based on XRT and hyperspectral imaging, which aims to solve the deficiencies in existing ore sorting technology. By combining X-ray transmission imaging with hyperspectral imaging technology, a comprehensive fusion of ore surface and internal information is achieved, thereby overcoming the problems of incomplete features and insufficient information caused by a single data source in traditional ore sorting technology.

[0004] The objective of the present invention is achieved through the following technical solutions:

[0005] According to a first aspect of an embodiment of the present invention, there is provided an automated fine sorting method for ore based on XRT and hyperspectral, which specifically comprises the following steps:

[0006] S1: Ore data collection: collect multimodal data of ore samples through XRT and hyperspectral imaging equipment to obtain X-ray transmission imaging data and hyperspectral images of ore;

[0007] S2: Ore contour extraction, by combining X-ray transmission imaging technology and deep learning models, high-precision extraction of ore contours is achieved;

[0008] S3: Feature fusion and optimization. In the processing of hyperspectral data, an independent convolutional neural network is also used to extract the spectral features of the ore surface. In the process of extracting hyperspectral features, the convolution layer can gradually optimize and extract high-dimensional features through the learning mechanism of the residual block, and the existence of the jump connection allows the original input features to be retained, avoiding information attenuation caused by gradual feature loss in deep convolution.

[0009] S4: Classification prediction of deep fusion features. After completing the independent feature extraction of the two modalities, the data is sent to the cross-modal attention mechanism module for deep fusion of XRT and hyperspectral features. The fused features are further extracted through a deep residual network to extract high-order features.

[0010] S5: Ore category identification and intelligent sorting.

[0011] In one embodiment of the present invention, in step S1, XRT data can reveal structural information such as internal density distribution, fractures and voids of the ore, while hyperspectral images provide spectral characteristics of the ore in different bands to reveal its surface chemical composition and mineral characteristics. The comprehensive use of the two modes provides a multi-level and rich data foundation for the system, ensuring that effective ore information can be obtained in different environments. Specifically, the following sub-steps are included:

[0012] S11: For XRT data, preliminary feature extraction is first performed through multiple convolutional layers of the convolutional neural network. XRT data mainly reflects the internal density distribution and structural information of the ore. Through layer-by-layer convolution, CNN can extract the basic morphology and edge features of the ore from a lower level;

[0013] S12: After multiple convolution operations, the feature map is further used to extract deep features through residual blocks. The residual mechanism makes deep feature extraction no longer rely on relearning all features, but instead performs incremental learning based on the input features. This further enhances the network's ability to extract complex features while retaining the original information.

[0014] In one embodiment of the present invention, the X-ray transmission imaging in step S2 is used to obtain a projection image of the ore. After the X-ray image is obtained, the image undergoes a series of preprocessing steps to improve data quality and provide optimized input for the subsequent deep learning algorithm, specifically including the following sub-steps:

[0015] S21: After using X-ray transmission imaging technology to obtain the ore projection image, due to the influence of factors such as the external environment and imaging equipment, noise interference will inevitably appear in the image. In order to eliminate these noises and make the image clearer, Gaussian filtering and median filtering are used. Gaussian filtering can effectively remove high-frequency noise while maintaining the continuity of the overall structure by smoothing the image. Combining these two filtering methods, the effective denoising of the ore image is finally achieved to ensure the integrity and clarity of the edge information;

[0016] S22: After noise removal, in order to enhance the contrast of the image, the histogram equalization technology is used to adjust the distribution of pixel values ​​in the image so that the difference between the ore and the background is more significant, thereby making the boundary of the ore clearer. For some images with large contrast changes in local areas, adaptive histogram equalization (CLAHE) is further introduced. By dividing the image into blocks and performing equalization adjustment on the contrast of different areas, the overall contrast of the image is enhanced and the noise problem introduced by over-enhancement is avoided.

[0017] S23: In the process of image preprocessing, bilateral filtering is used to retain edge information while removing noise. By combining the spatial distance and grayscale difference between pixels for smoothing, similar areas are smoothed while the edge parts remain clear. The introduction of bilateral filtering makes the ore edge more prominent in the subsequent extraction process, which helps the model to accurately locate the boundary of the ore.

