| Tailings dams are crucial components of mining operations,but they also pose significant environmental and safety hazards due to potential pollution.Therefore,safety research on tailings dams has become a crucial area of focus in the mining engineering field.The length of the dry beach is a critical indicator used to assess the safety and stability of a tailings dam.The aim of this thesis is to introduce a novel approach for measuring the length of dry beaches using depth learning techniques.The primary objective is to use depth learning training to accurately identify the boundary between the dry beach and the water surface,by segmenting the image model.Once the boundary is identified,the proposed method selects reference points on the dry beach water surface boundary based on the image segmentation results.These reference points are used to derive a conversion formula between the reference point and the actual dry beach length,using a monocular ranging model.By applying this conversion formula,the length of the dry beach can be measured with precision.The main research tasks of this thesis include:Firstly,a network structure of feature pyramids has been improved to enhance the accuracy of image segmentation.Firstly,the existing image segmentation methods and their characteristics are introduced.Then,the structure of the Mask R-CNN network is analyzed.In response to the problem that the recognition rate of the Mask R-CNN network is not optimal for large-scale target problems,the feature pyramid network structure is improved by adding a reverse side connection from bottom to top,and then combined with the fusion of multi-scale feature maps.Through experimental comparison,the improved network structure has improved by 6.3% compared to the original network structure in identifying large-scale target problems.Applying the network to the recognition and segmentation of dry beach images results in clear and accurate segmentation.Second,a dry beach length measurement method based on monocular ranging is proposed.Based on the traditional monocular ranging model,the actual situation of the dry beach in the tailings pond is analyzed,and a monocular ranging model with added pitch angle is established to calculate the dry beach length.The pixel coordinate values of reference points on the dry beach waterline boundary are converted into dry beach length using the dry beach length measurement formula,and the error source is analyzed to reduce the error.This method solves the shortcomings of traditional dry beach length measurement methods,this approach simplifies the calculation process and effectively addresses issues such as low accuracy and poor recognition rates.It was successfully applied to measure a tailings pond in Shaanxi Province,where the relative error of the method was found to be less than1.12%.Thirdly,a prototype system for measuring dry beach length based on deep learning was designed and developed.Develop a dry beach length measurement system based on Py Qt5 visual tool package on the pychar platform.The system calculates the length of dry beach by uploading dry beach images,mainly including image segmentation function,length calculation function,and parameter setting function.The system runs smoothly,is simple to use,and has small measurement error,which can meet the needs of dry beach length measurement. |
