A preselected waste discard rate calculation method based on YOLO series algorithm

Abstract

The application discloses a preselected waste-throwing raw ore waste-throwing rate calculation method based on a YOLO series algorithm, and comprises the following steps: obtaining a raw ore gray image and performing pretreatment to obtain an image dataset; based on the image dataset, a video in a preset coding format is generated to obtain a video dataset; a YOLO target detection or instance segmentation algorithm is used to classify and identify targets in each frame of image in the video dataset, and rectangular frame or mask (Mask) information of ore and waste rock categories is counted and output; according to the rectangular frame or mask information, the areas of the ore and the waste rock are calculated, and then the waste-throwing rate is calculated according to the area ratio to complete the waste-throwing rate calculation. The method can quickly and low-costly evaluate the effect of the preselected waste-throwing scheme, the YOLO target detection and instance segmentation algorithm can be collectively referred to as a YOLO series AI recognition algorithm, can provide decision support for XRT intelligent sorting equipment particle size control and process parameter optimization, and has good engineering adaptability and application prospect.

Description

Technical Field

[0001] This invention belongs to the field of intelligent identification technology for non-ferrous metal mineral resources, specifically involving a method for calculating the waste rate of pre-selected waste ore based on the YOLO series algorithm. Background Technology

[0002] In modern mineral resource utilization systems, photoelectric intelligent mineral processing, as an emerging sorting technology, directly impacts the optimization of mineral processing procedures and resource utilization efficiency through its recognition accuracy and sorting efficiency. This is particularly evident in the pre-selection and waste disposal stages of difficult-to-process ores such as low-grade tungsten ore, where the application value of intelligent sorting technology is increasingly prominent. However, most mainstream industrial sorting equipment currently employs dual-stage target detection algorithms, which have poor compatibility with YOLO series single-stage deep learning algorithms, making direct deployment impossible. This hinders the rapid migration of optimized models to the field, severely restricting the widespread application of intelligent mineral processing algorithms.

[0003] In practical applications, XRT technology has been widely used for the rapid sorting of high-density ores such as tungsten. However, due to the lack of color information, small grayscale value differences, and blurred target boundaries in XRT grayscale images, existing methods lack systematic modeling methods for the relationship between image target area, particle size differences, and waste disposal rate, making it difficult to scientifically calculate the waste disposal effect without hardware dependence.

[0004] Furthermore, before the industrial deployment of intelligent sorting equipment, how to achieve rapid simulation performance evaluation of the recognition algorithm has become a key issue in engineering applications. Traditional methods usually require repeated experiments and debugging using real raw ore materials and industrial sites, which is time-consuming and lacks flexibility. To address this issue, there is currently a lack of an accurate and rapid simulation calculation method for theoretically calculating the pre-selection and waste disposal effect of raw ore, particle size classification, and waste disposal rate prediction during the optimization stage of the YOLO series AI recognition single-stage algorithm. This would provide a reference for equipment parameter design, algorithm structure optimization, and particle size control strategies for the sorted ore.

[0005] In summary, there is an urgent need for an image recognition and statistical calculation method based on the YOLO series algorithms, which can identify target ore and non-target waste rock in images in a simulation environment and automatically calculate the waste rate, providing a low-cost and high-precision solution for the optimization, evaluation and industrial deployment of intelligent mineral processing algorithms. Summary of the Invention

[0006] To address the shortcomings of existing technologies, the present invention aims to provide a method for calculating the waste rate of pre-selected waste ore based on YOLO series algorithms. This method can quickly and cost-effectively evaluate the effectiveness of pre-selected waste disposal schemes, provide decision support for particle size control and process parameter optimization of sorting equipment, and has good engineering adaptability and prospects for widespread application.

