A method and system for salt lake mineral distribution identification

Abstract

The present application belongs to the technical field of mineral identification, and particularly relates to a method and system for salt lake mineral distribution identification. The method comprises: collecting a visible light image of a salt lake and extracting an ROI containing potential minerals; calculating a scene complexity factor according to the complexity of texture, calculating micro gradient consistency and crystal facet reflectivity index based on local gradient and gray difference, calculating a texture saliency weighting coefficient in combination with an environmental noise base; weighting and fusing texture and shape features by using the texture saliency weighting coefficient, and constructing an anisotropic hybrid feature vector in combination with a color moment feature; and finally inputting the classifier to obtain a mineral category and generate a distribution map. The present application can dynamically adjust the feature weight according to the physical properties of minerals, and improve the accuracy and robustness of identification in a complex environment.

Description

Technical Field

[0001] This invention relates to the field of mineral identification technology. More specifically, this invention relates to a method and system for identifying the distribution of minerals in salt lakes. Background Technology

[0002] In the exploration and development of salt lake resources, high-resolution industrial cameras mounted on drones or fixed supports are typically used to collect visible light images of the salt lake surface. Through image processing and pattern recognition technologies, the identification and location of minerals such as carnallite and potash are automated, replacing inefficient manual inspections. Mineral distribution information can provide data support for the optimization and control of salt field brine drying processes and the operation of subsequent harvesting equipment, which is crucial for improving the comprehensive utilization rate of salt lake resources.

[0003] However, the harsh and variable environment of salt lakes, including drastic fluctuations in light intensity, random disturbances in water ripples, and inherent thermal noise in photographic equipment, all affect the signal-to-noise ratio of images. Existing technologies typically employ algorithms such as Otsu for image segmentation, utilize gradient operators to calculate edge features, and combine texture and shape features with classifiers such as support vector machines for recognition.

[0004] The Otsu algorithm and support vector machine classifiers mentioned above have significant shortcomings in salt lake scenarios: On the one hand, traditional image segmentation and gradient calculation often rely on fixed thresholds or single global statistical features, lacking the ability to adaptively perceive environmental noise levels such as wave interference and the degree of scene texture disorder. This may lead to misjudging light waves as minerals under strong interference or missing targets under low light. On the other hand, in the feature extraction and fusion stage, traditional methods usually use fixed weights to splice texture and shape features, ignoring the significant differences in microscopic physical properties of different types of minerals. They cannot dynamically adjust feature weights according to changes in mineral morphology, making it difficult to accurately identify minerals of different morphologies. Summary of the Invention

[0005] To address the aforementioned technical problems of poor recognition adaptability and low accuracy, the present invention provides solutions in the following aspects.

[0006] In a first aspect, the present invention provides a method for identifying the distribution of minerals in a salt lake, comprising:

[0007] Raw visible light images of the salt lake surface are acquired using a high-resolution industrial camera. These raw visible light images are segmented to extract Regions of Interest (ROIs) containing potential minerals. A scene complexity factor is calculated based on the texture complexity of the raw visible light image. Microscopic gradient consistency and crystal facet reflectivity indices are calculated based on local neighborhood gradient information and grayscale differences of pixels within the ROI. Furthermore, the texture salience weighting coefficient for each pixel within the ROI is calculated by combining the drastic grayscale changes in the region outside the ROI in the raw visible light image. Texture features, shape features, and color moment features of the ROI are extracted. The texture and shape features are weighted and fused based on the mean of the texture salience weighting coefficients of all pixels within the ROI, and combined with the color moment features to obtain an anisotropic mixture feature vector of the ROI. This anisotropic mixture feature vector is input into a pre-defined classifier for classification to obtain mineral category information within the ROI. Coordinate mapping is then performed based on camera parameters to generate a mineral distribution map.

