Method for automatically identifying side slope rock types

Abstract

The invention discloses a method for automatically identifying the type of side slope rock. The classification of rock slope rocks is crucial to the analysis of slope stability. The conventional manual method is low in efficiency and is influenced by subjective factors. The method comprises the following steps of establishing a convolutional neural network model for rock slope image set analysis,performing feature information extraction and compression on 80000 rock slope images through convolution operation and pooling operation respectively, and then training the network model to realize automatic identification and classification of rock slope rocks. The rock slope images in the training set and the test set are adopted to test and analyze the model, the accuracy rate of the training set reaches 98%, the accuracy rate of the test set reaches 90%, it is shown that the trained network model has good robustness, and an ideal training effect can be achieved. Finally, a network model established by deep learning is adopted to realize the rapidness and automation of rock identification of the rock slope.

Description

technical field [0001] The invention relates to the technical field of construction engineering informatization, in particular to a method for automatic recognition of slope rock types. Background technique [0002] The delineation of the types and ranges of different rocks in slopes is the basic work in the scientific research of slopes. In the past, on-site sampling was often carried out through complex instruments and equipment, and the type of slope rocks and the calibration of different rock boundaries were artificially identified based on the color and structure of the rocks. This process was time-consuming and laborious. [0003] There are two main methods for conventional slope rock identification and classification. The first is the physical test method, which uses physical testing methods to detect slope rocks. For example, X-ray powder diffraction, scanning electron microscopy, infrared spectroscopy, differential thermal analysis, electron probe, hyperspectral i...

AI Technical Summary

Problems solved by technology

The above two types of methods have defects such as complex experiments, long periods, and greater subjective influence, and cannot meet the needs of on-site measurement and evaluation. In order to quickly and accurately extract rock information of rock slopes, many experts have used intelligent algorithms to analyze rock The images were studied: Zhang Xu et al. applied the naive Bayesian K-neighborhood algorithm to classify rock images; Kang Liping et al. used softmax multi-classifiers and multi-classification support vector machines (SVM) to realize image classification; Zhang Jiafan et al. proposed clustering-based Analytical algorithms for rock CT image segmentation and quantification; Li et al. used transfer learning methods to train sandstone images, and finally obtained a high-precision sandstone image classification model

Analysis of the above research progress found that there are some deficiencies: the images use the standard rock slices after subsequent processing, instead of the original rock slope images as the training data set; secondly, the boundaries of different rocks in the slope are not calibrated , it is impossible to determine the boundary range of various types of rocks on the slope

Images