[0018] S24: After the image preprocessing is completed, the image needs to be normalized to standardize the image pixel values ​​to a uniform range (for example, between 0 and 1). Normalization can eliminate the impact of brightness or intensity differences between different samples, so that the model can learn more essential features under the same input standard, thereby improving the robustness and adaptability of the overall model.

[0019] S25: During the model training process, a data augmentation module was added to help the model better cope with the various forms and complexities of ores by artificially expanding the diversity of the training dataset.

[0020] In one embodiment of the present invention, in step S3, in terms of processing hyperspectral data, an independent convolutional neural network is also used to extract the spectral characteristics of the ore surface. The hyperspectral data mainly describes the spectral response of the ore in different bands, and characterizes the surface chemical composition of the ore through spectral information of up to hundreds of bands. The convolutional neural network processes these spectral images through layer-by-layer convolution. The low-level convolutional layer is used to extract the basic response characteristics of the ore in each band. As the network depth increases, the high-level convolutional layer can gradually learn more complex spectral features.

[0021] In one embodiment of the present invention, in step S4, during the extraction of hyperspectral features, the convolution layer can gradually optimize and extract high-dimensional features through the learning mechanism of the residual block, and the existence of the jump connection allows the original input features to be retained, avoiding information attenuation caused by gradual feature loss in deep convolutions. The multi-layer convolution in the residual block introduces nonlinear transformations through nonlinear activation functions, so that the network can learn more complex spectral patterns from the hyperspectral images of the ore, which reflect the changes in the surface composition and internal correlations of the ore.

[0022] In one embodiment of the present invention, in step S5, the fused features are processed by the deep learning network and then sent to the prediction module. The features of each modality are independently predicted by the prediction head to obtain the preliminary judgment results of the ore category. The independent processing in the prediction stage ensures that the features of each modality are fully reflected in the decision. Subsequently, the prediction results of each modality are finally integrated using an adaptive weighted fusion mechanism. The mechanism dynamically adjusts the influence weight of each modality in the final decision according to the confidence of each modality prediction. The purpose of the adaptive weighting mechanism is to optimize the prediction results so that the model can dynamically adjust the influence of different modal features on the final prediction according to their contribution, thereby ensuring the accuracy and reliability of the prediction. This strategy achieves the optimal use of multimodal data and generates more accurate classification results.

[0023] After the final prediction is completed, the ore category and composition information is passed to the robotic arm sorting module. The sorting system combines visual sensors and control systems to accurately identify the category, location and shape of the ore, and drives the robotic arm to perform grasping and sorting operations. Based on the prediction results of the deep learning model, the robotic arm automatically adjusts its grasping posture and path to adapt to different types and sizes of ores, achieving high-precision grasping and sorting. The robotic arm has multi-degree-of-freedom movement capabilities and can adapt to complex ore shapes and layouts to ensure the accuracy and continuity of sorting operations.

[0024] According to the second aspect of the embodiment of the present invention, there is also provided an automated fine sorting system for ore based on XRT and hyperspectral, which accurately identifies the type and location of ore through an intelligent robotic arm combined with a visual sensor, and realizes automatic grabbing and sorting operations of ore, including a central control system, a robotic arm sorting system, and a communication network. The ore is stably transported from the input end to each functional module by a conveyor belt. The conveyor belt is responsible for moving the ore smoothly along the specified path, and through the entire device, it ensures that the ore can go through a series of complex detection and sorting processes.

[0025] In one embodiment of the present invention, at the initial stage of the conveyor belt, the ore will enter the detection area, where an X-ray source and a hyperspectral camera are arranged to scan the internal structure and surface elements of the ore respectively. The X-ray tomography technology is used to obtain the internal density characteristics and fracture gaps of the ore. By penetrating the ore and detecting the absorption changes of the X-rays, the internal characteristics of the ore can be accurately presented. At the same time, the hyperspectral camera performs multi-band spectral imaging on the surface of the ore to provide detailed information about the surface composition of the ore. By combining these two scanning methods, the system can capture comprehensive data on the internal physical characteristics and surface chemical composition of the ore. In order to ensure the accuracy of the detection process, multiple sensors are installed on the conveyor belt, which are responsible for monitoring the position and state of the ore and controlling the start and stop of the scanner. Through precise position perception, the ore can trigger the scanning of the X-ray source and the hyperspectral camera when it reaches the detection area, ensuring that each piece of ore can be completely collected. These real-time sensor data are directly connected to the central control system to provide dynamic scheduling and control support for the entire detection process.

