Software implementation of the algorithm for automatic detection of lineaments and their properties on open-pit dumps

Abstract

The paper presents an algorithm and a description of its software implementation for detecting lineaments (ground erosions or cracks) in aerial photography images of open-pits. The proposed approach is based on the apparatus of convolutional neural networks based on the semantic classification of binarized images of objects (lineaments), as well as graph theory for determining the geometric location of linearized objects, followed by determining their lengths and areas. Three-channel RGB images of high-resolution aerial photography (pixel 10x10 cm) were used as initial data. The software unit of the model is logically divided into three layers: pre-processing, detection and post-processing. The first level includes preprocessing of input data to form a training sample based on successive transformations of an RGB image into a binary one using the OpenCV library. A neural network of the U-Net type, which includes blocks of the convolutional (Encoder) and scanning parts (Decoder), represents the second level of the information model. At this level, automatic lineament detection (washouts) is implemented. The third level of the model is responsible for calculating the areas and lengths of lineaments. The result of the work of the convolutional neural network is transferred to the input. Lineament area is calculated by summing the total number of points multiplied by the pixel size. The length of the lineaments is computed by linearizing a plane object into a line segmental object with nodal points and then calculating the lengths between them, also relying on the resolution of the original image. The software module can work with input images, with their subsequent resulting merging to the size of the original image.