Research On Gradation Detection Method Of Earth-rock Dam Material Based On Digital Image Processing

The gradation of earth-rock dam m,aterial is the basic parameter for studying the performance of earth-rock dam,which has great significance to the dam quality control.At present,the gradation detection in engineering mainly adopts the screening method,which calculates the gradation data through random,sampling and manual screening,which is time-consuming and the results are not representative enough.With the development of computer technology,digital image processing has been widely used in various fields,which providing a new means for gradation detection.For soil and rock images,traditional image recognition algorithms use methods such as mathematics and topology to process pixel gray values.This theory is mature but the accuracy is difficult to meet the demand.Deep learning image recognition extracts the features of the image marked area through neural network,which accuracy is higher but the model is complex to consuming.Therefore,an efficient and accurate gradation detection method based on images has become an urgent need in the current earth-rock dams construction.This research took the earth-rockfill materials as the object,established a sample dataset of earth-rockfill material,image through field collection and experiment,compared and analyzed the image processing results through the traditional image recognition method based on gray level criterion and the image recognition model based on deep learning.Based on the quantitative analysis of the shape of the soil and rock materials in different particle sizes,the conversion between the two-dimensional characteristics of the earth-rockfill materials and the gradation data is studied.Combining traditional image recognition algorithm and deep learning data analysis model,an earth-rockfill material gradation detection model based on image was established,and the accuracy of this model was verified through experiments.The main research contents and results are as follows:(1)For the earth-rockfill material images,the detection results of the two traditional image recognition methods were compared with the OTUS thresholding algorithm and the watershed segmentation algorithm,and clarified the basic principle and process of gradation detection of earth-rockfill material based on images.(2)Established the earth-rockfill images dataset in different states,and analyzed the segmentation results of the deep learning method of Mask R-CNN and U-Net on the earth-rockfill image.(3)Quantitative analysis and research on the morphology of earth-rockfill material in different particle size ranges showed that the same batch of earth-rockfill material in the same material yard had a good similarity in the morphological quantitative characteristics.On this basis,through two methods of two-dimensional morphology analysis and three-dimensional volume reconstruction,a conversion model between the contours of earth-rockfill materials and gradation data was established,which verified the feasibility of gradation detection based on the two-dimensional morphological characteristics.(4)Combining the OTUS thresholding algorithm and Convolutional Neural Network,an earth-rockfill material gradation detection based on images-Deep Otus Convolutional Neural Network(DO-CNN)was established,designed and built the prototype system,which could detect the gradation rapidly base on images.Through 18 sets of screening samples,showed that the max MAPE of the DO-CNN was 2.45%,For particle below 5mm,the detection results were still exact.