[0026] In one embodiment of the present invention, during the detection process, the ore passes through a protective cabin arranged on the conveyor belt to ensure the safety of the X-ray during use. The protective cabin is a shielding structure designed around the scanning area to ensure that the rays will not leak into the external environment and avoid potential impact on operators and equipment. At the same time, during the transportation and scanning of the ore, the system will generate a large amount of dust, so the equipment is equipped with dust removal pipes and dust removal devices. These devices can effectively collect and remove dust emitted during the scanning and transportation of the ore, maintain the cleanliness of the equipment, and prevent dust from affecting the detection accuracy and mechanical operation. After the ore is scanned, the conveyor belt continues to send the ore to the robotic arm sorting system. The robotic arm obtains the previous X-ray and hyperspectral scanning data through the connection with the central control system, and classifies and sorts the ore based on these data. The operation of the robotic arm sorting system depends on the central control system and the communication network. The central control system receives data from sensors, X-ray scanning and hyperspectral cameras, and performs real-time analysis and judgment on the ore based on these data. Then, through the communication network, the central control system directs the robotic arm to perform sorting actions. After receiving the command, the robot arm can respond quickly and accurately, grab the ore and place it in the corresponding sorting area. Finally, the ore is divided into concentrate, middlings and tailings according to its characteristics. These ores enter different collection containers through their respective tracks to complete the entire sorting process. At the same time, the waste collection system in the system is used to collect and process ores that do not meet the requirements, ensure the effective use of valuable ore resources, and reduce the accumulation of waste and environmental pollution. The dust removal and waste management system ensures that the ore sorting process is not only accurate but also environmentally friendly and intelligent, thus providing strong hardware support for the efficient use of ore.

[0027] According to a third aspect of an embodiment of the present invention, there is also provided a computer-readable storage medium having computer executable instructions stored thereon, which, when executed by a processor, can implement the above-mentioned XRT and hyperspectral-based automated fine sorting method for ore.

[0028] Technical principle of the present invention: The present invention combines X-ray tomography (XRT) with hyperspectral imaging technology to achieve a comprehensive fusion of ore surface and internal information to improve the accuracy of identification and sorting efficiency. In this system, XRT is used to extract the internal structure and density characteristics of the ore, while hyperspectral imaging is used to identify the mineral composition characteristics of the ore surface, and the two information are complementary. The system uses a multimodal convolutional neural network (MM-CNN) to extract independent features from XRT and hyperspectral data, and fuses the extracted features through a deep residual network (ResNet). The fused features are processed by a cross-modal attention mechanism to ensure that the key features from different modalities are fully integrated and optimized, and finally achieve high-precision ore classification and sorting. In view of the complex structure and diverse components of the ore, a deep convolutional neural network (such as ResNet) is used for feature extraction and learning to achieve high-precision identification of ore components and boundaries, thereby further improving the accuracy and efficiency of ore sorting and reducing errors and deviations caused by traditional methods. Through the multimodal deep fusion method, the present invention can accurately determine the type and value of ore, support intelligent and automated sorting processes, and ultimately improve the rational utilization of mineral resources and sorting efficiency.

[0029] Beneficial effects: The technical solution of the present invention achieves a comprehensive characterization of the internal structure and surface composition of the ore by comprehensively applying two modal data, X-ray tomography (XRT) and hyperspectral imaging, and overcomes the limitations of a single data source in feature extraction. XRT technology provides the internal density and structural characteristics of the ore, while hyperspectral imaging reveals the surface mineral composition of the ore. Through the multimodal convolutional neural network (MM-CNN), the system performs independent feature extraction and cross-modal feature fusion on these two modal data, greatly improving the accuracy of ore identification and sorting.

[0030] The deep residual network (ResNet) is used to extract features from the two modal data, so that the internal physical properties and surface chemical characteristics of the ore are more comprehensively learned and expressed. By introducing the cross-modal attention mechanism, the system can achieve effective fusion of different modal features, and maximize the retention and utilization of the complementary characteristics of each modality. This deep fusion enables the ore features to achieve a better balance between information diversity and expression accuracy, and improves the robustness and reliability of the system in ore identification.

