CN-121980319-A - Coal gangue identification method and system based on distributed optical fiber vibration and image fusion

CN121980319ACN 121980319 ACN121980319 ACN 121980319ACN-121980319-A

Abstract

The invention discloses a coal gangue identification method and system based on distributed optical fiber vibration and image fusion, and relates to the technical field of coal gangue identification. The method comprises the steps of collecting vibration signals caused by coal or gangue striking a hydraulic support in a coal discharging process in real time, obtaining coal flow images according to set frame intervals, classifying the vibration signals according to vibration characteristic differences of coal block falling and gangue striking to obtain vibration identification results, identifying positions and sizes of gangue blocks in each frame of coal flow image by using a pretrained convolutional neural network to obtain image identification results, and carrying out association fusion on the vibration identification results and the image identification results in a time dimension and a space dimension according to a preset fusion strategy to obtain coal gangue identification results. The invention can better distinguish the complex situation that the coal block impact and the gangue impact are overlapped on the frequency spectrum, and improves the accuracy and adaptability of the gangue identification in the top coal caving process.

Inventors

LIANG MINFU
CHEN NINGNING
ZHENG DAQIAN
HUANG WEI
GONG HUAN
LU XINZE
LI SHUANG
WU GANG
FANG XINQIU
JIANG YUYE
ZHANG KUN
SI LEI
ZHANG NINGBO

Assignees

中国矿业大学

Dates

Publication Date: 20260505
Application Date: 20260128

Claims (8)