[0008] This invention perceives the level of interference from the environmental background by evaluating the texture complexity of the original visible light image. It combines the local micro-gradient information and grayscale differences of pixels within the ROI to obtain the micro-gradient consistency and crystal facet reflectivity index, which reflect the physical properties of mineral crystals. By combining the macro-environmental state with the micro-physical structure features, it obtains a texture saliency weighting coefficient that can dynamically reflect the signal-to-noise ratio of mineral features. Then, it uses the texture saliency weighting coefficient to perform weighted fusion of texture features and shape features, so that when minerals exhibit strong crystalline characteristics, the emphasis is on texture description, while when they exhibit an amorphous accumulation state, the emphasis is on shape description. At the same time, color moment features are combined to supplement chemical composition information, thereby restoring the essential physical properties of different types of minerals. By identifying them through a classifier and combining them with camera parameters for geospatial mapping, the risk of misjudgment under complex lighting changes or water surface ripple interference is reduced, and the accuracy and robustness of identifying the distribution of different mineral forms in dynamically changing salt lake scenes are improved.

[0009] Preferably, the process involves acquiring raw visible light images of the salt lake surface using a high-resolution industrial camera; segmenting the raw visible light images to extract Regions of Interest (ROIs) containing potential minerals, including:

[0010] High-resolution industrial cameras mounted on drones or fixed supports are used to acquire visible light images of the salt lake surface to obtain raw visible light images. The raw visible light images are then subjected to adaptive threshold segmentation using the Otsu algorithm to generate a preliminary mask. The preliminary mask is then subjected to a dilation-erosion operation to remove background areas and extract ROIs containing potential minerals.

[0011] This invention generates a preliminary mask through an adaptive threshold segmentation algorithm and further performs morphological operations of dilation followed by erosion. This not only removes irrelevant background brine regions but also utilizes the characteristics of morphological closing operations to connect the minute physical gaps between mineral crystals, repairing any holes and fractures in the mask. This results in a more complete ROI with smoother edges, reducing computational biases caused by background noise or broken target regions during feature extraction.

[0012] Preferably, obtaining the scene complexity factor includes:

[0013] Obtain the grayscale entropy of the original visible light image, and add 1 to twice the ratio of the grayscale entropy of the original visible light image to the maximum theoretical entropy value of the 8-bit grayscale image to obtain the scene complexity factor of the original visible light image.

[0014] Preferably, the micro-gradient consistency satisfies the following relationship:

[0015] ;

[0016] In the formula, Indicates the first ROI within the ROI Microscopic gradient consistency of each pixel; Indicates the first ROI within the ROI Within the local neighborhood of the nth pixel Gradient direction angle of each pixel; Indicates the first ROI within the ROI The total number of pixels in the local neighborhood of a pixel; Represents the cosine function; This represents the sine function.

[0017] This invention analyzes the gradient direction angle distribution of local neighborhoods of pixels within a Region of Interest (ROI), decomposes and synthesizes the gradient direction angle using trigonometric functions, and calculates the micro-gradient consistency. Micro-gradient consistency can capture the degree of order of the microstructure of mineral surfaces. When the gradient direction heights are similar in a local area, the values ​​approach a specific high value, corresponding to mineral crystal facets with regular geometric shapes. When the gradient directions are disordered, the values ​​approach zero, corresponding to amorphous mud or random noise. This reduces recognition errors when relying on brightness differences in uneven lighting or when the mineral color is similar to the background, and enhances the ability to identify minerals with regular geometric structures.

[0018] Preferably, obtaining the crystal facet reflectivity index includes:

[0019] The natural logarithm of the ratio of the gray value of any pixel within the ROI to the mean gray value of the ROI, plus the natural constant, is multiplied by the micro-gradient consistency to obtain the crystal facet reflectivity index of the pixel.

[0020] This invention combines microscopic gradient consistency and grayscale statistical characteristics to obtain the reflectivity index of crystal facets. By calculating the ratio of pixel grayscale value to the mean of the region and performing a logarithmic transformation using the natural constant, the contrast of the highlight area is enhanced. At the same time, by using the relative grayscale ratio instead of the absolute grayscale value, the algorithm has better adaptability to changes in ambient light. It can lock crystal areas that have both regular gradient structure and relatively high brightness characteristics even when the intensity of surface light fluctuates, reducing the possibility of missed detection due to cloudy days or shadows, and the possibility of misjudging water surface light as minerals due to simple brightness.

[0021] Preferably, the texture saliency weighting coefficients satisfy the following relationship:

[0022] ;

[0023] In the formula, Indicates the first ROI Texture saliency weighting coefficients for each pixel; Indicates the first ROI The reflectivity index of the crystal facets of each pixel; Indicates the first ROI Microscopic gradient consistency of each pixel; The scene complexity factor representing the original visible light image; This represents the standard deviation of gray levels in the region excluding the ROI in the original visible light image.