[0031] In practical applications, this scheme uses an adaptive weighted fusion mechanism to dynamically adjust the weight of features in the final decision according to the confidence of different modal data, which significantly improves the accuracy of ore identification and sorting. Theoretical analysis and experimental results show that multimodal information fusion can effectively reduce the error in ore identification, improve the automation and intelligence level of the system, make ore classification and sorting more accurate, reduce the possibility of human intervention, and thus improve the rational development and efficient utilization of mineral resources.

[0032] Compared with the existing single-modal technology solutions, the present invention provides a more comprehensive and accurate means of ore feature extraction and sorting by deeply integrating multi-modal data. Especially when the ore morphology is complex and diverse, its classification accuracy and sorting efficiency are significantly improved, which helps to promote the refinement and intelligent development of ore sorting. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a schematic diagram of the overall process structure of the present invention.

[0034] Figure 2 Schematic diagram of the residual network structure in one embodiment of the present invention.

[0035] Figure 3 The figure is a schematic diagram of the process of the intelligent sorting system in one embodiment of the present invention.

[0036] Figure 4 It is a schematic diagram of the overall hardware structure of the present invention. DETAILED DESCRIPTION

[0037] The following is a detailed description of an embodiment of the present invention in conjunction with the accompanying drawings: This embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation method and a specific operation process are given, but the protection scope of the present invention is not limited to the following embodiments.

[0038] like Figure 1-4 As shown, an automated fine separation method for ore based on XRT and hyperspectral includes the following steps:

[0039] S1: Ore data collection: collect multimodal data of ore samples through XRT and hyperspectral imaging equipment to obtain X-ray transmission imaging data and hyperspectral images of ore;

[0040] S2: Ore contour extraction, by combining X-ray transmission imaging technology and deep learning models, high-precision extraction of ore contours is achieved;

[0041] S3: Feature fusion and optimization. In the processing of hyperspectral data, an independent convolutional neural network is also used to extract the spectral features of the ore surface. In the process of extracting hyperspectral features, the convolution layer can gradually optimize and extract high-dimensional features through the learning mechanism of the residual block, and the existence of the jump connection allows the original input features to be retained, avoiding information attenuation caused by gradual feature loss in deep convolution.

[0042] S4: Classification prediction of deep fusion features. After completing the independent feature extraction of the two modalities, the data is sent to the cross-modal attention mechanism module for deep fusion of XRT and hyperspectral features. The fused features are further extracted through a deep residual network to extract high-order features.

[0043] S5: Ore category identification and intelligent sorting.

[0044] In one embodiment of the present invention, in step S1, XRT data can reveal structural information such as internal density distribution, fractures and voids of the ore, while hyperspectral images provide spectral characteristics of the ore in different bands to reveal its surface chemical composition and mineral characteristics. The comprehensive use of the two modes provides a multi-level and rich data foundation for the system, ensuring that effective ore information can be obtained in different environments. Specifically, the following sub-steps are included:

[0045] S11: For XRT data, preliminary feature extraction is first performed through multiple convolutional layers of the convolutional neural network. XRT data mainly reflects the internal density distribution and structural information of the ore. Through layer-by-layer convolution, CNN can extract the basic morphology and edge features of the ore from a lower level;

[0046] S12: After multiple convolution operations, the feature map is further used to extract deep features through residual blocks. The residual mechanism makes deep feature extraction no longer rely on relearning all features, but instead performs incremental learning based on the input features. This further enhances the network's ability to extract complex features while retaining the original information.

[0047] In one embodiment of the present invention, the X-ray transmission imaging in step S2 is used to obtain a projection image of the ore. After the X-ray image is obtained, the image undergoes a series of preprocessing steps to improve data quality and provide optimized input for the subsequent deep learning algorithm, specifically including the following sub-steps:

[0048] S21: After using X-ray transmission imaging technology to obtain the ore projection image, due to the influence of factors such as the external environment and imaging equipment, noise interference will inevitably appear in the image. In order to eliminate these noises and make the image clearer, Gaussian filtering and median filtering are used. Gaussian filtering can effectively remove high-frequency noise while maintaining the continuity of the overall structure by smoothing the image. Combining these two filtering methods, the effective denoising of the ore image is finally achieved to ensure the integrity and clarity of the edge information;

[0049] S22: After noise removal, in order to enhance the contrast of the image, the histogram equalization technology is used to adjust the distribution of pixel values ​​in the image so that the difference between the ore and the background is more significant, thereby making the boundary of the ore clearer. For some images with large contrast changes in local areas, adaptive histogram equalization (CLAHE) is further introduced. By dividing the image into blocks and performing equalization adjustment on the contrast of different areas, the overall contrast of the image is enhanced and the noise problem introduced by over-enhancement is avoided.

