Rock Sampling And Prediction Combined With Image Point Cloud On Mars

With the continuous development of human spaceflight technology,Mars exploration has achieved fruitful results.The autonomous geological analysis of the detector is a key technical problem to be solved in the remote environment.Through the automatic detection and identification of the rock in the scene,selective for individual rock samples analysis,researchers can effectively survey the geological profile of the area and maximize work efficiency.In rock detection,due to long time dust cover,some rocks are similar to the color of the Martian surface,which are hard to distinguish.At the same time,Mars probe to obtain image will be affected by light conditions and capturing conditions.Traditional target detections are mostly based on image processing.In the Martian scene,it is difficult to obtain good detection results based on this traditional image processing method.In this paper,focusing on the Martian rock scene,we put forward an image segmentation and extraction method combined with 3D point cloud information,sampling candidate region image of rock,thus optimizing the detection and extraction of rock objects in the scene.We use SVM classifier to further classify and identify the extracted rock objects,the main contents are as follows:1.The image is denoised by bilateral filtering and segmented by mean-shift algorithm to simplify the image information.By filtering out the ground point to obtain the three-dimensional information of the rock,and by using the correspondence between the three-dimensional information of rock and the image pixels,the obtained rock point cloud is projected to the image space,and the corresponding visual images are obtained.2.Obtain the gradient information of the above two images,extract and fuse the marked images by the extended minimum method.Fused marker images which need to be further segmented,are used to mark the watershed segmentation algorithm.Those images are clipped by solving the smallest rectangle in the segmentation area,and then the rock image part contained in the scene is obtained.3.We have enhanced the processing of the obtained rock images,used the gray level co-occurrence matrix,ULBP features,wavelet transform and other methods to get the image features,and constructed SVM classifier.By training and testing the actual sample data,the result of rock classification is obtained.Experimental results show that the presented images and three-dimensional point cloud data in combination with the method of image segmentation and extraction can effectively be applied to rocky planet in the scene test.At the same time,the classification method based on SVM achieves the purpose of rock classification.