Gold tailings drying sand quality detection method based on machine vision

Abstract

The present invention relates to the technical field of dried sand detection and analysis, and in particular to a method for detecting the quality of dried sand of gold tailings based on machine vision. The content includes: collecting multi-angle image data and infrared sensor data of dried sand particles, performing intelligent image processing on the multi-angle image data of dried sand particles, and obtaining the final comprehensive characteristics of dried sand particles; based on the infrared sensor data, performing moisture detection on dried sand particles to obtain the moisture content of dried sand particles; based on the moisture content of dried sand particles and the final comprehensive characteristics of dried sand particles, obtaining complete feature data of dried sand particles; performing multimodal fusion and analysis on the complete feature data of dried sand particles to generate comprehensive analysis results; based on the comprehensive analysis results, generating a detection report. The technical problem of insufficient detection accuracy of key quality indicators such as dried sand particle morphology and moisture content is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of dried sand detection and analysis, and in particular to a gold tailings dried sand quality detection method based on machine vision. Background Art

[0002] The quality inspection of dried gold tailings sand is particularly important in the field of mineral processing, involving the product quality and process efficiency of subsequent smelting, purification and other links. However, the traditional quality inspection method of dried gold tailings sand mainly relies on manual observation and mechanical screening, which not only has poor detection accuracy and consistency, but also has obvious deficiencies in automation. Manual inspection methods are highly subjective and time-consuming and labor-intensive, and cannot guarantee the stability of the morphology and moisture content of each dried sand particle. Although mechanical screening can improve efficiency, its accuracy in identifying dried sand particles is limited, especially for subtle indicators such as the morphology and moisture content of dried sand particles, which are difficult to accurately control.

[0003] The particle morphology and moisture content of gold tailings dried sand are important indicators that affect its quality. Differences in the morphology of dried sand particles, such as particle size, surface roughness, roundness, etc., will directly affect the uniformity and quality of subsequent processing. The moisture content plays a key role in the drying effect, stability during subsequent storage and transportation, especially in the drying process, where moisture control is one of the difficulties in quality inspection. Traditional methods are difficult to detect the morphology and moisture content of dried sand particles in real time and accurately, and cannot meet the quality control needs of large-scale continuous production.

[0004] In summary, the traditional quality detection method of gold tailings dried sand has the technical problem of insufficient detection accuracy of key quality indicators such as dried sand particle morphology and moisture content. Summary of the invention

[0005] The present invention provides a gold tailings dried sand quality detection method based on machine vision to solve the technical problem of insufficient detection accuracy of key quality indicators such as dried sand particle morphology and moisture content.

[0006] The gold tailings drying sand quality detection method based on machine vision includes the following steps:

[0007] S1. Collect multi-angle image data and infrared sensor data of dried sand particles, and perform intelligent image processing on the multi-angle image data of dried sand particles through a multi-partition multi-scale adaptive feature reconstruction algorithm to obtain the final comprehensive features of the dried sand particles;

[0008] S2. Based on the infrared sensor data, a moisture detection algorithm based on multi-dimensional spatiotemporal feature mapping and adaptive filtering is introduced to detect the moisture of the dried sand particles to obtain the moisture content of the dried sand particles; based on the moisture content of the dried sand particles and the final comprehensive characteristics of the dried sand particles, the complete feature data of the dried sand particles is obtained; multi-modal fusion and analysis are performed on the complete feature data of the dried sand particles to generate a comprehensive analysis result; based on the comprehensive analysis result, a detection report is generated.

[0009] Preferably, the S1 specifically includes:

[0010] The specific implementation steps of the multi-partition multi-scale adaptive feature reconstruction algorithm are as follows:

[0011] In the first step, an adaptive partitioning method is introduced to perform image multi-partitioning processing on the multi-angle image data of the dried sand particles to obtain the final partitioning;

[0012] The second step is to perform multi-scale feature extraction on the multi-angle image data of the dried sand particles in each area after the final partitioning to generate a feature tensor matrix;

[0013] The third step is to perform dynamic feature reconstruction and nonlinear fusion on the feature tensor matrix generated in each area after the final partitioning to generate the comprehensive features of the dried sand particles;

[0014] The fourth step is to optimize the comprehensive characteristics of the dried sand particles to obtain the final comprehensive characteristics of the dried sand particles.