[0024] This invention obtains texture saliency weighting coefficients by considering the environmental noise floor, scene complexity factor, and the physical properties of the minerals themselves. The texture saliency weighting coefficients use the environmental noise floor as the criterion and adjust the stringency of the screening by the scene complexity factor. In scenes with strong background interference or complex textures, low signal-to-noise ratio feature responses are automatically suppressed through exponential operations, while retaining the weights of pixels with clear structures and significant reflective features. This allows the subsequent feature fusion process to dynamically adapt to the current imaging environment, reducing the interference of environmental noise on feature expression and thus maintaining recognition stability under different wind and wave conditions.

[0025] Preferably, the anisotropic hybrid feature vector satisfies the following relation:

[0026] ;

[0027] In the formula, Represents anisotropic mixed eigenvectors; This represents the average of the texture saliency weighting coefficients for all pixels within the ROI; Texture features representing the gray-level co-occurrence matrix of the ROI; The shape characteristics of the Hu invariant moments in the ROI region; Represents the color moment characteristics of the ROI; This indicates the concatenation of vectors.

[0028] Preferably, the step of inputting the anisotropic mixture feature vector of the ROI into a preset classifier for classification to obtain the mineral category information within the ROI, and generating a mineral distribution map based on coordinate mapping according to camera parameters, includes:

[0029] The anisotropic hybrid feature vector is input into a support vector machine classifier pre-trained based on sample data, and the RBF kernel function is used for mapping and classification to output the mineral category.

[0030] Using the pre-acquired intrinsic and extrinsic parameter matrices of the camera, the coordinates of the mineral pixels identified in the image coordinate system are back-projected and mapped to the geographic coordinate system of the salt lake, generating a mineral distribution map containing geographic location information.

[0031] Preferably, obtaining the intrinsic and extrinsic parameter matrices of the camera includes:

[0032] The camera's intrinsic parameter matrix is ​​obtained in advance by calibrating the checkerboard calibration board using the Zhang Zhengyou calibration method; the camera's extrinsic parameter matrix is ​​obtained from the acquisition platform: if it is a drone, it is calculated in real time using the shooting position and attitude recorded by the airborne POS system; if it is a fixed-point support, it is calculated in reverse using the PnP algorithm based on the known ground control points.

[0033] Secondly, the present invention provides a system for identifying the distribution of minerals in salt lakes, including a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the aforementioned method for identifying the distribution of minerals in salt lakes is implemented.

[0034] By adopting the above technical solution, a computer program for identifying the distribution of minerals in salt lakes is generated and stored in a memory so that it can be loaded and executed by a processor. A terminal device can then be made based on the memory and processor for convenient use.

[0035] The beneficial effects of this invention are as follows: Addressing the identification challenges posed by variable natural lighting and significant differences in mineral physical morphology during salt lake mining, this invention assesses the texture complexity of the entire image to perceive the interference level of external wind and waves on the water surface texture. It also combines local micro-gradient information and grayscale differences to obtain a reflectivity index reflecting crystal physical properties. This allows for the identification of effective areas with high signal-to-noise ratios while suppressing environmental background noise. Furthermore, this invention generates a texture saliency weighting coefficient through the fusion of multiple features, automatically emphasizing texture feature description when minerals exhibit regular crystalline characteristics and shape feature description when they present amorphous powder accumulation. This reduces the risk of misjudgment by traditional static feature extraction methods when dealing with mixed distributions of carnallite crystals and potassium halite powders. It also improves the robustness of the algorithm under complex conditions such as strong light reflection, cloudy skies, and water surface ripples, achieving a mapping from surface visual data to mineral geographical distribution information, providing a reference decision-making basis for salt lake resource mining operations. Attached Figure Description

[0036] Figure 1 This schematically illustrates a flowchart of a method for identifying the distribution of minerals in a salt lake according to the present invention;

[0037] Figure 2 A schematic diagram of the preliminary mask is shown.

[0038] Figure 3 A schematic diagram illustrating an ROI containing potential minerals;

[0039] Figure 4 A statistical histogram illustrating the distribution of consistent micro-gradient values ​​within an ROI is shown.

[0040] Figure 5 A statistical histogram illustrating the distribution of reflectance index values ​​of crystal facets within the ROI is shown schematically.

