CN-122023340-A - Belt ore drawing cooperative control method based on visual identification and three-dimensional laser fusion

CN122023340ACN 122023340 ACN122023340 ACN 122023340ACN-122023340-A

Abstract

The invention discloses a belt ore drawing cooperative control method based on visual identification and three-dimensional laser fusion, which particularly relates to the technical field of mine automation control, and comprises the steps of acquiring a continuous image sequence and corresponding three-dimensional laser point cloud data of an ore drawing area, extracting an ore stacking edge contour based on the image sequence, identifying an ore effective stacking area by utilizing a semantic segmentation network, combining the area and the point cloud data, extracting a three-dimensional ore stack surface model and generating a fitting boundary point set, constructing an ore stacking geometric model through a spatial registration algorithm, extracting the gravity center, the volume change rate and the edge gradient characteristic of the ore stacking in the model as an input driving cooperative control model, adjusting the belt speed and the aperture of an ore drawing valve, dynamically correcting weight parameters in the control model based on real-time control errors and state feedback of an ore stack, and realizing self-adaptive ore drawing control.

Inventors

ZHOU GUOJUN
ZHANG ZHEN
ZHANG ZHIBIN
ZHENG FEI
CHEN SHANSHAN
WU MENGLIN
WANG DONGLIN
Che Yiting
QU DAPENG
WAN YINGTAO

Assignees

山东艾特智能技术有限公司
山东黄金矿业(莱州)有限公司焦家金矿

Dates

Publication Date: 20260512
Application Date: 20260130

Claims (7)

1. A belt ore drawing cooperative control method based on visual identification and three-dimensional laser fusion is characterized by comprising the following steps: acquiring a continuous image sequence of an ore drawing area and corresponding three-dimensional laser point cloud data, and constructing a time sequence registration data set; Extracting ore stacking edge contours based on the image sequences, and identifying an effective ore stacking area by utilizing a semantic segmentation network; extracting a three-dimensional heap surface model at a corresponding moment based on the three-dimensional laser point cloud data according to the identified effective ore stacking area, and generating a fitting boundary point set; Carrying out spatial registration on the ore stacking edge contour and the boundary point set, and constructing an ore stacking geometric model through a minimum error fusion algorithm; Extracting the gravity center position, the volume change rate and the edge gradient characteristics of the ore stacking in the ore stacking geometric model; Taking the gravity center position, the volume change rate and the edge gradient characteristics of ore accumulation as inputs, driving a cooperative control model of an ore drawing belt and an ore feeding device, and adjusting the belt speed and the opening of an ore drawing valve; And dynamically correcting weight parameters in the cooperative control model based on real-time control errors and pile state feedback to realize self-adaptive cooperative control.
2. The belt ore drawing cooperative control method based on visual identification and three-dimensional laser fusion according to claim 1, wherein the step of extracting the ore stacking edge contour based on the image sequence and identifying the effective ore stacking area by using a semantic segmentation network comprises the following steps: Performing edge enhancement pretreatment on each frame of image in the constructed time sequence registration data set, and extracting an initial contour of an ore stacking edge by adopting a multi-scale Sobel operator; Constructing a semantic segmentation neural network taking a U-Net structure as a core, classifying the preprocessed image at a pixel level, and outputting a probability map containing an ore region and a non-ore region; Fusing the probability map with the edge initial contour, and optimizing a segmentation result based on contour similarity and regional connectivity to generate a mask map of an effective ore stacking region; and extracting a boundary point set of the ore stacking area under an image coordinate system by using the mask map.
3. The belt ore drawing cooperative control method based on visual identification and three-dimensional laser fusion as defined in claim 1, wherein the step of generating a fitting boundary point set comprises the following steps: projecting the identified boundary point set of the effective ore stacking area under the image coordinate system to a three-dimensional laser point cloud coordinate system through an external parameter mapping relation, and screening out a point cloud subset of the corresponding area; Performing outlier rejection and noise filtering treatment on the point cloud subset, and uniformly sampling by adopting a voxel grid filtering method; and constructing a continuous three-dimensional heap surface model for the filtered point cloud subset by using a moving cube algorithm, extracting the surface boundary contour of the continuous three-dimensional heap surface model and generating a fitting boundary point set.
4. The belt ore drawing cooperative control method based on visual identification and three-dimensional laser fusion according to claim 1, wherein the step of constructing an ore stacking geometric model through a minimum error fusion algorithm comprises the following steps: Carrying out coordinate normalization processing on the ore stacking edge contour point set under the image coordinate system, and transforming the ore stacking edge contour point set to a three-dimensional laser point cloud coordinate system based on a camera external parameter matrix; Calculating an initial rotation matrix and a translation vector between the edge contour point set and the three-dimensional boundary point set by adopting an initial rigid registration algorithm so as to establish a space mapping relation between the edge contour point set and the three-dimensional boundary point set; performing fine registration based on an iterative nearest point algorithm, minimizing Euclidean distance errors between corresponding point pairs, and optimizing space alignment precision; And (3) jointly interpolating the registered edge contour and the boundary point set to construct a continuous closed ore stacking geometric model.
5. The belt ore drawing cooperative control method based on visual identification and three-dimensional laser fusion according to claim 1, wherein the step of extracting the gravity center position, the volume change rate and the edge gradient characteristics of the ore deposit in the ore deposit geometric model comprises the following steps: Dividing the model into regular three-dimensional voxel units based on the constructed ore stacking geometric model, endowing each voxel with corresponding space coordinates and occupation states, and calculating the integral gravity center position of ore stacking according to voxel distribution, wherein the gravity center position is a space average value of each voxel center coordinate weighted according to voxel volume; Under continuous time, carrying out differential calculation on the occupation quantity of voxels corresponding to the geometric model of the ore stacking of two adjacent frames, and combining the sampling time interval to obtain the volume change rate of the ore stacking volume along with the change of time; And extracting the normal vector change condition of adjacent surface units along the boundary of the outer surface of the ore stacking geometric model, and calculating the spatial gradient amplitude to obtain the edge gradient characteristic representing the steep degree of the ore stacking edge.
6. The belt ore drawing cooperative control method based on visual identification and three-dimensional laser fusion according to claim 1, wherein the step of adjusting the belt speed and the opening of the ore drawing valve comprises the following steps: uniformly mapping the extracted gravity center position, the volume change rate and the edge gradient characteristics of the ore stacking into standardized control input vectors, and carrying out normalization processing according to a preset characteristic range; Constructing a cooperative control model based on fusion of a fuzzy rule and linear weight, wherein the model adjusts the belt running speed in the gravity center deviation direction, adjusts the opening of a mineral drawing valve in the volume change rate, and adjusts the mineral drawing rhythm in the edge gradient change; dynamically updating characteristic input in a control period, calculating a belt speed regulating quantity and a target opening value of the ore drawing valve, and forming a control output instruction; And a control instruction is sent to the belt driving device and the ore feeding executing mechanism in real time, so that closed-loop cooperative control of ore flow, stacking form and ore drawing stability is realized.
7. The belt ore drawing cooperative control method based on visual identification and three-dimensional laser fusion according to claim 1 is characterized in that the step of dynamically correcting weight parameters in the cooperative control model comprises the steps of collecting actual belt speed, ore drawing valve opening and pile geometric change results in each control period, calculating error values of the actual belt speed, the ore drawing valve opening and pile geometric change results and controlling output of the actual belt speed, the ore drawing valve opening and pile geometric change results in the previous period to form a control error vector, constructing error sensitive factors according to the control error vector and historical change trend, and dynamically updating weight parameters corresponding to input features in the cooperative control model by adopting an exponential sliding average algorithm.

