The Study On The Intelligent Identification Way Of Remote Sensing In Typical Mine Environment

In order to reduce the damage of mine environment and protect the scarce minerals,related departments widely use remote sensing technology and geographic information technology to get a great number of remote sensing data so that they can have the detection work of mineral resources.However,at present the way to get the data of mine environment and minerals detection is still the artificial translation and manual field examination.With the gradual development of national geographic surveys,the current method cannot satisfy the image interpretation rate of the increasing data collection.And instance segmentation,as an important study field in computer vision,has a pixel division on object instance based on target detection.It is a visual technology which extract the pixel-level object contour.The study on the segmentation of remote sensing instance takes the image content as an object instance.It segments various collections of pixel in ground objects.It also has an important meaning to promote remote field from manual detection to automatic image or semi-automatic image detection of high resolution.This paper will deeply learn the study of technology which is applied in intelligent remote recognition in typical mine environment.And it aims to solve the dilemma of the collection of currently environment remote data and the interpretation work.It specifically study on two aspects: data generation and remote sensing detection of typical mine environment.Based on the mine remote detection data,this paper applies a method of automatic generation of instance segmentation data set to design and realize the remote sensing image segmentation by ArcGIS Engine,AE and Weiler-Atherton,and attribute data conversion by Newtonsoft.Json,Json.NET,aiming at the problems of deficiency of dataset and high cost of production.It solves the problem of automatic batch generation of samples,providing data for segmentation network of remote sensing object in typical mine environment.Aiming at the difficult problem of currently data collection of mine interpretation,the paper uses Resnet-101 to network extraction of image feature and builds a model of remote sensing object recognition in typical mine environment based on Mask R-CNN.It applies to thetarget detection of typical mine environment remote sensing and contour extraction.As the result shows,the applied result of automatic generation data of the insurance segmentation has no difference with the accuracy of data interpretation and it is superior to manually annotated in efficiency.This method which guarantees the data accuracy realizes the reuse of data interpretation and generates a deeply normative learning dataset.Based on the insurance segmentation method of Mask R-CNN,the paper can effectively build a model of remote sensing object recognition in typical mine environment through the applied result of automatic generation data of the insurance segmentation.It realizes the target recognition and segmentation of typical mine surface features.