[0007] This invention provides a method for calculating the waste rate of pre-selected waste ore based on YOLO series algorithms, comprising the following steps:

[0008] S1. Obtain the original ore grayscale image and perform preprocessing to obtain the image dataset;

[0009] S2. Based on the image dataset, generate a video in a preset encoding format to obtain a video dataset;

[0010] S3. Use the YOLO series AI recognition optimization algorithm to classify and identify targets in each frame of the video dataset, and output the bounding boxes or mask information of ore and waste rock categories;

[0011] S4. Calculate the area of ​​ore and waste rock based on the rectangular frame or mask information, and then calculate the waste disposal rate based on the area ratio to complete the waste disposal rate estimation.

[0012] Step S1 includes the following steps:

[0013] Obtain the grayscale image of the raw ore;

[0014] The original grayscale images of the ore were standardized to obtain an image dataset.

[0015] Step S1 is as follows:

[0016] Obtain the grayscale image of the raw ore, name the image according to a preset unified naming rule, and store it by sorting the file names.

[0017] The width and height of the first image in the named and sorted raw ore grayscale images are used as the standard resolution for subsequent images, while the size is fixed to the size of the first image, thus obtaining the image dataset.

[0018] The process of obtaining the grayscale image of the raw ore includes the following steps:

[0019] The raw mineral material to be processed is fed uniformly and quantitatively onto the material conveyor belt of the XRT intelligent sorting machine through a vibrating feeder to form a stable single layer of material.

[0020] As the raw ore moves along the conveyor belt past the X-ray emitter, high-energy and low-energy X-ray beams simultaneously penetrate the raw ore material, and the transmitted X-ray signal is received by a linear X-ray detector arranged below the conveyor belt.

[0021] The detector converts the received X-ray attenuation signal into grayscale image data and outputs it line by line at a fixed line scanning frequency to form a sequence of XRT grayscale images of the raw ore.

[0022] The sorting machine control system performs time synchronization and spatial calibration on the acquired grayscale images, and processes the images uniformly according to the set resolution and size to obtain a structured raw ore grayscale image dataset.

[0023] The grayscale image dataset is stored in the data processing module for subsequent target detection or instance segmentation and identification, and waste rate estimation.

[0024] Step S2 includes the following steps:

[0025] Preset video encoding format, set output path and frame rate;

[0026] Generate video based on image datasets;

[0027] The generated video is saved in a preset format to obtain a video dataset.

[0028] Step S2 is as follows:

[0029] The OpenCV library is used to initialize the video writing object, the video encoding format is specified as MP4, the output path and frame rate are set, and the total video duration is determined by the number of frames and the frame rate.

[0030] Read each image from the image dataset sequentially and call the corresponding OpenCV library function to write it to the video stream;

[0031] After the video is generated, resources are released first, then the video file is saved, and finally the memory space is cleaned up; the final generated dynamic video file is saved in MP4 format, resulting in a video dataset.

[0032] Step S3 includes the following steps:

[0033] The YOLO series AI recognition optimization algorithm is used to recognize video datasets and obtain bounding boxes or mask information with target ore and waste rock as recognition targets.

[0034] Based on the category labels corresponding to the bounding box or mask information, count the number of target ore label bounding boxes and waste rock label bounding boxes identified;

[0035] For YOLO series target detection algorithms, the two-dimensional projected area is calculated based on the center coordinates and side length parameters of the bounding box, and the granularity interval is divided according to the area value. For YOLO series instance segmentation algorithms, the pixel set of the original target on the image plane is obtained by using the output instance mask, and the true projected area is calculated by using the number of mask pixels and the single pixel area parameter calibrated by the camera, thus optimizing the area error problem caused by the bounding box.

[0036] By using a pre-defined dictionary recording structure, the data records of all rectangle or mask categories, sizes, areas, and granularity intervals in each frame of the video dataset are stored in text format.

[0037] Step S4 includes the following steps:

[0038] Based on the projected area of ​​the target ore and waste rock in the image output by the target detection or instance segmentation algorithm, the pixel area or mask area of ​​various targets is statistically analyzed.

[0039] Based on the statistical results of the projected area, the proportion of the total area of ​​pixels identified as waste rock and thrown out to the total area of ​​all raw ore pixels is calculated to obtain the raw ore waste rate.

