3D Segmentation And Identification Of Iron Ore Finished Product Warehouse Based On Voxel Network

Iron and steel output is one of the important criteria to measure a country’s industrial strength.With the deepening of the national industrialization process,the demand for iron ore materials is increasing,so the automatic transformation of the ore workshop and its finished product forming workshop is the trend of The Times.However,if you want to realize automation,the automatic acquisition of effective information for various targets is the prerequisite.Therefore,based on the idea of voxelization,this paper constructs an intelligent data information database of finished iron ore products,and completes point cloud segmentation,identification and tracking in this database.Specifically,the information base consists of the information of the ore material workshop(ore material information,harvester grab bucket information and open car information)and the information of the metal product forming area(metal solution level information).In the mineral workshop,in view of the fact that many traditional descriptions of mineral piles are simple global information(such as global volume,shape,etc.),the simple global information description is not applicable to the situation such as this paper where fine mineral information is needed for material purity control and multi-machine scheduling.Therefore,using the idea of voxelization,this paper proposed a new 5D ore material information description method,which can accurately describe the center of gravity position,purity and time stamp information of the unit voxel.This description method can provide a good information basis for the multi-machine optimal scheduling and ore material phase purity control of the reclaimer.Due to the sparsity of the grab point cloud and the fact that the grab center is not on the grab itself,many point cloud tracking methods are not ideal in the information acquisition of the harvester grab bucket.In view of this,this paper proposes a point cloud tracking method which combines the object detection algorithm of deep learning with Kalman filter.In this method,the concept of correlation value is proposed according to the actual situation,and the matching association method of Hungary is optimized.Due to the large space of the ore workshop,the distance between the radar and the train is too big compared with the distance between the open cars in the ore workshop.Under this condition,the point cloud collected by radar will be very disorderly and sparse and partially missing.With the progress of loading,the point cloud collected by radar will change constantly.And convertibles may come in different sizes.All these conditions make it difficult to automatically locate and partition the carriage.Based on this,this paper draws on the idea of point cloud voxel and innovativelyputs forward a point cloud automatic positioning and segmentation method based on plane rectangular grid of mine open car.Under the above unfavorable conditions,the relative error of this method can still be controlled within ±1% through actual data measurement,but in terms of actual project requirements,the relative error only needs to be controlled within ±8%.Thus,the superior performance of the proposed method can be seen in the above cases.In the metal solution information acquisition,this paper developed an automatic metal solution liquid level measurement system based on structured light triangulation.The filter algorithm based on statistics is adopted in this system to suppress the adverse effect of ingot line vibration on measurement.At the same time,by increasing the Angle between the optical axis of the camera and the laser plane,the influence of the strong reflection of the solution surface on the measurement is reduced.The relative error of the system can be controlled within 1.3% by comparison experiment,which can meet the actual measurement requirements in engineering.