Study On Key Technologies Of Foreign Object Detection In The Coal Mine Transportation

Coalmine underground belts are expensive,and the presence of foreign object may cause belt damage or even tear,which brings great economic losses.Therefore,it is very necessary to study and optimize the foreign object detection method of coalmine belt conveyor.The belt target detection uses intelligent analysis technology to extract the moving target by removing the background part,so as to effectively and accurately analyze the moving target.Taking the belt as the target,this paper optimizes the existing methods and focuses on the following aspects: video anti-shake mrthod,texture image segmentation,accurate target tracking and foreign object detection algorithm.Specific optimization methods include:1.Fast video anti-shake method based on binary descriptor.Due to the complicated environment under the mine,the camera will intermittently shake with the work,and it is impossible to obtain a clear and smooth video source.By comparing the performance of real-value descriptors and binary descriptors feature extraction algorithms in terms of time and space consumption,binary descriptors are used as feature extraction and matching algorithms.At the same time,RANSAC algorithm is utilized to remove the mismatch,performs motion compensation on the image through perspective transformation,and finally a stable sequence of video frames can be obtained.2.Texture image segmentation based on stationary directionlet domain probabilistic graphical model.Aiming at the problem of insufficient fine grain in the process of segmenting images,this paper proposes a probability graph model of stationary direction wave domain for texture image segmentation.In the stationary directionlet domain,the hidden markov chain and markov random field are combined,and the result of the final texture image segmentation is obtained by minimizing the energy function.Experimental results showed that the proposed method can achieve better performance than other algorithms,especially at the boundaries of homogeneous regions and different regions.3.Real-time visual tracking with compact shape and color feature.Visual tracking is to find and mark the location of the tracked object in each video sequence frame.This paper proposes a structured target tracking method based on compact color-coding.By blending the shape and color features of the candidate target regions,and using the hash function to reduce the dimensions of the combined features,a low-dimensional and compact color-coded feature is formed.Then the structured classification function performs target classification and prediction to achieve target tracking.The method enhances the feature description ability of the target,improves the accuracy of the target classification,effectively avoids the visual tracking drift,and thereby improves the tracking performance.4.Downhole belt foreign object detection algorithm based on multi-camera judgment.This paper proposes a downhole belt foreign object detection algorithm based on multi-camera judgment.On the basis of a pixel background modeling and a foreground detection mthod,this method constructs a multi-camera judgment algorithm by Bayesian estimation theory,and finally determine the type of foreign object through combing training the direction gradient histogram features of the image and the support vector machine classification.On the one hand,this method met the real-time nature of the detection of foreign objects in the down-hole belt.On the other hand,it achieved the intelligence of offline training and online classification.Based on the above algorithm,a prototype system for detecting foreign objects in underground coal belts was designed in this paper to realize real-time detection,alarm and image display of foreign objects transported by underground belts under multi-camera judgment.The experiment results showed that the correct rate of foreign detection could reach high accuracy in the real underground environment.