[0040] The projected area and quantity of targets in different particle size ranges are statistically analyzed to obtain the waste rate and particle size distribution by particle size.

[0041] The overall waste rate and the waste rate by particle size are output together as the final result of the raw ore waste rate calculation.

[0042] XRT imaging is essentially a two-dimensional projection, independent of thickness. It approximates the ore volume using the area of ​​a rectangle or mask, and is expressed by the following formula based on the density relationship between the target ore and waste rock in the original ore: Where T is the total mass of waste rock; O is the total mass of ore; The total area of ​​the rectangular bounding box or mask of the waste rock label in the image; The total area of ​​the rectangular bounding box or mask of the mineral label in the image; The density of waste rock; The target ore density.

[0043] If an object detection algorithm is used for AI recognition, then the total area of ​​the rectangular frame of the waste rock label in the image is... Calculate using the following formula: ; This represents the total number of rectangular frames labeled with waste rock. The width of the rectangle containing the i-th waste rock label; Here is the height of the j-th waste rock label rectangle; and the total area of ​​the target ore label rectangles in the image. Calculate using the following formula: ; This represents the total number of rectangular boxes labeled with ore. The width of the rectangle containing the j-th ore label; Let be the height of the rectangle containing the j-th ore label.

[0044] If an instance segmentation algorithm is used for AI recognition, then the total mask area of ​​waste rock label instances in the image is... Calculate using the following formula: ;in, The total number of waste rock label instance masks; The number of pixels within the mask of the i-th waste rock instance; This represents the actual projected area corresponding to a single pixel; similarly, it represents the total mask area of ​​mineral label instances in the image. Calculate using the following formula: ;in, The total number of ore label instance masks; Let be the number of pixels within the mask of the j-th ore instance. By using the above method of calculating the area based on the pixel count of the instance mask, the two-dimensional projected area of ​​the true outline of the original ore target can be directly reflected, avoiding the area error problem caused by the rectangular bounding box, thereby improving the accuracy and stability of the scrap rate estimation.

[0045] The total waste rate (WR) of raw ore is calculated using the following formula: Then, based on the rectangular frame or mask area data of different particle size ranges, the waste rate of different particle size ranges is calculated to obtain the distribution of the waste rate of raw ore at different particle sizes.

[0046] This invention discloses a method for calculating the waste rate of pre-selected waste ore based on YOLO series algorithms, which has the following beneficial effects:

[0047] This invention enables rapid assessment of the waste rate of raw ore by identifying the distribution areas of ore and waste rock in XRT grayscale images and calculating the target proportion by combining rectangular box or mask area information, without relying on actual sorting equipment.

[0048] This invention uses image recognition combined with surface accumulation and calculation mechanism to replace traditional weighing and testing operations. It can complete the statistical analysis of a large number of image samples in a short time. It is suitable for engineering simulation and strategy debugging of raw ores of different particle sizes and different categories, and has good versatility and scalability.

[0049] This invention, when used with visualization software, can output a particle size distribution statistical table, providing theoretical support and data reference for the particle size control of feed in intelligent mineral processing equipment and the strategy of waste rock resource reuse.

[0050] The theoretical waste rejection rate obtained in the experimental verification is within a reasonable range from the actual data in the sorting workshop. The results are reliable and the calculations are scientific. It can be widely applied to the sorting optimization of high-density ores such as tungsten, and contribute to the intelligent and green development of mineral processing technology. Attached Figure Description

[0051] Figure 1 This is a schematic flowchart of the method of the present invention;

[0052] Figure 2This is an example of a video result image identified using the YOLO series target detection algorithm in an embodiment of the present invention.