1. A coal gangue identification method for integrating distributed optical fiber vibration and images is characterized by comprising the following steps: The method comprises the steps that vibration signals caused by coal or gangue striking the hydraulic support in the coal caving process are collected in real time through distributed optical fiber sensors arranged on the hydraulic support of a fully mechanized coal face, and coal flow images are obtained according to set frame intervals through image collecting devices arranged in a coal caving area; Classifying the vibration signals according to the vibration characteristic differences of coal block falling and gangue striking so as to determine the corresponding positions and time of the vibration signals and preliminarily judge the material types, thereby obtaining a vibration identification result; Identifying the position and the size of gangue blocks in each frame of coal flow image by adopting a pre-trained convolutional neural network so as to determine the gangue content of the current frame of coal flow image and obtain an image identification result; and carrying out association fusion on the vibration identification result and the image identification result in the time dimension and the space dimension according to a preset fusion strategy to obtain a coal gangue identification result.
2. The method for identifying coal gangue by using distributed optical fiber vibration and image fusion according to claim 1, wherein the method is characterized in that vibration signals are classified according to vibration characteristic differences of coal block falling and gangue impact so as to determine positions and time corresponding to the vibration signals and preliminarily judge material types, and a vibration identification result is obtained, and specifically comprises the following steps: filtering the vibration signal to remove background noise and abnormal noise in the vibration signal and obtain an effective vibration section; The method comprises the steps of performing feature extraction on an effective vibration section by adopting a pre-trained BP neural network or a random forest classifier to obtain vibration features reflecting time domain statistical features, frequency domain features and time frequency features of vibration data, splicing the vibration features according to a preset sequence, and performing zero mean unit variance normalization processing to obtain a preprocessed feature vector; Inputting the feature sequence into a pre-trained support vector machine model, outputting the probability that each window belongs to three types of pure coal, pure gangue and coal gangue, taking the category corresponding to the maximum probability as a preliminary classification result, carrying out three-window majority voting on the preliminary classification result, combining with the constraint of the minimum duration to obtain a vibration identification category sequence, and aligning the vibration identification category sequence with the actual collision event time according to the window center time to obtain a vibration identification result.
3. The method for identifying coal gangue by using distributed optical fiber vibration and image fusion as claimed in claim 2, wherein training the BP neural network or random forest classifier and supporting vector machine model comprises the following steps: acquiring original vibration signals and corresponding vibration marking results of the mixed dropping of pure coal, pure gangue and gangue in the actual top coal caving process for a plurality of times; Inputting original vibration signals of pure coal, pure gangue and coal gangue mixed falling in the actual top coal caving process for a plurality of times into a BP neural network or a random forest classifier and a support vector machine model to obtain a vibration prediction result; And (3) taking a minimized loss function between the vibration labeling result and the vibration predicting result as a target, and adjusting model parameters of the BP neural network or the random forest classifier and the support vector machine model to obtain the trained BP neural network or the random forest classifier and the trained support vector machine model.
4. The method for identifying coal gangue by using distributed optical fiber vibration and image fusion as claimed in claim 1, wherein training the convolutional neural network comprises the following steps: acquiring underground live-action coal gangue mixed image data, and marking gangue targets in the underground live-action coal gangue mixed image data to obtain a real result; inputting the underground live-action coal-gangue mixed image data into a convolutional neural network to obtain a predicted result of a gangue target in the underground live-action coal-gangue mixed image data; And training the convolutional neural network by taking a loss function between a real result and a predicted result of a gangue target in the minimum underground live-action gangue mixed image data as a target to obtain the trained convolutional neural network.
5. The method for identifying coal gangue by using distributed optical fiber vibration and image fusion according to claim 4, wherein the step of inputting the underground live-action coal gangue mixed image data into a pre-trained convolutional neural network to obtain a predicted result of a gangue region in the underground live-action coal gangue mixed image data comprises the following steps: reading in the underground live-action coal gangue mixed image according to a fixed frame rate to obtain a multi-frame coal gangue mixed image; Denoising and contrast enhancement preprocessing is carried out on the current frame of coal gangue mixed image aiming at each frame of coal gangue mixed image so as to cut a coal flow interest area and scale to a network input size, so that a preprocessed image is obtained; Inputting the preprocessed image into a pretrained convolutional neural network, wherein the convolutional neural network comprises a multi-layer convolutional and pooled feature extraction layer, a feature fusion layer and an output layer which are sequentially connected; Screening the preprocessed images through the multi-layer feature extraction layer according to the confidence threshold value, and performing non-maximum suppression and overlap removal detection to obtain a bounding box or a mask set of the gangue target; Calculating the position, the size and the area proportion of the gangue target to the material area according to the boundary frame or the mask set of the gangue target through the feature fusion layer to obtain the gangue target detection result of the current frame of coal gangue mixed image; And carrying out time sequence smoothing and consistency constraint on gangue target detection results of all frames of the coal gangue mixed image through an output layer to obtain a predicted result of gangue targets in underground live-action coal gangue mixed image data.
6. The method for recognizing coal gangue by using distributed optical fiber vibration and image fusion according to claim 1, wherein the method for recognizing coal gangue by using the distributed optical fiber vibration and image fusion is characterized by performing association fusion on a vibration recognition result and an image recognition result in a time dimension and a space dimension according to a preset fusion strategy, and comprises the following steps: The preset fusion strategy comprises a synchronous verification strategy and a conflict processing strategy; For the same coal discharging process, if the vibration identification result is a gangue impact signal and the image identification result is that the gangue duty ratio is improved or gangue blocks are identified in a set time period, outputting a gangue identification result as a real gangue falling event; The conflict processing strategy comprises the steps of re-collecting new data for rechecking after a set time interval if a vibration identification result is inconsistent with an image identification result, and respectively assigning the reliability of the vibration identification result and the reliability of the image identification result according to an identification reliability evaluation system if the vibration identification result is inconsistent with the image identification result after rechecking, wherein the result with the highest reliability value is used as a coal gangue identification result.
7. A distributed optical fiber vibration and image fusion coal gangue recognition system is characterized by comprising: The distributed optical fiber vibration sensing subsystem is used for acquiring vibration signals caused by coal or gangue striking the hydraulic support in the coal caving process in real time through a distributed optical fiber sensor arranged on the hydraulic support of the fully mechanized coal mining face; The image acquisition subsystem is used for acquiring coal flow images according to set frame intervals through an image acquisition device arranged in the coal discharging area; The vibration identification subsystem is used for classifying the vibration signals according to the vibration characteristic differences of coal block falling and gangue striking so as to determine the corresponding positions and time of the vibration signals and preliminarily judge the material types, thereby obtaining a vibration identification result; The image recognition subsystem is used for recognizing the position and the size of gangue blocks in each frame of coal flow image by adopting a pre-trained convolutional neural network so as to determine the gangue content of the current frame of coal flow image and obtain an image recognition result; And the decision subsystem is used for carrying out association fusion on the vibration identification result and the image identification result in the time dimension and the space dimension according to a preset fusion strategy to obtain a coal gangue identification result.
8. The distributed optical fiber vibration and image fusion coal gangue recognition system according to claim 7, further comprising a control execution unit for controlling a hydraulic support coal discharging valve of the coal discharging area according to the coal gangue recognition result.