[0015] Preferably, the S1 specifically includes:

[0016] In the process of implementing the adaptive partitioning method, the multi-angle image data of the dried sand particles are converted into gradient vectors, and the local mean brightness value and local grayscale variance of each pixel in the multi-angle image data of the dried sand particles are combined to assign a significance weight to each pixel in the multi-angle image data of the dried sand particles; the specific calculation formula for the significance weight of each pixel is:

[0017] ,

[0018] in, Represents the significance weight of each pixel; and It is the multi-angle image data of dried sand particles exist and Directional gradient; is the local mean brightness value of the current pixel; is the local grayscale variance of the current pixel; is a smoothing parameter that prevents the denominator from being zero;

[0019] Based on the significance weight of each pixel point, the partition ratio parameter is introduced to perform a preliminary division on the multi-angle image data of the entire dried sand particles to obtain preliminary partitions, and the significance weights of each area after the preliminary partitions are calculated.

[0020] Preferably, the S1 specifically includes:

[0021] The edges of the regions after the preliminary partitioning are closed to obtain the final partitioning; based on the final partitioning, the significance weights of the regions are recalculated and updated.

[0022] Preferably, the S1 specifically includes:

[0023] Based on the multi-angle image data of the dried sand particles in each area after the final partition, the local contrast, surface roughness and average brightness are defined to generate the partition feature tensor; based on the partition feature tensor, the feature tensor matrix is ​​constructed; the generation formula of the partition feature tensor is as follows:

[0024] ,

[0025] in, After the final partition Areas in location The multi-scale feature tensor value at represents the partition feature tensor; Indicates the final partition The local contrast of each region; Indicates the final partition Surface roughness of the area; After the final partition The average brightness of the area; After the final partition The local edge orientation angle of the dried sand particles in each area; After the final partition The scaling exponent of the partition feature tensor for each region.

[0026] Preferably, the S1 specifically includes:

[0027] The feature tensor matrix is ​​dynamically reconstructed through nonlinear transformation to generate the reconstructed features within the region; the reconstructed features in all regions after the final partition are weighted and merged to generate the comprehensive features of the dried sand particles.

[0028] Preferably, the S2 specifically includes:

[0029] Based on the infrared sensor data, the absorption data, reflectivity data and transmittance data of the dried sand particles are obtained to generate an infrared feature matrix containing time and space dimensions; the spatial offset range radius is introduced to generate a spatial distribution matrix.

[0030] Preferably, the S2 specifically includes:

[0031] The infrared feature matrix and the spatial distribution matrix are converted into a feature space matrix through mapping processing; the feature space matrix is ​​subjected to convolution filtering processing to obtain a feature space matrix after convolution filtering.

[0032] Preferably, the S2 specifically includes:

[0033] Based on the feature space matrix and spatial distribution matrix after convolution filtering, the moisture content calculation formula of dried sand particles is introduced to obtain the moisture content of dried sand particles. The specific formula is:

[0034] ,

[0035] in, yes Always at the location The moisture content of the dried sand particles; , is the adjustment coefficient; represents mean calculation; It is the feature space matrix after convolution filtering; , , Respectively Always dry sand particles in position Absorption data, reflectivity data and transmittance data at; is the spatial distribution matrix in The element value at .

[0036] Preferably, the S2 specifically includes:

[0037] Based on the moisture content of the dried sand particles and combined with the final partition, the moisture content of the dried sand particles in each area is calculated, and the moisture content of the dried sand particles in each area is added to the final comprehensive features of the dried sand particles in the corresponding area to obtain the complete feature data of the dried sand particles; the complete feature data of the dried sand particles are multimodally fused to generate a comprehensive quality score of the dried sand particles; based on the comprehensive quality score of the dried sand particles and the complete feature data of the dried sand particles, the dried sand particles are quality inspected and classified to obtain a comprehensive analysis result; based on the comprehensive analysis result, a test report including the particle qualified rate and the proportion of unqualified particles is generated; when the proportion of unqualified particles exceeds the preset threshold, an alarm is triggered and production adjustments are made.

