CN-122023927-A - Gangue classification and identification method and system based on image vision

CN122023927ACN 122023927 ACN122023927 ACN 122023927ACN-122023927-A

Abstract

The invention relates to the technical field of gangue identification, in particular to a gangue classification identification method and system based on image vision, comprising the steps of acquiring a mixed image of coal and gangue by adopting a camera array, and dividing a plurality of candidate areas; the method comprises the steps of determining a mixing state of coal and gangue at each candidate region based on an overlapping position of each candidate region, distinguishing a blocking region and a non-blocking region under the mixing of the gangue by using characteristic loss, synchronizing the characteristic loss of the blocking region into an adjacent non-blocking region based on continuous frame images of the blocking region, determining a spatial similarity threshold of the adjacent non-blocking region, carrying out multi-target cooperative processing on the spatial position of the blocking region, unifying classification labels of the gangue in each candidate region, defining region characteristics of each candidate region according to a value range of the classification labels, and taking the region characteristics as synchronous output data. The accuracy and the efficiency of coal gangue identification are realized.

Inventors

LIU YUEMU
Wen Jiangmin
QIAO JIANG
ZOU JIE
MENG YI
LI XUE

Assignees

国能榆林能源有限责任公司

Dates

Publication Date: 20260512
Application Date: 20260204

Claims (10)

1. The gangue classification and identification method based on image vision is characterized by comprising the following steps of: S1, acquiring a mixed image of coal and gangue by adopting a camera array, and dividing a plurality of candidate areas by taking position coordinates of any element in the mixed image as a basis; S2, determining a mixing state of coal and gangue at the overlapping position based on the overlapping position of each candidate region, quantifying a characteristic loss value of each candidate region under overlapping shielding according to the mixing state, and distinguishing a shielding region and a non-shielding region under mixing of the gangue by using the characteristic loss; S3, based on continuous frame images of the shielding areas, carrying out grid fitting on the shielding areas in the current frame and the adjacent non-shielding areas, synchronizing the feature loss of the shielding areas into the adjacent non-shielding areas, and determining a spatial similarity threshold of the adjacent non-shielding areas; s4, carrying out multi-target cooperative processing according to the space positions of the shielding areas, and unifying classification labels of the coal gangue in each candidate area by combining the space similarity threshold values between adjacent non-shielding areas; and S5, defining the region characteristics of each candidate region by using the value range of the classification label, and taking the region characteristics as the data synchronously output.
2. The image vision-based gangue classification and identification method as claimed in claim 1, wherein the implementation manner of the step S1 comprises: S11, performing time stamp alignment on the acquired mixed images according to the acquired time sequence, and determining the position of the same element when outputting after each acquisition; S12, dividing a plurality of areas in a gridding mode according to the positions of the same elements, and determining that each area at least contains one element; and S13, mapping the positions of the elements and the pixel coordinates of the mixed image, and setting candidate areas for the identified coal or gangue.
3. The image vision-based gangue classification and identification method as claimed in claim 2, wherein when the candidate region is set in step S13, implementation manner further comprises: S131, the size, the height and the pixel value of the coal gangue during identification are called, and an original feature pool of the coal gangue is constructed; s132, taking pixel points in the original feature pool as data points, and carrying out clustering processing according to pixel values corresponding to each data point to generate a semantically segmented label graph; S133, selecting contents corresponding to coal or gangue in the tag map, and aggregating the tag map with a similarity measure to form a preliminary candidate region; S134, screening the preliminary candidate regions by acquiring size constraint and overlapping degree between the preliminary candidate regions, and determining the output candidate regions.
4. The image vision-based gangue classification recognition method as claimed in claim 3, wherein the implementation manner of step S134 further comprises: performing size filtering on all the preliminary candidate areas, and filtering the preliminary candidate areas which do not accord with the description of the tag map; And calculating the overlapping degree of the filtered preliminary candidate regions, and screening the preliminary candidate regions according to the coordinate values mapped by the preliminary candidate regions to obtain output candidate regions.
5. The image vision-based gangue classification and identification method as claimed in claim 1, wherein the implementation manner of the step S2 comprises: S21, for each overlapping position, determining the characteristic proportion of the coal gangue in the overlapping position by using the characteristic value of each overlapping position and the characteristic clustering center of the coal gangue on the whole mixed image; S22, dividing the mixing state of each overlapping position based on the characteristic proportion of the coal gangue; s23, based on the mixed state of the current overlapping position, calculating loss values under corresponding dimensions of colors, textures and shapes, and obtaining characteristic loss of the corresponding overlapping position; and S24, regarding a part with the characteristic loss larger than the threshold value as a shielding area in the overlapped position, regarding the rest of the overlapped position which is not overlapped or the overlapping position with the loss not exceeding the threshold value as a non-shielding area, and reserving the label of the mixed state.
6. The image vision-based gangue classification and identification method as claimed in claim 5, wherein the implementation manner of step S23 further comprises: The conditional probability is calculated for each candidate region in the mixed state, and the rationality of each mixed state configuration is verified by the probability distribution of each candidate region.
7. The image vision-based gangue classification and identification method as claimed in claim 1, wherein the implementation manner of the step S3 comprises: s31, calling continuous frame images corresponding to the shielding region, determining the position of the same gangue target in the continuous frame images based on the position coordinates of the shielding region and the movement track of the gangue, and screening out a reference frame containing the gangue target and in a non-shielding region state; S32, checking the distance between the non-shielding area and the shielding area of the current frame, and determining the adjacent non-shielding area of the current frame; S33, dividing the shielding area and the adjacent non-shielding area according to grids to sequentially obtain a plurality of shielding grids and adjacent non-shielding grids, distributing the characteristic loss of the shielding area to each shielding grid according to the grid area occupation ratio and the distance weight, and correcting the characteristic value of the adjacent non-shielding grids by using the characteristic loss of the shielding grids; S34, calling the reference frame as a complement template, combining the corrected characteristic values of the adjacent non-occlusion grids, carrying out characteristic complement on each occlusion grid, and calibrating the spatial similarity threshold value of all the adjacent non-occlusion grids after the complement.
8. The image vision-based gangue classification and identification method as claimed in claim 1, wherein the implementation manner of the step S4 comprises: s41, extracting the space position of the shielding area, and dividing the space position into a plurality of sorting paths according to the position of the transmission belt; S42, setting classification labels for each sorting path by combining the characteristic value of the shielding area on each sorting path and the space similarity threshold value between adjacent non-shielding areas.
9. The image vision-based gangue classification and identification method as claimed in claim 1, wherein the implementation manner of step S5 further comprises: And obtaining classification labels of the candidate areas under multiple iterations, marking the candidate areas by using the corresponding shielded areas and the non-shielded areas of the candidate areas, and taking the combined data as the output area characteristics.
10. Gangue classification and identification system based on image vision, which is characterized by comprising: The area dividing module is used for acquiring a mixed image of coal and gangue by adopting a camera array and dividing a plurality of candidate areas by taking the position coordinates of any element in the mixed image as a basis; the shielding identification module is used for determining the mixing state of coal and gangue at the overlapping position based on the overlapping position of each candidate region, quantifying the characteristic loss value of each candidate region under overlapping shielding according to the mixing state, and distinguishing the shielding region and the non-shielding region under mixing of the gangue by using the characteristic loss; The feature synchronization module is used for carrying out grid fitting on the shielding region and the adjacent non-shielding region in the current frame based on the continuous frame images of the shielding region, synchronizing the feature loss of the shielding region into the adjacent non-shielding region and determining a spatial similarity threshold of the adjacent non-shielding region; the classification labeling module is used for carrying out multi-target cooperative processing according to the space positions of the shielding areas, and unifying classification labels of the coal gangue in each candidate area by combining the space similarity threshold values between adjacent non-shielding areas; The data output module is used for defining the region characteristics of each candidate region according to the value range of the classification label and taking the region characteristics as the data synchronously output.