[0041] Figure 6 This diagram illustrates the numerical distribution of the weighted coefficients for texture saliency within a Region of Interest (ROI). Detailed Implementation

[0042] This invention discloses a method for identifying the distribution of minerals in salt lakes, referring to... Figure 1 This includes steps S100-S500:

[0043] S100. Acquire raw visible light images of the salt lake surface using a high-resolution industrial camera; segment the raw visible light images and extract ROIs containing potential minerals.

[0044] It should be noted that raw visible light images often contain a large number of irrelevant background brine regions, and there may be tiny physical gaps between mineral crystals. To reduce computational load and background interference, and to avoid invalid backgrounds participating in subsequent complex feature calculations, image purification is necessary. Furthermore, the initial mask generated by the Otsu algorithm often contains holes and breaks; if used directly, it will cause the mineral regions to break. Therefore, this invention combines morphological operations to repair the mask, focusing on the effective regions.

[0045] Specifically, a high-resolution industrial camera mounted on a drone or fixed support is used to acquire visible light images of the salt lake surface, obtaining raw visible light images. These raw visible light images are then converted to grayscale images, and adaptive threshold segmentation is performed to generate a preliminary mask. The preliminary mask is then repaired by dilation followed by erosion to remove background areas, extracting Regions of Interest (ROIs) containing potential minerals. The ROIs are then filtered and denoised to remove high-frequency random noise. For example, the adaptive threshold segmentation uses the Otsu algorithm, and the denoising uses Gaussian filtering. Both the Otsu algorithm and Gaussian filtering are existing technologies and will not be elaborated upon here.

[0046] For example, such as Figure 2 It is a preliminary mask generated by adaptive threshold segmentation. Figure 3 This represents the ROI containing potential minerals extracted after expansion corrosion operation and Gaussian filtering for noise reduction.

[0047] It should be noted that the high-frequency random noise still contained in the partitioned ROI can lead to deviations in feature calculation; therefore, this invention performs Gaussian filtering on the ROI to eliminate unstructured noise interference.

[0048] Thus, the ROI containing potential minerals is obtained.

[0049] S200. Calculate the scene complexity factor based on the texture complexity of the original visible light image; calculate the micro-gradient consistency of the pixels based on the local neighborhood gradient information of each pixel within the ROI; and calculate the crystal facet reflectance index of the pixels by combining the micro-gradient consistency with the grayscale difference of the pixels.

[0050] It should be noted that the texture of the salt lake surface is greatly affected by wind speed and precipitation. When the wind is calm, the water surface is smooth, but when there is wind, ripples will appear. If fixed algorithm parameters are used, it is difficult to adapt to both calm and rippled water surfaces at the same time, which can easily lead to misjudgment. Therefore, this invention introduces a scene complexity factor to characterize the degree of disorder of the current image texture. By calculating the grayscale information entropy of the entire image, it is possible to determine from a macroscopic perspective whether the current scene is simple or complex, thereby automatically tightening the screening criteria in complex scenes and preventing ripples from being misjudged as minerals.

[0051] Specifically, the scene complexity factor is calculated based on the complexity of the texture in the original visible light image, including:

[0052] Obtain the grayscale entropy of the original visible light image;

[0053] The scenario complexity factor satisfies the following relationship:

[0054] ;

[0055] In the formula, The scene complexity factor representing the original visible light image; The grayscale entropy represents the grayscale information of the original visible light image; This represents the maximum theoretical entropy value of an 8-bit grayscale image.

[0056] In this relation, The normalized index representing grayscale information entropy reflects the degree of disorder in image texture; Linearly map the normalized index to interval; Entropy is used to measure the degree of disorder in a scene; the more complex the scene, the higher the entropy. A value close to 3 indicates more interference textures, requiring stricter screening of the crystal structure; conversely, a lower value indicates a simpler scene. If the value approaches 1, the screening requirements will be relaxed accordingly.

[0057] Thus, the scene complexity factor of the original visible light image was obtained.

[0058] It should be noted that, since mineral crystals have regular geometric facets, the normal direction of the mineral crystal facets is highly consistent in a local area, while the gradient direction of ripples or amorphous mud and sand is randomly distributed. Therefore, this invention distinguishes mineral crystals with regular geometric shapes from chaotic background interference by calculating the composite modulus of the gradient direction angles of all pixels in the local neighborhood through structural features.