[0038] The beneficial effects of the technical solution of the present invention are:

[0039] 1. A multi-partition and multi-scale adaptive feature reconstruction algorithm is introduced to accurately extract key features such as the size, surface features, and morphology of dried sand particles, thereby effectively reducing the detection bias caused by a single perspective.

[0040] 2. The infrared sensor combined with the moisture detection algorithm based on multi-dimensional spatiotemporal feature mapping and adaptive filtering can obtain accurate data on the moisture content of the dried sand particles. Through the spatiotemporal analysis of multi-dimensional features such as absorptivity, reflectivity and transmittance, the thermal response changes of the dried sand particles during the drying process can be captured, thus improving the accuracy of the moisture content determination of the dried sand particles. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a flow chart of the gold tailings drying sand quality detection method based on machine vision described in the present invention. DETAILED DESCRIPTION

[0042] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiment of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0044] The specific scheme of the gold tailings drying sand quality detection method based on machine vision provided by the present invention is described in detail below with reference to the accompanying drawings.

[0045] See attached Figure 1 , which shows a flow chart of a gold tailings drying sand quality detection method based on machine vision provided by an embodiment of the present invention, the method comprising the following steps:

[0046] S1. Collect multi-angle image data and infrared sensor data of dried sand particles, and perform intelligent image processing on the multi-angle image data of dried sand particles through a multi-partition multi-scale adaptive feature reconstruction algorithm to obtain the final comprehensive features of the dried sand particles;

[0047] Multimodal data collection of gold tailings dried sand is carried out. According to the expert experience method, professional technicians select high-resolution industrial cameras (such as 5 million pixels or above), and install them around the dried sand particles from multiple angles to ensure that the complete morphological information of the dried sand particles can be collected from different angles, and multi-angle image data of the dried sand particles can be obtained. At the same time, the clarity of the image details is ensured to facilitate the subsequent analysis of the characteristics of the dried sand particles; professional infrared sensors, such as infrared moisture sensors, are selected and installed above the conveyor belt that transports the dried sand, before or after the camera shooting position, to ensure that both image data and infrared sensor data can be obtained when the dried sand particles pass through the conveyor belt;

[0048] Furthermore, a multi-partition multi-scale adaptive feature reconstruction algorithm is introduced to perform intelligent image processing on the multi-angle image data of the dried sand particles to obtain the final comprehensive features of the dried sand particles; the multi-partition multi-scale adaptive feature reconstruction algorithm realizes dynamic feature reconstruction and fusion through a multi-partition multi-scale method, so as to more accurately capture the surface features, size and morphology of the dried sand particles in the multi-angle image data of the dried sand particles, and generate the final comprehensive features of the dried sand particles. The specific implementation process is as follows:

[0049] The first step is image multi-partition processing;

[0050] The adaptive partitioning method is introduced to process the multi-angle image data of the dried sand particles according to the adaptive partitioning rules and divide them into regions, so that each region can independently characterize its edge features and internal structure. Through image multi-partition processing, the image data of dried sand particles taken at various angles are converted into partition features to ensure that subsequent feature extraction can more comprehensively cover the information of dried sand particles. The process of the adaptive partitioning method is as follows:

[0051] First, the multi-angle image data of the dried sand particles is input Convert to gradient vector , obtain the gradient changes of multi-angle image data of dried sand particles at different coordinates, and combine the local mean brightness value of each pixel and local grayscale variance , a significance weight is assigned to each pixel in the multi-angle image data of the dried sand particles. The specific formula is:

[0052] ,

[0053] in, Represents the saliency weight of each pixel, which is used to determine edge saliency to ensure that edge features are easier to partition; and It is the multi-angle image data of dried sand particles exist and The gradient of the direction is used to represent the degree of change of the pixel value in the multi-angle image data of the dried sand particles; is the local mean brightness value of the current pixel; is the local grayscale variance of the current pixel; is a smoothing parameter to prevent the denominator from being zero. The significance weights of different pixels are obtained by calculation. Based on the significance weight of each pixel, the multi-angle image data of the entire dried sand particles are initially divided into regions, and obtain preliminary partitions so that each region can evenly cover the more significant parts; the specific formula of the constraint condition is:

[0054] ,

[0055] in, After the initial partition Regions The total weight of Regions The significance weight of After the initial partition The partition ratio parameter of each region is used to control the proportion of the significance weight of each region in the overall significance weight; After the initial partition All the regions constitute the multi-angle image data of the whole dried sand particles; After the initial partition Partitions The saliency weight of each pixel in ; is the significance weight of each pixel in the multi-angle image data of the entire dried sand particles; Describes the relative proportion of the significance weight of each partition, that is, in the multi-angle image data of the whole dried sand particles, the of the region The significance weight Overall significance weight The ratio is determined by the parameter control;

[0056] Through the above constraints, each region can contain both significant edge features and weak edge information of the internal structure, so as to ensure that the different morphological features of the dried sand particles can be fully expressed in the image partition. The calculation of the significance weight of each region ensures that the edge features in the region are effectively divided, laying the foundation for subsequent feature extraction and analysis;

[0057] Furthermore, in order to ensure the smoothness and continuity of the edges between regions, after the initial partitioning, it is necessary to perform a closing operation on the edges of each partition to obtain the final partition, so as to eliminate the gaps and breaks on the edges and form a smooth edge boundary. The formula for the closing operation is as follows:

[0058] ,

[0059] in, Represents the image after the closing operation, that is, the image after the final partition Regions; After the initial partition Regions, namely the original image of preliminary partitioning; As a morphological structural element, a small rectangle can be selected according to the specific scene; in the closing operation, the expansion operation is performed first , so that the boundary is connected, and then the corrosion operation is performed , remove boundary redundancy, ensure the smooth edge of each area, and reduce the influence of noise;

[0060] Based on the final partitioned area division, recalculate the significance weight of the updated area;

[0061] The second step is to extract multi-scale features within the region;

[0062] In each area, the surface morphology, roughness, and brightness of the dried sand particles are fully captured from different scales. To this end, the fine-grained features such as local contrast, surface roughness, and average brightness are defined for the pixel points of the multi-angle image data of the dried sand particles in each area after the final partitioning, and the partition feature tensor is generated. The generation formula of the partition feature tensor is as follows:

[0063] ,

[0064] in, After the final partition Areas in location The multi-scale feature tensor value at , that is, the partition feature tensor; Indicates the first The local contrast of each region; Indicates the first Surface roughness of the area; After the final partition The average brightness of the area; After the final partition The local edge orientation angle of the dried sand particles in each area; After the final partition The scaling index of the partition feature tensor of the region; Used to prevent the denominator from being zero;

[0065] Furthermore, based on the partition feature tensor, a feature tensor matrix is ​​constructed; each element in the feature tensor matrix reflects the contrast, surface roughness and average brightness at the local scale; The feature tensor matrix of the region It is expressed as:

[0066] ,

[0067] Feature tensor matrix The elements represent the first The multi-scale feature tensor value of each pixel in the region provides a high-precision partition feature description; It is the first Regions The number of columns represents the number of horizontal pixels of the image in the final partitioned area; It is the first Regions The number of rows represents the number of vertical pixels of the image in the final partitioned area; It is the first The feature tensor matrix of the region is The multi-scale feature tensor value of the pixel at . The feature tensor matrix will be used as the basic data source for dynamic feature reconstruction in subsequent steps;

[0068] The third step is dynamic feature reconstruction and nonlinear fusion;

[0069] After generating the feature tensor matrix in each region after the final partition, dynamic feature reconstruction processing is performed to generate the reconstructed features in the region. are combined to produce a reconstructed signature for each region:

[0070] ,

[0071] in, It is the first The reconstructed features of the region are used to represent the comprehensive features of the region, which comprehensively expresses the final partitioning. Regions Internal All The multi-scale feature tensor value of the pixel at the position; It is the first The feature tensor matrix of the region is The multi-scale feature tensor value of the pixel at ; It is the first The feature tensor matrix of the region is The multi-scale feature tensor value of the pixel at ; is the nonlinear amplification factor, which makes the change of multi-scale feature tensor value more significant; is a regularization parameter to prevent instability caused by a small denominator; is the total number of regions; Indicates that after the final partition Regions All The sum is calculated at the position, that is, the multi-scale feature tensor values ​​of the entire region are fused. Each element of the feature tensor matrix of each region is fused under nonlinear operation to generate the final partitioned The reconstruction characteristics of the region;

