Multi-target Detection And Separation Methods For Automatic Transport Equipment

With the rapid development of e-commerce in China,intelligent logistics based on automated equipment has become one of the hotspots in the field of artificial intelligence and Internet of Things as a product of the combination of a new generation of intelligent technology and traditional industries.In the process of express delivery and sorting,single-piece separation systems with high-precision detection and positioning and realtime separation functions can replace manual work at the front end of the logistics production line to complete the express delivery ”queue” and make the express delivery output in the form of a single column in an orderly manner.However,due to plant equipment limitations,detection algorithms that do not meet real-time requirements,and external light variations,the final efficient and stable separation poses a great challenge.Therefore,to address the above problems,thesis designs and implements a multi-objective separation strategy and system based on automatic transfer equipment and completes field tests and other work,based on improving the relevant algorithms.The main research work in thesis has the following three points:Firstly,to address the problem of uneven distribution of external ambient light causing subsequent detection algorithms to miss and mis-detect,thesis uses an extended＿Lagrange multiplier alternating direction minimisation algorithm based on the light-reflection model to extract global light information from the image,and optimises the control parameters of the gamma correction algorithm by introducing light judgement thresholds and combining the light distribution characteristics of each pixel of the image,so that it can finely and autonomously adjust the image light distribution The Gamma Correction algorithm is capable of finely adjusting the image light distribution.The image correction experiments on the self-harvested image data prove that the algorithm in thesis has better light equalisation capability than other algorithms,and can better improve the overall image quality and eliminate external light interference.Secondly,to address the problem that the detection and localization accuracy and speed of the multi-target detection algorithm cannot meet the demand of real-time detection,thesis lightens the Yolov5 n model based on GhostNet small convolutional computing network,and corrects the distance cost in the original loss function by the angle-sensitive SIoU＿Loss loss function,finally enables the algorithm to improve detection accuracy nondestructively while speeding up detection localisation and reducing the number of model parameters.The algorithm’s multi-target detection capability is tested on a self-built shipment dataset,verifying that the model can accurately detect and locate multiple targets in real time,and that it can still maintain good performance on low performance computer equipment.Finally,this thesis designs and implements a multi-target separation strategy based on the target location state information output by the detection algorithm and builds a shipment separation system,encapsulating each part of the algorithm in a software system,and using the actual logistics separation production line transmission equipment as the experimental platform to conduct practical functional tests,proving the feasibility of the implementation of the strategy and the functional integrity of the system.The separation system has been officially launched and put into use,and its stable peak efficiency is6500 pieces/hour in the process of separating multiple express pieces with varying degrees of standardisation.Compared with the more advanced separation equipment currently in mass production on the market,the system designed in thesis has a high separation efficiency and stability,and can provide an efficient and customised solution for major logistics enterprises and warehousing centres.