[0059] Preferably, the micro-gradient consistency of all pixels within the ROI is calculated based on local texture structure features, including:

[0060] Perform convolution operations on the ROI to obtain the gradient information of each pixel, and then use the gradient of the pixel within the ROI as the basis for the gradient. A local neighborhood is defined centered on a given pixel. Based on the gradient information of the given pixel, the gradient direction angles of all pixels within the local neighborhood are calculated. For example, the size of the local neighborhood can be set to... The convolution operation is the Sobel operator, which is existing technology and will not be described in detail here.

[0061] ROI within The micro-gradient consistency of each pixel satisfies the following relationship:

[0062] ;

[0063] In the formula, Indicates the first ROI Microscopic gradient consistency of each pixel; Indicates the first ROI Within the local neighborhood of the nth pixel Gradient direction angle of each pixel; Indicates the first ROI The total number of pixels in the local neighborhood of a pixel; Represents the cosine function; This represents the sine function.

[0064] In this relationship, the double-angle treatment eliminates the ambiguity of the gradient angle. and These represent the decomposition of all gradient direction angles within the local neighborhood onto an orthogonal coordinate system and their accumulation; This represents the combined magnitude of all gradient direction angles within the local neighborhood in double-angle space; if the gradient directions within the local neighborhood are disordered, such as noise, then... and The gradients cancel each other out, and the combined modulus approaches 0; if the gradient directions are highly consistent, such as in a crystal facet, the combined modulus approaches 0. .when When it approaches 1, it indicates that the first... The local neighborhood of each pixel has a highly uniform gradient direction, corresponding to regular crystal facets; when When it approaches 0, it represents the first The local neighborhood texture of each pixel is messy, corresponding to amorphous powder or noise.

[0065] For example, such as Figure 4 This is a statistical histogram of the distribution of micro-gradient consistency values ​​within the ROI. The bar regions represent the distribution of the number of pixels within different micro-gradient consistency value intervals; the dashed line in the middle represents the mean position of the micro-gradient consistency of all pixels in this region, and the dashed line on the right represents the position of the maximum value of the micro-gradient consistency.

[0066] At this point, the micro-gradient consistency of all pixels has been achieved.

[0067] It should be noted that, considering the specular reflection characteristics of the facet of a mineral crystal, the crystal surface is smooth and its reflected light intensity is higher than that of the surrounding diffusely reflected background. Therefore, this invention calculates the logarithmic gain of the pixel gray value relative to the average gray value of the ROI region, thereby adapting to changes in light intensity under different weather conditions. In strong light or cloudy conditions, it can lock the crystal area that is brighter than the background.

[0068] Preferably, the crystal facet reflectance index of all pixels within the ROI is calculated based on the grayscale difference, and the ROI is... The reflectivity index of the crystal facets of each pixel satisfies the following relationship:

[0069] ;

[0070] In the formula, Indicates the first ROI within the ROI The reflectivity index of the crystal facets of each pixel; Indicates the first ROI within the ROI Microscopic gradient consistency of each pixel; Indicates the first ROI within the ROI The grayscale value of each pixel; This represents the average grayscale value within the ROI; Represents the natural constant; This represents the natural logarithm function.

[0071] In this relation, Indicates the first The degree of brightness of each pixel relative to the global background; The larger the value, the higher the value. The gray value of each pixel is higher than the gray value of the mean gray value, exhibiting strong specular reflection characteristics, corresponding to the highly reflective facets of mineral crystals; The smaller the value, the higher the value. The gray values ​​of individual pixels are close to or lower than the gray value average, exhibiting a dim diffuse reflection characteristic, corresponding to non-reflective brine backgrounds or shadow areas. This enables adaptive compensation for different weather lighting intensities, eliminating invalid areas that, although having high micro-gradient consistency, have low brightness.

[0072] For example, such as Figure 5 This is a statistical histogram of the distribution of crystal facet reflectance index values ​​within the ROI. The bar regions represent the distribution of the number of pixels within different crystal facet reflectance index value ranges; the dashed lines represent the mean position of the crystal facet reflectance index of all pixels within this region.

[0073] At this point, the crystal facet reflectance index of all pixels has been obtained.