[0072] Furthermore, the reconstructed features of all regions after the final partition are weighted and merged to generate the comprehensive features of the dried sand particles. , ;in, is the weight coefficient, which is dynamically calculated based on the significance weight of the final partitioned area, so that the feature fusion process can reflect the internal information of the dried sand particles and strengthen the performance of the edge structure. The above nonlinear fusion method avoids the information loss caused by simple feature averaging, thereby achieving effective fusion of features in each region;

[0073] The fourth step is feature integration and comprehensive particle characterization;

[0074] Comprehensive characteristics of dried sand particles after nonlinear fusion It represents the main morphological characteristics of the dried sand particles, but further probability optimization is still needed to eliminate the influence of noise. Therefore, the comprehensive characteristics of the dried sand particles are optimized through the existing maximum likelihood estimation, and finally the final comprehensive characteristics of the dried sand particles are obtained by maximizing the logarithm of the likelihood function. .

[0075] S2. Based on the infrared sensor data, a moisture detection algorithm based on multi-dimensional spatiotemporal feature mapping and adaptive filtering is introduced to detect the moisture of the dried sand particles to obtain the moisture content of the dried sand particles; based on the moisture content of the dried sand particles and the final comprehensive characteristics of the dried sand particles, the complete feature data of the dried sand particles is obtained; multi-modal fusion and analysis are performed on the complete feature data of the dried sand particles to generate a comprehensive analysis result; based on the comprehensive analysis result, a detection report is generated.

[0076] In order to accurately obtain the moisture content of dried sand particles, a moisture detection algorithm based on multidimensional spatiotemporal feature mapping and adaptive filtering is introduced, combined with the spatiotemporal multidimensional distribution of infrared features (such as absorptivity, reflectivity, and transmittance) and the convolution filtering method, through in-depth analysis of absorptivity data, reflectivity data, and transmittance data, accurate moisture content estimation is gradually performed. The specific implementation process is as follows:

[0077] First, the absorption rate data, reflectivity data and transmittance data of the dried sand particles are obtained based on the infrared sensor data, and the infrared feature matrix containing time and space dimensions is obtained to capture the infrared features of the dried sand particles at different positions and times. The infrared feature matrix is ​​defined as ,in is the location of the dried sand particles; Indicates the acquisition time. Any element in the infrared feature matrix Contains the following:

[0078] ,

[0079] in, , , Respectively Always dry sand particles in position Absorption data, reflectivity data, and transmittance data at the location; , , Is the dried sand particles in position The rate at which the absorptivity data, reflectivity data, and transmittance data at a location change with time is used to reflect the thermal response changes of the dried sand particles; , , It is the time-accumulated absorptivity data, reflectivity data and transmittance data, which are used to capture the cumulative response of the dried sand particles in the time dimension; is the integration variable;

[0080] Further, establish the spatial distribution matrix , which is used to analyze the diffusion of infrared characteristics of dried sand particles in space, so as to more accurately characterize the moisture characteristic distribution of dried sand particles. Any element in the spatial distribution matrix The calculation method is:

[0081] ,

[0082] in, is the spatial distribution matrix in The element value at ; is the number of dried sand particles analyzed; Indicates that the dried sand particles surrounding adjacent locations; , , They are Time drying sand particles Absorption data, reflectivity data, and transmittance data at the location; is in position The surrounding space offset; Indicates the radius of the spatial offset range;

[0083] The time and space characteristics of infrared sensor data are considered simultaneously through the infrared feature matrix and the spatial distribution matrix, so that the spatiotemporal infrared response of each dried sand particle is included in the detection range, ensuring that the potential moisture distribution information can be captured in the multi-dimensional features;

[0084] Furthermore, the infrared feature matrix and the spatial distribution matrix are converted into a feature space matrix through mapping processing for unified processing. The specific formula is expressed as:

[0085] ,

[0086] in, Representing the feature space matrix, all infrared features are unified in a spatiotemporal mapping space, which not only maintains the multi-dimensional information of infrared features, but also allows for unified filtering operations;

[0087] Furthermore, in order to eliminate noise and ensure the smoothness of the feature space matrix, the feature space matrix Perform convolution filtering to obtain the feature space matrix after convolution filtering ,Convolution filtering adopts existing technology, which will not be described here;

[0088] Based on the feature space matrix after convolution filtering And the spatial distribution matrix , combined with the absorption rate data, reflectivity data and transmittance data in the infrared characteristics, the moisture content calculation formula of the dried sand particles is introduced to obtain the moisture content of the dried sand particles. The calculation formula is as follows:

[0089] ,

[0090] in, yes Always at the location The moisture content of the dried sand particles; , is the adjustment coefficient, which is determined based on simulation experiments; Indicates mean calculation.

[0091] Finally, based on the final partition, the moisture content of the dried sand particles in each area is calculated (i.e., the moisture content of the dried sand particles at all positions in the area is accumulated), and the moisture content of the dried sand particles in each area is added to the final comprehensive characteristics of the dried sand particles in the corresponding area to obtain complete characteristic data of the dried sand particles.

[0092] Furthermore, the existing multimodal data fusion method (such as the multimodal data fusion method based on Bayesian reasoning or weighted average) is used to perform multimodal fusion on the complete feature data of the dried sand particles to generate a comprehensive quality score of the dried sand particles; Bayesian reasoning can be weighted according to the weight of the influence of different features on the quality, for example, appropriate weights are given to the morphology and moisture content of the dried sand particles to ensure the accuracy of multimodal fusion; based on the comprehensive quality score of the dried sand particles and the complete feature data of the dried sand particles, the existing random forest model is used for training and verification to establish a quality classification model for the dried sand particles. After the quality classification model training is completed, the random forest model is applied to the actual detection task to perform quality detection and classification on the dried sand particles to obtain a comprehensive analysis result, which includes the classification label of the dried sand particles (qualified or unqualified) and the confidence score of the classification; according to the comprehensive analysis result, a test report containing the particle qualified rate and the proportion of unqualified particles is generated. If the proportion of unqualified particles exceeds the threshold preset according to the expert experience method, an alarm reminder is triggered and production adjustments are made.

[0093] In summary, the quality detection method of gold tailings drying sand based on machine vision was completed.

[0094] The order of the embodiments of the invention is for description only and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0095] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0096] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.

Claims

1. A method for detecting the quality of dried gold tailings sand based on machine vision, characterized in that: The following steps are involved: S1. Collect multi-angle image data and infrared sensor data of dried sand particles, and perform intelligent image processing on the multi-angle image data of dried sand particles through a multi-partition multi-scale adaptive feature reconstruction algorithm to obtain the final comprehensive features of the dried sand particles; the specific implementation steps of the multi-partition multi-scale adaptive feature reconstruction algorithm are as follows: In the first step, an adaptive partitioning method is introduced to perform image multi-partitioning processing on the multi-angle image data of the dried sand particles to obtain the final partitioning; The second step is to perform multi-scale feature extraction on the multi-angle image data of the dried sand particles in each area after the final partitioning to generate a feature tensor matrix; The third step is to perform dynamic feature reconstruction and nonlinear fusion on the feature tensor matrix generated in each area after the final partitioning to generate the comprehensive features of the dried sand particles; The fourth step is to optimize the comprehensive characteristics of the dried sand particles to obtain the final comprehensive characteristics of the dried sand particles; S2. Based on the infrared sensor data, a moisture detection algorithm based on multidimensional spatiotemporal feature mapping and adaptive filtering is introduced to detect moisture in the dried sand particles to obtain the moisture content of the dried sand particles; the specific implementation process of the moisture detection algorithm based on multidimensional spatiotemporal feature mapping and adaptive filtering is as follows: Based on the infrared sensor data, the absorption rate data, reflectivity data and transmittance data of the dried sand particles are obtained to generate an infrared feature matrix containing time and space dimensions; the spatial offset range radius is introduced to generate a spatial distribution matrix; The infrared characteristic matrix and the spatial distribution matrix are converted into a characteristic space matrix through mapping processing; Performing convolution filtering on the feature space matrix to obtain a feature space matrix after convolution filtering; Based on the feature space matrix and spatial distribution matrix after convolution filtering, the moisture content calculation formula of dried sand particles is introduced, and the moisture content of dried sand particles is detected. The specific formula is: ， in, yes Always at the location The moisture content of the dried sand particles; , is the adjustment coefficient; represents mean calculation; It is the feature space matrix after convolution filtering; , , Respectively Always dry sand particles in position Absorption data, reflectivity data and transmittance data at; is the spatial distribution matrix in The element value at ; Based on the moisture content of the dried sand particles and the final comprehensive characteristics of the dried sand particles, complete characteristic data of the dried sand particles are obtained; multi-modal fusion and analysis are performed on the complete characteristic data of the dried sand particles to generate comprehensive analysis results; based on the comprehensive analysis results, a test report is generated.