[0050] S23: In the process of image preprocessing, bilateral filtering is used to retain edge information while removing noise. By combining the spatial distance and grayscale difference between pixels for smoothing, similar areas are smoothed while the edge parts remain clear. The introduction of bilateral filtering makes the ore edge more prominent in the subsequent extraction process, which helps the model to accurately locate the boundary of the ore.

[0051] S24: After the image preprocessing is completed, the image needs to be normalized to standardize the image pixel values ​​to a uniform range (for example, between 0 and 1). Normalization can eliminate the impact of brightness or intensity differences between different samples, so that the model can learn more essential features under the same input standard, thereby improving the robustness and adaptability of the overall model.

[0052] S25: During the model training process, a data augmentation module was added to help the model better cope with the various forms and complexities of ores by artificially expanding the diversity of the training dataset.

[0053] In one embodiment of the present invention, in step S3, in terms of processing hyperspectral data, an independent convolutional neural network is also used to extract the spectral characteristics of the ore surface. The hyperspectral data mainly describes the spectral response of the ore in different bands, and characterizes the surface chemical composition of the ore through spectral information of up to hundreds of bands. The convolutional neural network processes these spectral images through layer-by-layer convolution. The low-level convolutional layer is used to extract the basic response characteristics of the ore in each band. As the network depth increases, the high-level convolutional layer can gradually learn more complex spectral features.

[0054] In one embodiment of the present invention, in step S4, during the extraction of hyperspectral features, the convolution layer can gradually optimize and extract high-dimensional features through the learning mechanism of the residual block, and the existence of the jump connection allows the original input features to be retained, avoiding information attenuation caused by gradual feature loss in deep convolutions. The multi-layer convolution in the residual block introduces nonlinear transformations through nonlinear activation functions, so that the network can learn more complex spectral patterns from the hyperspectral images of the ore, which reflect the changes in the surface composition and internal correlations of the ore.

[0055] In one embodiment of the present invention, in step S5, the fused features are processed by the deep learning network and then sent to the prediction module. The features of each modality are independently predicted by the prediction head to obtain the preliminary judgment results of the ore category. The independent processing in the prediction stage ensures that the features of each modality are fully reflected in the decision. Subsequently, the prediction results of each modality are finally integrated using an adaptive weighted fusion mechanism. The mechanism dynamically adjusts the influence weight of each modality in the final decision according to the confidence of each modality prediction. The purpose of the adaptive weighting mechanism is to optimize the prediction results so that the model can dynamically adjust the influence of different modal features on the final prediction according to their contribution, thereby ensuring the accuracy and reliability of the prediction. This strategy achieves the optimal use of multimodal data and generates more accurate classification results.

[0056] After the final prediction is completed, the ore category and composition information is passed to the robotic arm sorting module. The sorting system combines visual sensors and control systems to accurately identify the category, location and shape of the ore, and drives the robotic arm to perform grasping and sorting operations. Based on the prediction results of the deep learning model, the robotic arm automatically adjusts its grasping posture and path to adapt to different types and sizes of ores, achieving high-precision grasping and sorting. The robotic arm has multi-degree-of-freedom movement capabilities and can adapt to complex ore shapes and layouts to ensure the accuracy and continuity of sorting operations.

[0057] According to the second aspect of the embodiment of the present invention, there is also provided an automated fine sorting system for ore based on XRT and hyperspectral, which accurately identifies the type and location of ore through an intelligent robotic arm combined with a visual sensor, and realizes automatic grabbing and sorting operations of ore, including a central control system, a robotic arm sorting system, and a communication network. The ore is stably transported from the input end to each functional module by a conveyor belt. The conveyor belt is responsible for moving the ore smoothly along the specified path, and through the entire device, it ensures that the ore can go through a series of complex detection and sorting processes.