[0074] S300. Calculate the texture saliency weighting coefficient of each pixel within the ROI based on the crystal facet reflectivity index, micro-gradient consistency, the degree of grayscale change in the region excluding the ROI in the original visible light image, and the scene complexity factor.

[0075] It should be noted that, considering that the light generated in strong wind and wave environments may exhibit bright, crystal-like features in some areas, in order to further improve the anti-interference capability of recognition, this invention combines the environmental noise baseline and the scene complexity factor to construct a texture saliency weighting coefficient. This invention uses the standard deviation of the pixel grayscale in the background area as a noise benchmark and the scene complexity as an exponential constraint to apply more stringent verification pressure to suspected features in high-noise environments; thereby automatically suppressing false feature responses caused by environmental noise, which helps to make high-weight feature points correspond to real minerals with high signal-to-noise ratio and clear structure.

[0076] Specifically, the texture saliency weighting coefficients for all pixels within the ROI are calculated based on the joint constraints of environmental noise and scene complexity, including:

[0077] Extract the grayscale values ​​of all pixels within the region excluding the ROI in the original visible light image. Normalize the pixel grayscale values ​​to the [0,1] interval by dividing them by the maximum theoretical grayscale value, and then calculate the grayscale standard deviation. For example, the maximum theoretical grayscale value for an 8-bit grayscale image is 255.

[0078] ROI within The weighted coefficients of texture saliency for each pixel satisfy the following relationship:

[0079] ;

[0080] In the formula, Indicates the first ROI Texture saliency weighting coefficients for each pixel; Indicates the first ROI The reflectivity index of the crystal facets of each pixel; Indicates the first ROI Microscopic gradient consistency of each pixel; The scene complexity factor representing the original visible light image; This represents the standard deviation of gray levels in the region excluding the ROI in the original visible light image.

[0081] In this relation, As an inflection point of the function, this utilizes the statistically common 3σ criterion; only when the 3σ-th inflection point is reached... When the reflectivity index of a single pixel's crystal facet exceeds three times that of the ambient noise substrate, The value will rise and approach 1; in environments with high noise or low light, the algorithm will automatically shift the inflection point to the right, raising the judgment threshold and preventing the light wave from being misjudged as a mineral. This indicates the stringency of the adaptive structure selection. A higher value indicates that the pixel maintains a high degree of gradient consistency even in complex scenes. The algorithm will apply higher screening pressure through the exponential effect, retaining pixels with high signal-to-noise ratios; conversely, a lower value indicates a lower gradient consistency. The smaller the value, the simpler the scene or the less obvious the texture structure. The algorithm automatically relaxes the screening criteria to avoid missing valid features.

[0082] For example, such as Figure 6 This is a schematic diagram of the numerical distribution of texture saliency weighting coefficients within the ROI. The horizontal axis represents the magnitude of the texture saliency weighting coefficient, which ranges from [0,1]. The vertical axis represents the number of pixels within the corresponding range of texture saliency weighting coefficients. The dashed line in the diagram represents the position of the arithmetic mean of the texture saliency weighting coefficients of all pixels within the ROI region.

[0083] At this point, the texture saliency weighting coefficients for all pixels have been obtained.

[0084] S400. Extract the texture features, shape features, and color moment features of the ROI. Based on the mean of the texture saliency weighting coefficients of all pixels in the ROI, perform weighted fusion of the texture features and shape features, and combine them with the color moment features to obtain the anisotropic mixed feature vector of the ROI.

[0085] It should be noted that this invention fuses features based on the significant differences in the physical morphology of different types of salt lake minerals. For example, carnallite typically exhibits a distinct crystalline structure, and its texture features are highly distinguishable; while potash often appears as powder or amorphous aggregates, making its shape and contour features more crucial. This invention utilizes a texture saliency weighting coefficient as a dynamic weight to automatically adjust the proportion of texture and shape features in the mixed feature vector. When the ROI exhibits strong crystalline characteristics, this invention emphasizes texture features; when the ROI exhibits amorphous aggregates, this invention emphasizes shape features, thereby ensuring that the feature vector input to the classifier contains the essential physical properties of the mineral crystals.

[0086] Specifically, this invention constructs an anisotropic hybrid feature vector based on the adaptive saliency difference between mineral crystal properties and texture and shape features, including:

[0087] Calculate the arithmetic mean of the texture saliency weighted coefficients for all pixels within the ROI.

