Design And Implementation Of Freight Information Intelligent Management System Based On Deep Learning

With the development of economy and trade,the total amount of freight continues to rise,and intelligent freight platform has become an important development trend.Traditional weighing requires freight drivers to input pound sheet data one by one and audit personnel to check and verify one by one.The process is cumbersome and inefficient,which cannot meet the demands of efficient and fast modern freight industry.In recent years,optical character recognition(OCR)technology based on deep learning is becoming more and more mature,which has a broad application prospect in intelligent extraction and processing of freight bill information.In the actual scene of freight pound bill,there are often interference factors such as fuzzy pound bill,deformation and shadow,which brings great challenges to OCR recognition.In this paper,the information recognition technology and system application of pound sheet under the above nonideal conditions are studied,and the main achievements and innovations are as follows:(1)TextDeblurDAE,an image deblurring processing model,was proposed,which adopted multi-scale architecture design and introduced space and channel attention mechanism to enhance text features of fuzzy images and suppress other invalid features.On the basis of the primary blind deblurring of the first level network,the cascade of the second level network realizes the fine-grained blind deblurring,and the peak signal-tonoise ratio(PSNR)reaches 30.01dB.The maximum-minimum filtering is combined with OTSU algorithm to remove the influence of shadow on the pound sheet and enhance the text.(2)Based on the one-stage text detection model TextBoxes++,dilated convolution was introduced,and lightweight backbone network MobileNet-V3 was used to replace VGG-16.The detection accuracy,recall rate and reasoning speed of the model were significantly improved,with the detection accuracy reaching 96.8%.CRNN network with STN module is used as text recognizer to improve the recognition accuracy under perspective distortion condition,and the recognition rate is up to 94.81%.(3)Developed a set of intelligent information management system of freight pound bill,realized the image deblurring,shadow removal,text detection and recognition and structured processing of pound bill,and provided abnormal data alarm,pound bill data persistence,specific data query and other functions.