[0053] Figure 3 This is an example of a video recognition result image obtained using the YOLO series instance segmentation algorithm in an embodiment of the present invention. Detailed Implementation

[0054] This invention provides a method for calculating the waste rate of pre-selected waste ore based on YOLO series algorithms, the flowchart of which is shown below. Figure 1 As shown, it includes the following steps:

[0055] S1. Obtain the original ore grayscale image and perform preprocessing to obtain the image dataset;

[0056] Step S1 includes the following steps:

[0057] Obtain the grayscale image of the raw ore;

[0058] The original grayscale images of the ore were standardized to obtain an image dataset.

[0059] Step S1 is as follows:

[0060] Obtain the grayscale image of the raw ore, name the image according to a preset unified naming rule, and store it by sorting the file names.

[0061] The width and height of the first image in the named and sorted raw ore grayscale images are used as the standard resolution for subsequent images, while the size is fixed to the size of the first image, thus obtaining the image dataset.

[0062] The process of obtaining the grayscale image of the raw ore includes the following steps:

[0063] The raw mineral material to be processed is fed uniformly and quantitatively onto the material conveyor belt of the XRT intelligent sorting machine through a vibrating feeder to form a stable single layer of material.

[0064] As the raw ore moves along the conveyor belt past the X-ray emitter, high-energy and low-energy X-ray beams simultaneously penetrate the raw ore material, and the transmitted X-ray signal is received by a linear X-ray detector arranged below the conveyor belt.

[0065] The detector converts the received X-ray attenuation signal into grayscale image data and outputs it line by line at a fixed line scanning frequency to form a sequence of XRT grayscale images of the raw ore.

[0066] The sorting machine control system performs time synchronization and spatial calibration on the acquired grayscale images, and processes the images uniformly according to the set resolution and size to obtain a structured raw ore grayscale image dataset.

[0067] The grayscale image dataset is stored in the data processing module for subsequent target detection or instance segmentation and identification, and waste rate estimation.

[0068] S2. Based on the image dataset, generate a video in a preset encoding format to obtain a video dataset;

[0069] Step S2 includes the following steps:

[0070] Preset video encoding format, set output path and frame rate;

[0071] Generate video based on image datasets;

[0072] Save the generated video in a preset format to obtain a video dataset.

[0073] Step S2 is as follows:

[0074] The OpenCV library is used to initialize the video writing object, the video encoding format is specified as MP4, the output path and frame rate are set, and the total video duration is determined by the number of frames and the frame rate.

[0075] Read each image from the image dataset sequentially and call the corresponding OpenCV library function to write it to the video stream;

[0076] After the video is generated, resources are released first, then the video file is saved, and finally the memory space is cleaned up; the final generated dynamic video file is saved in MP4 format, resulting in a video dataset.

[0077] S3. Use the YOLO series AI recognition optimization algorithm to classify and identify targets in each frame of the video dataset, and output the bounding boxes or mask information of ore and waste rock categories;

[0078] Step S3 includes the following steps:

[0079] The YOLO series AI recognition optimization algorithm is used to recognize video datasets and obtain rectangular boxes or mask information with target ore and waste rock as recognition targets.

[0080] Based on the category labels corresponding to the bounding box or mask information, count the number of target ore label bounding boxes and waste rock label bounding boxes identified;

[0081] For YOLO series target detection algorithms, the two-dimensional projected area is calculated based on the center coordinates and side length parameters of the bounding box, and the granularity interval is divided according to the area value. For YOLO series instance segmentation algorithms, the pixel set of the original target on the image plane is obtained by using the output instance mask, and the true projected area is calculated by using the number of mask pixels and the single pixel area parameter calibrated by the camera, thus optimizing the area error problem caused by the bounding box.

[0082] By using a pre-defined dictionary recording structure, the data records of all rectangle or mask categories, sizes, areas, and granularity intervals in each frame of the video dataset are stored in text format.

[0083] S4. Calculate the area of ​​ore and waste rock based on the rectangular frame or mask information, and then calculate the waste disposal rate based on the area ratio to complete the waste disposal rate estimation.

[0084] Step S4 includes the following steps:

[0085] Step S4 includes the following steps:

[0086] Based on the projected area of ​​the target ore and waste rock in the image output by the target detection or instance segmentation algorithm, the pixel area or mask area of ​​various targets is statistically analyzed.