[0088] Extract the texture features of the gray-level co-occurrence matrix of the ROI; extract the shape features of the Hu invariant moments of the ROI; extract the color moment features of the ROI.

[0089] The anisotropic mixture eigenvectors of a ROI satisfy the following relationship:

[0090] ;

[0091] In the formula, Represents anisotropic mixed eigenvectors; This represents the average of the texture saliency weighting coefficients for all pixels within the ROI; Texture features representing the gray-level co-occurrence matrix of the ROI; The shape characteristics of the Hu invariant moments in the ROI region; Represents the color moment characteristics of the ROI; This indicates the concatenation of vectors.

[0092] In this relation, Represents the texture component weighted by crystal saliency; Represents the shape component weighted by the saliency of the amorphous material; when When the value approaches 1, corresponding to a strong crystalline structure such as carnallite, the anisotropic mixing feature vector automatically tilts towards the texture features of the gray-level co-occurrence matrix, dominating the recognition process and ignoring irregular stacking shapes; when When the value approaches 0, corresponding to powdery or amorphous forms, such as potassium halite, the anisotropic mixed feature vector automatically switches to be dominated by the shape features of Hu invariant moments. This dynamic weight allocation mechanism makes the classifier tend to focus on the main physical features. The color moment feature represents the ROI. The larger the color moment feature, the stronger the color distribution contrast within the ROI. For example, carnallite is red due to iron impurities, corresponding to a mineral category with specific chemical composition. The smaller the color moment feature, the more uniform or colorless and transparent the color within the ROI is, such as pure halite. This serves as a chemical composition verification dimension independent of the texture features of the gray-level co-occurrence matrix and the shape features of the Hu invariant moment, preventing misjudgments caused by morphological similarity.

[0093] Thus, the anisotropic mixed feature vector is obtained.

[0094] S500: Input the anisotropic mixed feature vector of the ROI into the preset classifier for classification, obtain the mineral category information within the ROI, and generate a mineral distribution map based on the coordinate mapping of the camera parameters.

[0095] It should be noted that, because this invention enhances and reduces noise at the physical level, the separability of different mineral categories in the feature space is improved. Therefore, this invention uses a support vector machine with low computational complexity as a classifier to achieve mineral recognition, meeting the real-time requirements of embedded devices. In addition, this invention combines the intrinsic and extrinsic parameter matrices of the camera to convert pixel coordinates on the image plane into geospatial coordinates, thereby directly converting the visual recognition results into a mineral geographic distribution map to guide production operations, realizing the mapping from image data to spatial location information.

[0096] Specifically, the anisotropic mixed feature vector is input into a support vector machine classifier pre-trained based on sample data, and the RBF kernel function is used for mapping and classification to output mineral categories, such as carnallite, potassium halite, and hydrated magnesium chlorite.

[0097] Using pre-acquired intrinsic and extrinsic parameter matrices of the camera, the coordinates of mineral pixels identified in the image coordinate system are back-projected and mapped to the geographic coordinate system of the salt lake, generating a mineral distribution map containing geographic location information. It should be noted that the camera's intrinsic parameter matrix is ​​pre-calibrated using the Zhang Zhengyou calibration method on a checkerboard calibration board; the extrinsic parameter matrix is ​​obtained based on the data acquisition platform: if it is a drone, it is calculated in real-time using the shooting position and attitude recorded by the onboard POS system; if it is a fixed-point support, the camera's fixed pose is calculated back-by-back using known ground control points and the PnP algorithm. The Zhang Zhengyou calibration method and the PnP algorithm are existing technologies and will not be elaborated upon here.

[0098] This completes the identification of mineral distribution in the salt lake.

[0099] This invention also discloses a system for identifying the distribution of minerals in salt lakes, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement a method for identifying the distribution of minerals in salt lakes according to the present invention.

[0100] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0101] While this specification has shown and described numerous embodiments of the invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention. It should be understood that various alternatives to the embodiments of the invention described herein may be employed in the practice of this invention.

