Parcel Identification System Based On Improved Faster R-CNN

With the continuous development of e-commerce,the volume of parcel delivery services in my country has also grown rapidly,while parcel delivery identification and sorting technology has also undergone diversified development.Automatic parcel sorting systems based on radio frequency technology,barcodes and QR codes have also been used.But when it comes to parcel delivery station scenes,these technologies cannot yet distinguish the packages of different express companies.At present,there are precedents of using convolutional neural network to identify the signs of express delivery,coupled with the rapid development of deep learning,it can be seen that the object detection technology based on deep learning has broad application prospects in parcel identification and sorting.This study uses Faster R-CNN model to realize the identification of parcel delivery,and distinguishes the packages of different express companies by learning the characteristics of express signs,which has an important theoretical basis for the subsequent improvement of express sorting and parcel delivery efficiency.The specific work of this paper is as follows:First of all,since there is no public express delivery dataset,this article uses seven common express delivery: youzheng,yuantong,yunda,shentong,baishi,zhongtong and shunfeng as the research samples to make the express delivery dataset.Through a series of work such as image acquisition,data labeling,data classification,the creation of the dataset is completed.The author introduces the centralized method of data enhancement,and further improved the dataset by rotating,adding noise,and adjusting brightness.Finally,the author uses the Faster R-CNN model to complete the training and testing of the model on the self-built express dataset.Experiments show that the average accuracy rate of the express delivery recognition system based on Faster R-CNN model reached 87.36%,and achieved good recognition results.Secondly,in view of the problem that the anchor points in the original model are not suitable for the express dataset,the k-means++ algorithm is used to cluster the target frames in the parcel dataset to obtain a suitable aspect ratio,and the scale of the initial candidate frame is improved to adapt the size of parcel.Experiments show that the average accuracy of the system after adding the clustering strategy reaches 89.19%,which is 1.83% higher than the original.It can be seen that the appropriate size of anchor points can reduce redundancy and help the model to accurately locate the frame.Finally,in the express delivery identification scene,there will be problems such as overlapping and occlusion for file express delivery.Using the non-maximum suppression algorithm in the original model will cause missed detection.Aiming at this problem,this paper proposes a dual threshold-nonmaximum suppression algorithm based on the aspect ratio of the candidate frame.Firstly,some borders that are obviously not express targets are eliminated by the aspect ratio of the candidate frame,and then double threshold-non-maximum suppression is performed.Compared with the original algorithm,the improved non-maximum suppression improves the average accuracy of the system by3.48%,and the missed detection rate of the system is also reduced by 3.4%.Taken together,the improved Faster R-CNN with clustering strategy and improved non-maximum suppression method is more robust than the standard network,and finally achieves an average accuracy of 92.07%.This method can effectively improve the accuracy of express delivery identification,thereby further improving the delivery efficiency.