Description

Belt ore drawing cooperative control method based on visual identification and three-dimensional laser fusion Technical Field The invention relates to the technical field of mine automation control, in particular to a belt ore drawing cooperative control method based on visual identification and three-dimensional laser fusion. Background Traditional belt ore drawing operation realizes accurate discharging of ore or material through the coordinated operation of control belt feeder start-stop and ore feeding device often. However, in complex, space-constrained or environmentally dynamic mining areas, the following key problems often occur: the spatial coupling error between the ore drawing port and the ore receiving belt is large, and especially in the belt running process, the ore drawing deviation and the material scattering and even the belt deviation are easy to be caused due to the uncertainty of ore accumulation and the foundation settlement or the tiny deviation of ore drawing equipment. The method lacks of accurate sensing means for real-time geometric states of ores, and is characterized in that the current method relies on monocular vision or infrared detection, is obviously influenced by dust shielding and illumination variation, and is difficult to stably extract ore stacking boundaries and volume contour features. The three-dimensional laser point cloud and the visual image are difficult to fuse efficiently, the existing algorithm depends on external parameter calibration and fixed scenes, fusion drift caused by dynamic vibration or visual angle perturbation of equipment is difficult to deal with, and a control system cannot obtain an accurate heap model for feedback in real time. In a complex mining area, a small ore drawing error may cause serious hidden dangers such as collapse of a ore heap, blockage, equipment damage and the like, and how to realize stable identification, accurate modeling and high-reliability fusion sensing of geometric characteristics of ore heap and feed back sensing results to a cooperative control system in real time is a key problem to be solved urgently in the field of current belt ore drawing automation control. Disclosure of Invention The invention aims to provide a belt ore drawing cooperative control method based on visual identification and three-dimensional laser fusion, which aims to solve the defects in the background technology. In order to achieve the purpose, the belt ore drawing cooperative control method based on visual identification and three-dimensional laser fusion comprises the following steps: acquiring a continuous image sequence of an ore drawing area and corresponding three-dimensional laser point cloud data, and constructing a time sequence registration data set; Extracting ore stacking edge contours based on the image sequences, and identifying an effective ore stacking area by utilizing a semantic segmentation network; extracting a three-dimensional heap surface model at a corresponding moment based on the three-dimensional laser point cloud data according to the identified effective ore stacking area, and generating a fitting boundary point set; Carrying out spatial registration on the ore stacking edge contour and the boundary point set, and constructing an ore stacking geometric model through a minimum error fusion algorithm; Extracting the gravity center position, the volume change rate and the edge gradient characteristics of the ore stacking in the ore stacking geometric model; Taking the gravity center position, the volume change rate and the edge gradient characteristics of ore accumulation as inputs, driving a cooperative control model of an ore drawing belt and an ore feeding device, and adjusting the belt speed and the opening of an ore drawing valve; And dynamically correcting weight parameters in the cooperative control model based on real-time control errors and pile state feedback to realize self-adaptive cooperative control. Preferably, the step of extracting the ore stacking edge contour based on the image sequence and identifying the effective ore stacking area by using the semantic segmentation network comprises the following steps: Performing edge enhancement pretreatment on each frame of image in the constructed time sequence registration data set, and extracting an initial contour of an ore stacking edge by adopting a multi-scale Sobel operator; Constructing a semantic segmentation neural network taking a U-Net structure as a core, classifying the preprocessed image at a pixel level, and outputting a probability map containing an ore region and a non-ore region; Fusing the probability map with the edge initial contour, and optimizing a segmentation result based on contour similarity and regional connectivity to generate a mask map of an effective ore stacking region; and extracting a boundary point set of the ore stacking area under an image coordinate system by using the mask map. Preferably, the step of generating the f