[0087] Based on the statistical results of the projected area, the proportion of the total area of ​​pixels identified as waste rock and thrown out to the total area of ​​all raw ore pixels is calculated to obtain the raw ore waste rate.

[0088] The projected area and quantity of targets in different particle size ranges are statistically analyzed to obtain the waste rate and particle size distribution by particle size.

[0089] The overall waste rate and the waste rate by particle size are output together as the final result of the raw ore waste rate calculation.

[0090] XRT imaging is essentially a two-dimensional projection, independent of thickness. It approximates the volume of the original ore by using a rectangular frame or mask area. Based on the density relationship between the target ore and waste rock in the original ore, it is expressed by the following formula: Where T is the total mass of waste rock; O is the total mass of target ore; The total area of ​​the rectangular bounding box or mask of the waste rock label in the image; The total area of ​​the target ore label rectangle or mask in the image; The density of waste rock; The target ore density.

[0091] If an object detection algorithm is used for AI recognition, then the total area of ​​the rectangular frame of the waste rock label in the image is... Calculate using the following formula: ; This represents the total number of rectangular frames labeled with waste rock. The width of the rectangle containing the i-th waste rock label; Here is the height of the j-th waste rock label rectangle; and the total area of ​​the target ore label rectangles in the image. Calculate using the following formula: ; Total number of target ore label rectangles; The width of the rectangle containing the j-th target ore label; Let be the height of the rectangle containing the j-th target ore label.

[0092] If an instance segmentation algorithm is used for AI recognition, then the total mask area of ​​waste rock label instances in the image is... Calculate using the following formula: ;in, The total number of waste rock label instance masks; The number of pixels within the mask of the i-th waste rock instance; This represents the actual projected area corresponding to a single pixel; similarly, it represents the total mask area of ​​the target ore label instances in the image. Calculate using the following formula: ;in, The total number of masks for the target ore label instances; Let be the number of pixels within the mask of the j-th target ore instance. By using the above method of area calculation based on the pixel count of the instance mask, the two-dimensional projected area of ​​the true outline of the original ore target can be directly reflected, avoiding the area error problem caused by the rectangular bounding box, thereby improving the accuracy and stability of the scrap rate estimation.

[0093] The total waste rate (WR) of raw ore is calculated using the following formula: Then, based on the rectangular frame or mask area data of different particle size ranges, the waste rate of different particle size ranges is calculated to obtain the distribution of the waste rate of raw ore at different particle sizes.

[0094] The method of the present invention will be further described below with reference to the embodiments:

[0095] A sample of raw scheelite was randomly selected in a production workshop, and 225 XRT static grayscale images were collected. These images were then synthesized into a dynamic MP4 video at a fixed frame rate of 6.25 FPS to simulate the actual pre-selection and disposal process.

[0096] The scheelite has a grade of only 0.25%. The density of individual ore and waste rock in the raw ore is close and can be regarded as the same. Therefore, in this embodiment, the area ratio of the rectangular frame is used as the mass ratio of waste ore to raw ore.

[0097] Table 1 summarizes the statistical results obtained from identifying objects in dynamic videos using the YOLO series object detection algorithms. The algorithms identified a total of 940 ore targets and 1018 waste targets, with the proportion of ore targets being roughly equal to the number of waste rock targets. These statistical results not only quantify the classification ability of the YOLO series object detection algorithms in simulated real-world waste disposal scenarios but also provide a quantitative basis for extrapolating waste disposal rates in a simulation environment.

[0098] The discard rate of the scheelite ore sample was estimated to be 50.97% based on the target detection algorithm. However, in actual industrial production scenarios, the high-speed solenoid valve of the XRT intelligent sorting machine is affected by factors such as missed detection by the identification algorithm, lag in solenoid valve response, and target positioning error when executing the blowing command. The actual discard rate will be 10-12% lower than the theoretically estimated value.

