Research On Detection And Quantitative Statistics Of Open-pit Slope Fractures Based On Machine Vision

The slope stability of Open-pit mine is closely related to ensuring the safety of mine production,guaranteeing the safety of mine employees' lives and property,and improving the economic efficiency of the mine.With the increase of open-pit mining depth and slope angle,the problem of slope stability becomes more and more prominent,and slope landslide disasters occur from time to time.The reason is often that there is no timely and effective slope crack detection.Although the slope cracks will not have a direct and huge impact on the stability of the slope,the untimely detection of cracks will lead to serious hazards such as landslide of the open-pit mine slope.Therefore,it is very important to select scientific and reasonable methods for timely and accurate detection and quantitative statistics of open-pit mine slope cracks,which can provide data support for the safety production management of open-pit mine.The specific work of this paper mainly includes the following aspects:(1)This paper briefly expounds the deformation monitoring technology,digital image processing technology and crack detection method of open-pit mine,and introduces the related depth learning theory and framework,which provides a way of thinking for intelligent detection and quantification model of slope crack in open pit.(2)Aiming at the problems of poor detection effect of the open-pit mine slope,low segmentation accuracy of the fractures' edge and serious false detection in the direct application of the crack detection methods based on threshold segmentation,edge detection and machine learning,this paper selects the deep learning framework Mask R-CNN which integrates the characteristics of target detection and segmentation as the basic network of fracture detection.In order to solve the problems of unclear crack edge and false detection in mask branch,the atrous spatial pyramid pooling is introduced.The multi-scale feature extraction of slope cracks is realized by adding the classify segmentation iterative up-sampling operation in the mask branch,which improves the accuracy of slope crack segmentation.(3)Aiming at the problem that there is no unified and appropriate quantitative statistical method to evaluate the damage degree of the detected cracks,this paper combines the fracture morphology classification model and the quantitative statistical model to achieve the goal of statistics quantifying the detected and segmented cracks.The core idea of this method is as follows: firstly,based on convolution neural network,a pixel level classification model of open pit slope fracture morphology is constructed,and the detected cracks are divided into transverse,longitudinal and oblique single cracks,and irregular cracks.Then,based on the classification of fracture morphology,different statistical quantification methods are used for different types of fractures.For a single fracture image,the relevant attribute information such as fracture length,minimum critical width and maximum critical width are statistically quantified based on projection mapping method.For the irregular fracture image,the related attribute information of the irregular fracture is quantified based on the fracture connected domain labeling method and fracture skeleton extraction method,and then the corresponding damage degree can be inferred.It provides data index for slope stability control and mine safety production management of open pit mine.The experimental results show that the intelligent detection model of open-pit slope crack constructed in this paper has better crack target detection results and higher crack edge segmentation accuracy.Meanwhile,the quantitative statistical method based on fracture morphology classification designed on the basis of fracture detection and segmentation basically meets the quantitative statistical requirements of slope fractures of open-pit mine,and can provide data index for safe production management of open-pit mine,which has certain applicability and feasibility.