Research On Automatic Extrapolation Method Of Cold-Rolled Strip Stockpile Based On Monitoring View Images

Cold-Rolled Strip(hereinafter referred to as "CRS")is one of the important cargoes in the breakbulk port,and the accurate and efficient management of their stocking in the port area is an important task of the breakbulk port.At present,the management of CRS adopts the record method of incoming and outgoing ledger,and the inventory is carried out regularly by hand(to check the stacking position and quantity of CRS).Due to the different specifications of CRS,various packaging forms and uneven stacking,manual checking is prone to errors and costly.It is of practical significance and theoretical value to study the detection and automatic projection method of CRS based on the image of the monitoring viewpoint(about 45 degrees oblique in the vertical direction),which can realize the automatic inventory of CRS and thus improve the efficiency and real-time of CRS management while reducing the cost.For the problem of automatic CRS inventory projection,relevant research has been carried out,and the main research contents are summarized as follows.(1)The detection of CRS in the port warehouse is achieved by improving the data enhancement and NECK module in the YOLOv5 algorithm.The lack of target feature information and blurred boundaries caused by mutual occlusion between CRS can lead to possible leakage and false detection in the YOLOv5 algorithm,so a target detection algorithm for CRS(Cold-Rolled Strip YOLOv5,CRS YOLOv5)is proposed.The algorithm is improved on the basis of YOLOv5 by adding Simulate Mask,a single-sample data enhancement method to simulate CRS masking,and adding jump connections to the NECK module of YOLOv5.Experiments show that for the CRS detection task,the CRS YOLOv5 algorithm has a higher accuracy rate with similar recall,i.e.,it has a higher Average Precision(AP).(2)A Cold-Rolled Strip Calculation Algorithm Based on Image(CRSCABI)is designed to realize automatic inventory of CRS in the warehouse.Due to the stacking of CRS and the occlusion of adjacent shell views(shells are the basic property of a CRS stacking field,each shell contains several rows,each row can hold one CRS),some CRS cannot be captured completely by the camera and therefore cannot be recognised by the target detection algorithm,therefore an automatic CRS stacking inventory projection algorithm is proposed to project the occluded CRS from the CRS YOLOv5 detection results.Firstly,the stacking form of CRS in the stacking yard is modelled to obtain a CRS stacking model.Secondly,a prediction frame de-weighting and non-conventional prediction frame correction algorithm is designed to correct overlapping and misaligned prediction frames.Next,a CRS clustering algorithm was designed to classify CRS into different bays.Subsequently,the row and layer numbers of CRS were calculated based on their bay numbers and positions to obtain the visible stacking of CRS.Then,different extrapolation algorithms are designed for each of the two packaging forms,lying rolls and standing rolls,to derive the actual stack inventory based on the visible stack inventory of CRS.Finally,it is reconciled with the CRS incoming ledger,thus realising the automatic inventory of CRS.(3)The CRSCABI algorithm is tested and evaluated with video surveillance data from an inland port warehouse with an oblique 45-degree(vertical direction)view as the input object.The results show that the algorithm has an accuracy rate of over 90% and can effectively project the actual stacking situation of CRS.At the same time,its accuracy rate in the task of counting the number of CRS in and out of storage in each work shift is above 99%,and it can complete the automatic verification of the stacking position of CRS and the number of CRS in and out of storage in each work shift,thus realising the automatic inventory of CRS.The work in this paper is a preliminary exploration of the problem of automatic inventory of CRS.The field test in an inland river port warehouse shows that the method of this paper can reduce the workload of the on-site stocker and improve the accuracy and real-time of CRS inventory,which has certain practical significance and value.