[0099] Taking the target ore and waste rock in the XRT grayscale image dataset of raw ore as the research object, an algorithm was designed to statistically analyze the waste rate corresponding to their particle size distribution, as shown in Table 1.

[0100] Table 1. Statistical Table of Raw Ore Grain Size Zoning Waste Rate for Target Detection Algorithm

[0101]

[0102] As shown in Table 1, there are significant differences in the distribution of different particle sizes in the raw ore sample. The ore with a particle size of 30–40 mm is the most abundant, accounting for 50.41%, making it the primary particle size for selection. This is followed by the 40–50 mm particle size (27.58%), the 20–30 mm particle size (9.65%), and the 50–60 mm particle size (9.19%). The proportion of raw ore with a particle size greater than 60 mm is extremely low, at only 3.17%.

[0103] To demonstrate the generalization ability of the algorithm, 225 grayscale images were reselected to synthesize another MP4 format dynamic video. The statistical results obtained by using the YOLO series instance segmentation algorithm to identify the dynamic video are summarized in Table 2. The algorithm identified a total of 696 ore targets and 1277 waste targets.

[0104] Using the ore and waste rock in the XRT grayscale image dataset of raw ore as the research object, an algorithm was designed to statistically analyze the waste rate corresponding to their particle size distribution, as shown in Table 2. The waste rate of this scheelite raw ore sample was calculated to be 55.70% using an instance segmentation algorithm.

[0105] Table 2. Statistical Table of Raw Ore Grain Size Division Waste Rate in Instance Segmentation Algorithm

[0106]

[0107] The instance segmentation algorithm also exhibits a clear granularity pattern in the raw ore particle size distribution and waste rate statistics: fine-grained (10–40 mm) waste rock has a high proportion and a high waste rate; coarse-grained (≥40 mm) ore has a high proportion and a significantly lower waste rate. This phenomenon is consistent with the particle size distribution and XRT transmission characteristics of the raw ore, verifying the effectiveness and rationality of the instance segmentation algorithm in ore identification and waste rate estimation. Comparing the actual waste data from the experimental workshop with the results obtained from the two embodiments, the simulated waste rate based on area proportion proposed in this invention is basically within a reasonable error range. This estimation method can provide a reference for the particle size control of XRT intelligent mineral processing equipment and the formulation of waste disposal strategies.