[0058] In one embodiment of the present invention, at the initial stage of the conveyor belt, the ore will enter the detection area, where an X-ray source and a hyperspectral camera are arranged to scan the internal structure and surface elements of the ore respectively. The X-ray tomography technology is used to obtain the internal density characteristics and fracture gaps of the ore. By penetrating the ore and detecting the absorption changes of the X-rays, the internal characteristics of the ore can be accurately presented. At the same time, the hyperspectral camera performs multi-band spectral imaging on the surface of the ore to provide detailed information about the surface composition of the ore. By combining these two scanning methods, the system can capture comprehensive data on the internal physical characteristics and surface chemical composition of the ore. In order to ensure the accuracy of the detection process, multiple sensors are installed on the conveyor belt, which are responsible for monitoring the position and state of the ore and controlling the start and stop of the scanner. Through precise position perception, the ore can trigger the scanning of the X-ray source and the hyperspectral camera when it reaches the detection area, ensuring that each piece of ore can be completely collected. These real-time sensor data are directly connected to the central control system to provide dynamic scheduling and control support for the entire detection process.

[0059] In one embodiment of the present invention, during the detection process, the ore passes through a protective cabin arranged on the conveyor belt to ensure the safety of the X-ray during use. The protective cabin is a shielding structure designed around the scanning area to ensure that the rays will not leak into the external environment and avoid potential impact on operators and equipment. At the same time, during the transportation and scanning of the ore, the system will generate a large amount of dust, so the equipment is equipped with dust removal pipes and dust removal devices. These devices can effectively collect and remove dust emitted during the scanning and transportation of the ore, maintain the cleanliness of the equipment, and prevent dust from affecting the detection accuracy and mechanical operation. After the ore is scanned, the conveyor belt continues to send the ore to the robotic arm sorting system. The robotic arm obtains the previous X-ray and hyperspectral scanning data through the connection with the central control system, and classifies and sorts the ore based on these data. The operation of the robotic arm sorting system depends on the central control system and the communication network. The central control system receives data from sensors, X-ray scanning and hyperspectral cameras, and performs real-time analysis and judgment on the ore based on these data. Then, through the communication network, the central control system directs the robotic arm to perform sorting actions. After receiving the command, the robot arm can respond quickly and accurately, grab the ore and place it in the corresponding sorting area. Finally, the ore is divided into concentrate, middlings and tailings according to its characteristics. These ores enter different collection containers through their respective tracks to complete the entire sorting process. At the same time, the waste collection system in the system is used to collect and process ores that do not meet the requirements, ensure the effective use of valuable ore resources, and reduce the accumulation of waste and environmental pollution. The dust removal and waste management system ensures that the ore sorting process is not only accurate but also environmentally friendly and intelligent, thus providing strong hardware support for the efficient use of ore.

[0060] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. An automated fine separation method for ore based on XRT and hyperspectral, characterized in that: The specific steps include: S1: Ore data collection: collect multimodal data of ore samples through XRT and hyperspectral imaging equipment to obtain X-ray transmission imaging data and hyperspectral images of ore; S2: Ore contour extraction, by combining X-ray transmission imaging technology and deep learning models, high-precision extraction of ore contours is achieved; S3: Feature fusion and optimization. In the processing of hyperspectral data, an independent convolutional neural network is also used to extract the spectral features of the ore surface. In the process of extracting hyperspectral features, the convolution layer can gradually optimize and extract high-dimensional features through the learning mechanism of the residual block, and the existence of the jump connection allows the original input features to be retained, avoiding information attenuation caused by gradual feature loss in deep convolution. S4: Classification prediction of deep fusion features. After completing the independent feature extraction of the two modalities, the data is sent to the cross-modal attention mechanism module for deep fusion of XRT and hyperspectral features. The fused features are further extracted through a deep residual network to extract high-order features. S5: Ore category identification and intelligent sorting.

2. The method for automatic fine sorting of ore based on XRT and hyperspectral according to claim 1, characterized in that: In step S1, XRT data can reveal structural information such as internal density distribution, fractures and voids of the ore, while hyperspectral images provide spectral characteristics of the ore in different bands to reveal its surface chemical composition and mineral characteristics. The comprehensive use of the two modes provides a multi-level and rich data foundation for the system, ensuring that effective ore information can be obtained in different environments. Specifically, the following sub-steps are included: S11: For XRT data, preliminary feature extraction is first performed through multiple convolutional layers of the convolutional neural network. XRT data mainly reflects the internal density distribution and structural information of the ore. Through layer-by-layer convolution, CNN can extract the basic morphology and edge features of the ore from a lower level; S12: After multiple convolution operations, the feature map is further used to extract deep features through residual blocks. The residual mechanism makes deep feature extraction no longer rely on relearning all features, but instead performs incremental learning based on the input features. This further enhances the network's ability to extract complex features while retaining the original information.

