Anomaly Recognition Of Electronic Dossiers Based On Deep Learning

With the reform of the case registration system and the acceleration of the judicial reform process,the electronic dossiers,as an important deployment in the intelligent law enforcement project,has been popularized throughout the country.Electronic dossiers refer to the real-time conversion of the collected and generated paper materials into electronic data in the process of case handling.In the process of converting paper dossiers into electronic dossiers,there will be anomalous electronic dossiers that need to be identified,including checking whether there is a department seal at the place where the seal should be signed and whether there is a person’s fingerprint at the place where the fingerprint should be signed.In addition,there is a lack of systematic identification technology in the face of anomalous electronic dossiers.Therefore,it is very necessary to study the technology of timely and efficient identification of anomalous electronic dossiers.This paper focuses on the anomalous identification technology of electronic dossiers,and designs three modules on how to distinguish anomalous electronic dossiers,including seal and fingerprint detection module,text detection module and text recognition module.The details are as follows:The seal and fingerprint detection module is used to detect whether there is a seal or fingerprint in the electronic dossiers.Based on the deep learning technology,this paper proposes to use the object detection algorithm for detection,analyzes the YOLOv5 s algorithm,SSD algorithm and Faster R-CNN algorithm which have good detection effect on the object,and carries out the simulation experiment on the seal and fingerprint detection data set.The experimental results show that the detection accuracy is as high as0.924 and the detection speed is only 23 milliseconds.The text detection module realizes the detection of the text area in the electronic dossiers.The color channel separation method is used to pre-treated the electronic dossiers,and the text is separated from the red seal and fingerprint to avoid the color interference with the subsequent text detection and recognition.Based on the DBNet detection algorithm and optimizing the loss function in the network,an improved DBNet-Tve electronic dossiers text detection technology is proposed.The algorithm is tested on the text detection data set,and the results show that its accuracy reaches 0.867,which is improved by 3 percentage points compared with DBNet algorithm.The text recognition module carries out text recognition for the detected text area of the electronic dossiers.This paper presents an improved CRNN-FPN electronic dossiers text recognition algorithm for recognition.Based on CRNN text recognition algorithm,FPN feature pyramid Network is added to strengthen the ability of network to extract image features.The CRNN-FPN electronic dossiers text recognition technology is simulated on the text recognition data set.The experimental results show that the text recognition accuracy is 0.901,which is nearly 3 percentage points higher than the CRNN algorithm.Electronic dossiers anomaly identification technology in order to identify anomalous electronic dossiers,according to the seals and fingerprints detected in the seal and fingerprint detection module,the seal anomaly keyword and fingerprint anomaly keyword identified in the text detection and identification module,identify whether the number of seal and seal anomaly keyword,fingerprint and fingerprint anomaly keyword is consistent.If not,it is determined as anomalous electronic dossiers,If it is consistent,the position information will be identified,and the shortest Euclidean Distance between seal and seal anomaly keyword,fingerprint and fingerprint anomaly keyword will be calculated.If the distance is more than the set threshold,it is determined as anomalous electronic dossiers,otherwise it is the normal electronic dossiers.