Description

Gangue classification and identification method and system based on image vision Technical Field The invention relates to the technical field of gangue identification, in particular to a gangue classification identification method and system based on image vision. Background Gangue is a typical associated solid waste discharged in the coal mining and washing processes, and has obvious regional cause differences and mineral diversity. The mineral composition mainly comprises two major classes, namely clay minerals such as kaolinite, illite and chlorite, and gangue minerals typically represented by quartz, feldspar and the like. In addition, there is often a proportion of non-mineral phase components such as pyrite, calcite, and residual carbonaceous materials. Aiming at the identification process between coal and gangue, the method generally distinguishes the coal and the gangue based on the mode of image acquisition and analysis, when the images are inconsistent, the images are marked and the corresponding gangue is removed, but the identification is difficult to be carried out on the mixed state of the coal and the gangue under the treatment, so that the image identification precision is reduced due to the shielding condition of the coal and the gangue, and the coal resources are wasted. As disclosed in chinese patent publication No. CN119180974a, an intelligent recognition analysis system for coal gangue based on machine vision and image processing is disclosed, which analyzes the fit score of each target object relative to the coal gangue by combining the first, second and third order matching degrees of each target object relative to the coal gangue, so as to identify whether each target object is the coal gangue, wherein the first order matching degree comprehensively considers the color features, texture features and morphological features in the appearance image of each target object, the second order matching degree comprehensively considers the thermal distribution effect, the thermal boundary effect and the thermal conductivity effect fed back in the thermal imaging image of each target object, the third order matching degree comprehensively considers the internal uniformity of the X-ray imaging partition of each target object and the uniformity between partitions, and the matching degrees of different orders evaluate the target objects from different angles, so that the coal gangue recognition has strong complementarity and accuracy. The method comprises the steps of embedding a CBAM attention mechanism into a whole network core module C2f, increasing small target detection capability, adopting a rapid space pyramid pooling module in a backbone network, fusing more front and rear features, enhancing interaction of feature information, further improving multi-scale detection capability, increasing multi-scale detection capability and preventing small target features from being lost due to network depth through a bidirectional feature fusion module with more jump connection, merging convolution blocks for feature fusion in a detection head to be light, setting regression Loss to EIoU _loss, and avoiding errors caused by consistent aspect ratio of a prediction frame and a real frame. In the prior art, the characteristics of the coal gangue identified under three dimensions are used as distinguishing points to finish the division of the coal gangue through appearance images, thermal imaging and X-ray imaging as the subjects of detection analysis, and pyramid pooling analysis is used for finishing multi-scale detection processing of the coal gangue, but the processing modes ignore inaccurate classification caused by factors such as element overlapping shielding, characteristic missing and the like, emphasize characteristic pooling convolution processing under a single frame image, and easily enable captured characteristic values to be difficult to adapt to spatial information corresponding to a transmission belt, so that the situation of misidentification when coal and the coal gangue are mixed is caused. Disclosure of Invention In order to solve the technical problems, the method for classifying and identifying the coal gangue based on image vision comprises the following steps of S1, acquiring a mixed image of coal and the coal gangue by adopting a camera array, and dividing a plurality of candidate areas by taking position coordinates of any element in the mixed image as a basis. S2, based on the overlapping position between each candidate region, determining the mixing state of coal and gangue at the overlapping position, quantifying the characteristic loss value of each candidate region under overlapping shielding according to the mixing state, and distinguishing the shielding region and the non-shielding region under mixing of the gangue by using the characteristic loss. And S3, performing grid fitting on the shielding region and the adjacent non-shielding region in the current frame based on co