3. The method for automatic fine sorting of ore based on XRT and hyperspectral according to claim 1, characterized in that: In step S2, X-ray transmission imaging is used to obtain a projection image of the ore. After the X-ray image is obtained, the image will undergo a series of preprocessing steps to improve data quality and provide optimized input for the subsequent deep learning algorithm, specifically including the following sub-steps: S21: After using X-ray transmission imaging technology to obtain the ore projection image, due to the influence of factors such as the external environment and imaging equipment, noise interference will inevitably appear in the image. In order to eliminate these noises and make the image clearer, Gaussian filtering and median filtering are used. Gaussian filtering can effectively remove high-frequency noise while maintaining the continuity of the overall structure by smoothing the image. Combining these two filtering methods, the effective denoising of the ore image is finally achieved to ensure the integrity and clarity of the edge information; S22: After noise removal, in order to enhance the contrast of the image, the histogram equalization technology is used to adjust the distribution of pixel values ​​in the image so that the difference between the ore and the background is more significant, thereby making the boundary of the ore clearer. For some images with large contrast changes in local areas, adaptive histogram equalization (CLAHE) is further introduced. By dividing the image into blocks and performing equalization adjustment on the contrast of different areas, the overall contrast of the image is enhanced and the noise problem introduced by over-enhancement is avoided. S23: In the process of image preprocessing, bilateral filtering is used to retain edge information while removing noise. By combining the spatial distance and grayscale difference between pixels for smoothing, similar areas are smoothed while the edge parts remain clear. The introduction of bilateral filtering makes the ore edge more prominent in the subsequent extraction process, which helps the model to accurately locate the boundary of the ore. S24: After the image preprocessing is completed, the image needs to be normalized to standardize the image pixel values ​​to a uniform range (for example, between 0 and 1). Normalization can eliminate the impact of brightness or intensity differences between different samples, so that the model can learn more essential features under the same input standard, thereby improving the robustness and adaptability of the overall model. S25: During the model training process, a data augmentation module was added to help the model better cope with the various forms and complexities of ores by artificially expanding the diversity of the training dataset.

4. The method for automatic fine sorting of ore based on XRT and hyperspectral according to claim 1, characterized in that: In step S3, in terms of processing hyperspectral data, an independent convolutional neural network is also used to extract the spectral features of the ore surface. The hyperspectral data mainly describes the spectral response of the ore in different bands, and characterizes the surface chemical composition of the ore through spectral information of up to hundreds of bands. The convolutional neural network processes these spectral images through layer-by-layer convolution. The low-level convolutional layer is used to extract the basic response characteristics of the ore in each band. As the network depth increases, the high-level convolutional layer can gradually learn more complex spectral features.

5. The method for automatic fine sorting of ore based on XRT and hyperspectral according to claim 1, characterized in that: In step S4, during the extraction of hyperspectral features, the convolution layer can gradually optimize and extract high-dimensional features through the learning mechanism of the residual block, and the existence of the jump connection allows the original input features to be retained, avoiding information attenuation caused by gradual feature loss in deep convolutions. The multi-layer convolution in the residual block introduces nonlinear transformations through nonlinear activation functions, so that the network can learn more complex spectral patterns from the hyperspectral images of ore, which reflect the changes in the surface composition and internal correlations of the ore.

6. The method for automatic fine sorting of ore based on XRT and hyperspectral according to claim 1, characterized in that: In step S5, the fused features are processed by the deep learning network and then sent to the prediction module. The features of each modality are independently predicted by the prediction head to obtain a preliminary judgment result of the ore category. The independent processing in the prediction stage ensures that the features of each modality are fully reflected in the decision. Subsequently, the prediction results of each modality are finally integrated using an adaptive weighted fusion mechanism. The mechanism dynamically adjusts the influence weight of each modality in the final decision according to the confidence of each modality prediction. The purpose of the adaptive weighted mechanism is to optimize the prediction results so that the model can dynamically adjust its contribution to the different modal features according to their contribution. The influence of the final prediction is used to ensure the accuracy and reliability of the prediction. After the final prediction is completed, the ore category and composition information is passed to the robotic arm sorting module. The sorting system combines visual sensors and control systems to accurately identify the category, location and shape of the ore, and drive the robotic arm to perform grasping and sorting operations. Based on the prediction results of the deep learning model, the robotic arm automatically adjusts its grasping posture and path to adapt to ores of different types and sizes, and achieves high-precision grasping and sorting. The robotic arm has multi-degree-of-freedom movement capabilities and can adapt to complex ore shapes and layouts to ensure the accuracy and continuity of sorting operations.