Claims

1. A method for calculating the waste rate of pre-selected waste ore based on YOLO series algorithms, characterized in that, Includes the following steps: S1. Obtain the original ore grayscale image and perform preprocessing to obtain the image dataset; S2. Based on the image dataset, generate a video in a preset encoding format to obtain a video dataset; S3. Use the YOLO series AI recognition optimization algorithm to classify and identify targets in each frame of the video dataset, and output the bounding boxes or mask information of ore and waste rock categories; S4. Calculate the area of ​​ore and waste rock based on the rectangular frame or mask information, and then calculate the waste rejection rate based on the area ratio to complete the waste rejection rate estimation; providing a reference for the particle size control of XRT intelligent mineral processing equipment and the formulation of waste rejection strategies; The process of obtaining the grayscale image of the raw ore includes the following steps: The raw mineral material to be processed is fed uniformly and quantitatively onto the material conveyor belt of the XRT intelligent sorting machine through a vibrating feeder to form a stable single layer of material. As the raw ore moves along the conveyor belt past the X-ray emitter, high-energy and low-energy X-ray beams simultaneously penetrate the raw ore material, and the transmitted X-ray signal is received by a linear X-ray detector arranged below the conveyor belt. The detector converts the received X-ray attenuation signal into grayscale image data and outputs it line by line at a fixed line scanning frequency to form a sequence of XRT grayscale images of the raw ore. The sorting machine control system performs time synchronization and spatial calibration on the acquired grayscale images, and processes the images uniformly according to the set resolution and size to obtain a structured raw ore grayscale image dataset. The grayscale image dataset is stored in the data processing module for subsequent target detection or instance segmentation and identification, and waste rate estimation. Step S3 includes the following steps: The YOLO series AI recognition optimization algorithm is used to identify video datasets and obtain rectangular boxes or masks with ore and waste rock as the recognition targets. Based on the category labels corresponding to the rectangles or masks, count the number of ore label rectangles and waste rock label rectangles that are identified; For YOLO series target detection algorithms, the two-dimensional projected area is calculated based on the center coordinates and side length parameters of the rectangular box, and the granularity interval is divided according to the area value; for YOLO series instance segmentation algorithms, the pixel set of the original target on the image plane is obtained by using the output instance mask, and the true projected area is calculated by using the number of mask pixels and the single pixel area parameter calibrated by the camera, thus optimizing the area error problem caused by the bounding box. By using a pre-defined dictionary recording structure, the data records of all rectangle or mask categories, sizes, areas, and granularity intervals in each frame of the video dataset are stored in text form. Step S4 includes the following steps: Based on the projected area of ​​the target ore and waste rock in the image output by the target detection or instance segmentation algorithm, the pixel area or mask area of ​​various targets is statistically analyzed. Based on the statistical results of the projected area, the proportion of the total area of ​​pixels identified as waste rock and thrown out to the total area of ​​all raw ore pixels is calculated to obtain the raw ore waste rate. The projected area and quantity of targets in different particle size ranges are statistically analyzed to obtain the waste rate and particle size distribution by particle size. The overall waste rate and the waste rate by grain size are output together as the final result of the raw ore waste rate estimation; the total waste rate WR of the ore is calculated using the following formula: Then, based on the rectangular frame or mask area data of different particle size ranges, the waste rate of different particle size ranges is calculated to obtain the distribution of the waste rate of raw ore at different particle sizes.

2. The method for calculating the waste rate of pre-selected waste ore based on YOLO series algorithms according to claim 1, characterized in that, Step S1 includes the following steps: Obtain the grayscale image of the raw ore; The original grayscale images of the ore were standardized to obtain an image dataset.

3. The method for calculating the waste rate of pre-selected waste ore based on YOLO series algorithms according to claim 1, characterized in that, Step S2 includes the following steps: Preset video encoding format, set output path and frame rate; Generate video based on image datasets; Save the generated video in a preset format to obtain a video dataset.

4. The method for calculating the waste rate of pre-selected waste ore based on YOLO series algorithms according to claim 1, characterized in that, In step S4, XRT imaging is essentially a two-dimensional projection, independent of thickness. Ignoring the thickness difference of the original ore, the volume of the original ore is approximated by the area of ​​a rectangular frame. Based on the density relationship between the target ore and waste rock in the original ore, the following formula is used: Where T is the total mass of waste rock; O is the total mass of target ore; The total area of ​​the rectangular bounding box or mask of the waste rock label in the image; The total area of ​​the target ore label rectangle or mask in the image; The density of waste rock; This represents the density of the ore.

5. The method for calculating the waste rate of pre-selected waste ore based on YOLO series algorithms according to claim 4, characterized in that, When using object detection algorithms for AI recognition, the total area of ​​the rectangular frames containing waste rock labels in the image... Calculate using the following formula: ; This represents the total number of rectangular frames labeled with waste rock. The width of the rectangle containing the i-th waste rock label; Let the height be the rectangle of the j-th waste rock label; Total area of ​​the target ore label rectangle in the image Calculate using the following formula: ; Total number of target ore label rectangles; The width of the rectangle containing the j-th target ore label; Let the height be the rectangle of the j-th target ore label; When using instance segmentation algorithms for AI recognition, the total area of ​​the waste rock label instance mask in the image is... Calculate using the following formula: ;in, The total number of waste rock label instance masks; The number of pixels within the mask of the i-th waste rock instance; This represents the actual projected area corresponding to a single pixel; similarly, it represents the total mask area of ​​the target ore label instances in the image. Calculate using the following formula: ;in, The total number of masks for the target ore label instances; The number of pixels within the mask of the j-th target ore instance.

Images