Claims

1. A method for salt lake mineral distribution identification, characterized in that, The method comprises the following steps: acquiring an original visible light image of a salt lake surface by using a high-resolution industrial camera; segmenting the original visible light image to extract a ROI containing potential minerals; calculating a scene complexity factor according to the texture complexity of the original visible light image, which comprises the following steps: obtaining the gray information entropy of the original visible light image, adding 1 and twice the ratio of the gray information entropy of the original visible light image to the maximum theoretical entropy value of an 8-bit gray image, to obtain the scene complexity factor of the original visible light image; The micro gradient consistency and the crystal facet reflectance index are calculated based on the local neighborhood gradient information and the gray difference of the pixel points in the ROI, and then the texture saliency weighting coefficient of each pixel point in the ROI is calculated by combining the gray variation degree in the region except the ROI in the original visible light image, satisfying the relationship: , the texture saliency weighting coefficient of the i-th pixel point in the ROI, the texture saliency weighting coefficient of the i-th pixel point in the ROI, the crystal facet reflectance index of the i-th pixel point in the ROI, the crystal facet reflectance index of the i-th pixel point in the ROI, the micro gradient consistency of the i-th pixel point in the ROI, the micro gradient consistency of the i-th pixel point in the ROI, the scene complexity factor of the original visible light image, the gray standard deviation in the region except the ROI in the original visible light image; Micro-gradient consistency satisfies the following relation: , Indicates the first ROI within the ROI Within the local neighborhood of the nth pixel Gradient direction angle of each pixel Indicates the first ROI within the ROI The total number of pixels in the local neighborhood of a given pixel. Represents the cosine function. Represents the sine function; obtaining a crystal facet reflectivity index, which comprises the following steps: multiplying the natural logarithm value of the ratio of the gray value of an arbitrary pixel point in the ROI to the average gray value of the ROI by a micro gradient consistency, to obtain the crystal facet reflectivity index of the pixel point; extracting texture features, shape features and color moment features of the ROI, weighting and fusing the texture features and the shape features according to the average value of the texture saliency weighting coefficients of all pixel points in the ROI, and combining the color moment features to obtain an anisotropic hybrid feature vector of the ROI; inputting the anisotropic hybrid feature vector of the ROI into a preset classifier to classify the ROI and obtain the category information of the minerals in the ROI, and performing coordinate mapping based on the camera parameters to generate a mineral distribution map.

2. The method for identifying mineral distribution of salt lake according to claim 1, characterized in that, The method comprises the following steps: acquiring a visible light image of a salt lake surface by using a high-resolution industrial camera mounted on a UAV or a fixed-point support, to obtain an original visible light image; performing adaptive threshold segmentation on the original visible light image by using an Otsu algorithm to generate a preliminary mask; performing a first inflation and then a corrosion operation on the preliminary mask to remove the background area and extract a ROI containing potential minerals.

3. The method for identifying mineral distribution of salt lake according to claim 1, characterized in that, The anisotropic hybrid feature vector satisfies the following relationship: ； In the formula, denotes an anisotropic mixed feature vector; denotes the average value of the texture saliency weighting coefficients of all pixel points in the ROI; denotes the texture feature of the gray level co-occurrence matrix of the ROI; denotes the shape feature of the Hu invariant moment of the ROI region; denotes the color moment feature of the ROI; denotes the splicing of vectors.

4. The method for salt lake mineral distribution identification according to claim 1, characterized in that, The method comprises the following steps: inputting the anisotropic hybrid feature vector into a support vector machine classifier trained based on sample data in advance, mapping and classifying by using an RBF kernel function, and outputting the mineral category; projecting and mapping the mineral pixel coordinates recognized in the image coordinate system to a geographical coordinate system of the salt lake in reverse by using the intrinsic matrix and the extrinsic matrix of the camera obtained in advance, to generate a mineral distribution map containing geographical position information.

5. The method for salt lake mineral distribution identification according to claim 4, characterized in that, The method comprises the following steps: The intrinsic matrix of the camera is obtained by Zhang Zhengyou calibration method by calibrating a checkerboard calibration plate in advance; the extrinsic matrix of the camera is obtained according to the acquisition platform: if it is a UAV, the shooting position and attitude recorded by the onboard POS system are used for real-time calculation; if it is a fixed-point support, the fixed pose of the camera is inversely solved by using the known control points on the ground through the PnP algorithm.

6. A system for salt lake mineral distribution identification, characterized in that, The method comprises the following steps: A processor and a memory, the memory storing computer program instructions which, when executed by the processor, implement a method for identifying a distribution of minerals in a salt lake according to any one of claims 1-5.

Images