7. An automated fine ore sorting system based on XRT and hyperspectral, characterized in that: Through the combination of intelligent robotic arms and visual sensors, the type and location of ore can be accurately identified, and automatic grabbing and sorting operations of ore can be realized. Including the central control system, robotic arm sorting system, and communication network, the ore is stably transported from the input end to each functional module by the conveyor belt. The conveyor belt is responsible for moving the ore smoothly along the designated path through the entire equipment to ensure that the ore can undergo a series of complex detection and sorting processes.

8. The XRT and hyperspectral-based automated fine ore sorting system according to claim 7, characterized in that: At the initial stage of the conveyor belt, the ore will enter the detection area, where an X-ray source and a hyperspectral camera are arranged to scan the internal structure and surface elements of the ore respectively. X-ray tomography technology is used to obtain information such as the internal density characteristics and fracture voids of the ore. By penetrating the ore and detecting the absorption changes of X-rays, the internal characteristics of the ore can be accurately presented. At the same time, the hyperspectral camera performs multi-band spectral imaging on the surface of the ore to provide detailed information about the surface composition of the ore. By combining these two scanning methods, the system can capture comprehensive data on the internal physical characteristics and surface chemical composition of the ore. In order to ensure the accuracy of the detection process, multiple sensors are installed on the conveyor belt. These sensors are responsible for monitoring the position and status of the ore and controlling the start and stop of the scanner. Through precise position perception, the ore can trigger the scanning of the X-ray source and the hyperspectral camera when it reaches the detection area, ensuring that the data of each piece of ore can be fully collected. These real-time sensor data are directly connected to the central control system to provide dynamic scheduling and control support for the entire detection process.

9. The XRT and hyperspectral-based automated fine ore sorting system according to claim 7, characterized in that: During the detection process, the ore passes through a protective cabin arranged on the conveyor belt to ensure the safety of X-rays during use. The protective cabin is a shielding structure designed around the scanning area to ensure that the rays will not leak into the external environment and avoid potential impacts on operators and equipment. At the same time, during the transportation and scanning of the ore, the system will generate a large amount of dust. Therefore, the equipment is equipped with dust removal pipes and dust removal devices. These devices can effectively collect and remove dust emitted during the scanning and transportation of the ore, maintain the cleanliness of the equipment, and prevent dust from affecting the detection accuracy and mechanical operation. After the ore is scanned, the conveyor belt continues to send the ore to the robotic arm sorting system. The robotic arm obtains the previous X-ray and hyperspectral scanning data through the connection with the central control system, and classifies and sorts the ore based on these data. The operation of the robotic arm sorting system depends on the central control system. The central control system receives data from sensors, X-ray scanners and hyperspectral cameras, and performs real-time analysis and judgment on the ore based on these data. Then, through the communication network, the central control system instructs the robotic arm to perform sorting. After receiving the command, the robotic arm can respond quickly and accurately, grab the ore and place it in the corresponding sorting area. Finally, the ore is divided into concentrate, middlings and tailings according to its characteristics. These ores enter different collection containers through their respective tracks to complete the entire sorting process. At the same time, the waste collection system in the system is used to collect and process ores that do not meet the requirements, ensuring the effective use of valuable ore resources and reducing waste accumulation and environmental pollution. The dust removal and waste management system ensures that the ore sorting process is not only accurate but also environmentally friendly and intelligent, thereby providing strong hardware support for the efficient use of ore.

10. A computer-readable storage medium having computer-executable instructions stored thereon, characterized in that: When the executable instructions are executed by the processor, the automated fine sorting method for ore based on XRT and hyperspectrum as described in any one of claims 1 to 6 is implemented.

Images