2. The gold tailings drying sand quality detection method based on machine vision according to claim 1 is characterized in that: The S1 specifically includes: In the process of implementing the adaptive partitioning method, the multi-angle image data of the dried sand particles are converted into gradient vectors, and the local mean brightness value and local grayscale variance of each pixel in the multi-angle image data of the dried sand particles are combined to assign a significance weight to each pixel in the multi-angle image data of the dried sand particles; the specific calculation formula for the significance weight of each pixel is: ， in, Represents the significance weight of each pixel; and It is the multi-angle image data of dried sand particles exist and Directional gradient; is the local mean brightness value of the current pixel; is the local grayscale variance of the current pixel; is a smoothing parameter that prevents the denominator from being zero; Based on the significance weight of each pixel point, the partition ratio parameter is introduced to perform a preliminary division on the multi-angle image data of the entire dried sand particles to obtain preliminary partitions, and the significance weights of each area after the preliminary partitions are calculated.

3. The gold tailings drying sand quality detection method based on machine vision according to claim 2 is characterized in that: The S1 specifically includes: The edges of the regions after the preliminary partitioning are closed to obtain the final partitioning; based on the final partitioning, the significance weights of the regions are recalculated and updated.

4. The gold tailings drying sand quality detection method based on machine vision according to claim 3 is characterized in that: The S1 specifically includes: Based on the multi-angle image data of the dried sand particles in each area after the final partition, the local contrast, surface roughness and average brightness are defined to generate the partition feature tensor; based on the partition feature tensor, the feature tensor matrix is ​​constructed; the generation formula of the partition feature tensor is as follows: ， in, After the final partition Areas in location The multi-scale feature tensor value at represents the partition feature tensor; Indicates the final partition The local contrast of each region; Indicates the final partition Surface roughness of the area; After the final partition The average brightness of the area; After the final partition The local edge orientation angle of the dried sand particles in each area; After the final partition The scaling exponent of the partition feature tensor for each region.

5. The gold tailings drying sand quality detection method based on machine vision according to claim 4 is characterized in that: The S1 specifically includes: The feature tensor matrix is ​​dynamically reconstructed through nonlinear transformation to generate the reconstructed features within the region; the reconstructed features in all regions after the final partition are weighted and merged to generate the comprehensive features of the dried sand particles.

6. The gold tailings drying sand quality detection method based on machine vision according to claim 1 is characterized in that: The S2 specifically includes: Based on the moisture content of the dried sand particles and combined with the final partition, the moisture content of the dried sand particles in each area is calculated, and the moisture content of the dried sand particles in each area is added to the final comprehensive features of the dried sand particles in the corresponding area to obtain the complete feature data of the dried sand particles; the complete feature data of the dried sand particles are multimodally fused to generate a comprehensive quality score of the dried sand particles; based on the comprehensive quality score of the dried sand particles and the complete feature data of the dried sand particles, the dried sand particles are quality inspected and classified to obtain a comprehensive analysis result; based on the comprehensive analysis result, a test report including the particle qualified rate and the proportion of unqualified particles is generated; when the proportion of unqualified particles exceeds the preset threshold, an alarm is triggered and production adjustments are made.

Images