Description

Coal gangue identification method and system based on distributed optical fiber vibration and image fusion Technical Field The invention relates to the technical field of coal gangue identification, in particular to a coal gangue identification method and system based on distributed optical fiber vibration and image fusion. Background In the thick coal seam caving coal mining process, realizing automatic identification and separation of gangue is an important target of intelligent mining. Conventionally, underground workers judge when coal caving should be stopped in the process of caving the top coal by means of experience of auditory and visual observation to avoid gangue mixing, on one hand, by hearing the sound characteristics of coal blocks or gangue striking the hydraulic support, and on the other hand, observing the condition of coal dropping on the rear conveyor, and estimating gangue content according to experience. However, the manual judgment has great limitation, so that the coal caving is stopped too early to cause undercaving, or too late to cause excessive gangue mixing, and the intelligent fully-mechanized caving mining requirement is difficult to meet. Therefore, development of an automatic coal gangue recognition technology is imperative. The invention relates to a patent application with the patent number of ZL200910152006.X and the name of a coal gangue recognition and automatic coal discharging control system, which specifically comprises a voiceprint sensor and an image sensor which are arranged at a coal discharging port and are used for collecting audio and video signals when a coal gangue mixture falls, a DSP (digital signal processor) fast processor is used for carrying out matching analysis on an audio frequency spectrum and carrying out digital image processing on video, and a singlechip controller is used for sending a control instruction to drive a hydraulic support coal discharging mechanism to act so as to realize automatic top coal discharging. The system simulates the coal discharge judging process of manual hearing and watching, and aims to automatically close coal discharge when gangue appears so as to reduce gangue mixing. In the prior art, in the process of carrying out matching analysis on a sound spectrum through a DSP (digital signal processor), a threshold value is required to be set manually or characteristics are extracted, but a large number of overlapping areas exist on the spectrum of the sound of coal block impact and gangue impact, and the accurate and robust distinction of coal gangue is difficult to be achieved by simply relying on a plurality of characteristics and fixed threshold values which are set manually. Disclosure of Invention Based on the above, it is necessary to provide a coal gangue identification method and system for integrating distributed optical fiber vibration and images. The embodiment of the invention provides a coal gangue identification method for distributed optical fiber vibration and image fusion, which comprises the following steps: The method comprises the steps that vibration signals caused by coal or gangue striking the hydraulic support in the coal caving process are collected in real time through distributed optical fiber sensors arranged on the hydraulic support of a fully mechanized coal face, and coal flow images are obtained according to set frame intervals through image collecting devices arranged in a coal caving area; Classifying the vibration signals according to the vibration characteristic differences of coal block falling and gangue striking so as to determine the corresponding positions and time of the vibration signals and preliminarily judge the material types, thereby obtaining a vibration identification result; Identifying the position and the size of gangue blocks in each frame of coal flow image by adopting a pre-trained convolutional neural network so as to determine the gangue content of the current frame of coal flow image and obtain an image identification result; and carrying out association fusion on the vibration identification result and the image identification result in the time dimension and the space dimension according to a preset fusion strategy to obtain a coal gangue identification result. Optionally, the vibration signals are classified according to the vibration characteristic differences of coal block falling and gangue striking, so as to determine the corresponding position and time of the vibration signals and preliminarily judge the material types, and a vibration identification result is obtained, and the method specifically comprises the following steps: filtering the vibration signal to remove background noise and abnormal noise in the vibration signal and obtain an effective vibration section; The method comprises the steps of performing feature extraction on an effective vibration section by adopting a pre-trained BP neural network or a random forest classifier to obtain vibration