Express Form Information Processing And Application Based On Deep Learning

Recently,the industry of express service has developed rapidly and continuously.A large number of automatic package sorting systems based on the recognition of barcode and QR code has been applied.However,the environment of courier sorting site is so complex that barcode contamination occurs from time to time.Generally,parcels that cannot be read by the barcode are sorted by the way of manual sorting which is time-consuming and labor-intensive.The methods of objects detection and Chinese character recognition in deep learning have the advantages of fast speed,low cost,and automation,which have broad application prospects in express parcel sorting systems.Based on deep learning,this paper designs a system for extracting value information on the pictures of express delivery forms.Firstly,this thesis designs a fast R-CNN based scheme to locate the valuable information on the express form image rapidly.This scheme is able to obtain the coordinates of the key information at a higher accuracy with deep feature extraction,cascade detection model,cascade model acceleration strategy and improved positive sample supplementation method.Secondly,this paper proposes a method based on the improved CRNN model to recognize the valuable information such as names,phone numbers,and addresses on the pictures.The approach applies a network based on improved Efficient Net to extract text feature maps,a bidirectional long and short-term memory network for the extraction of feature sequences,and a weighted CTC method with time-step loss to translate the feature sequences into text sequences.Finally,in order to improve the efficiency of the package sorting and delivery,this article designs a system which consists of an image preprocessing module,a positioning module,a recognizing module,and a results correcting module.Moreover,the system could be widely used in fields such as invoice identification,identity card identification,etc.