CN-121982476-A - Illegal operation vehicle rapid identification system and method based on denoising convolutional neural network

CN121982476ACN 121982476 ACN121982476 ACN 121982476ACN-121982476-A

Abstract

The invention discloses a rapid identification system and method for illegal operation vehicles based on a denoising convolutional neural network, wherein the rapid identification system comprises a multi-module AI-ISP denoising convolutional neural network module, an LSTM and GRU circulating neural network model, an instant internal and external network penetration mapping PHTunnel module, a handheld terminal, a second-generation identification card identification module and an RFID module, wherein the multi-module AI-ISP denoising convolutional neural network module is used for processing and denoising vehicle related information acquired at the front end in real time, the LSTM and GRU circulating neural network module is used for analyzing vehicle space-time track data processed by the multi-module AI-ISP denoising convolutional neural network module so as to identify a high-frequency abnormal behavior mode of the illegal operation vehicles, the instant internal and external network penetration mapping PHTunnel module is used for ensuring that data are transmitted between the internal and external networks in real time and safely, the handheld terminal is connected with the LSTM and GRU circulating neural network model, the handheld terminal is integrated with the second-generation identification card identification module and the RFID module, and the customization APP is deeply integrated with the handheld terminal to realize on-site data acquisition, encryption uploading and instant case processing. The system can distinguish normal operation vehicles from potential illegal operation vehicles, and greatly improves the identification accuracy and case processing efficiency.

Inventors

CHEN SHIBIN

Assignees

重庆腾旭测控技术有限公司

Dates

Publication Date: 20260505
Application Date: 20260121

Claims (9)

1. An illegal operation vehicle rapid identification system based on a denoising convolutional neural network, which is characterized by comprising: The multi-module AI-ISP denoising convolutional neural network module is used for processing and denoising the vehicle related information acquired by the front end in real time; The LSTM and GRU circulating neural network model is used for analyzing the space-time track data of the vehicle processed by the multi-module AI-ISP denoising convolutional neural network module so as to identify a high-frequency abnormal behavior mode of the illegally operated vehicle; the real-time internal and external network penetration mapping PHTunnel component ensures that data is transmitted between the internal and external networks in real time and safely, and connects the LSTM and GRU circulating neural network model with the handheld terminal; the handheld terminal is integrated with a second-generation identity card identification module and an RFID module and is used for acquiring key evidence information on site; and customizing the APP, and deeply integrating with the handheld terminal to realize on-site data acquisition, encryption uploading and instant case processing.
2. The system of claim 1, wherein the AI-ISP module integrates DnCNN algorithms, dnCNN is a deep learning framework that employs a supervised learning strategy using convolutional neural network structures with residual learning for removing gaussian noise in vehicle images.
3. The system according to claim 1, wherein the DnCNN algorithm specifically comprises the steps of: Step 1, firstly, preparing a vehicle image data set, wherein the vehicle images comprise corresponding cleaning images and noise-added versions, step 2, network training, namely, training in a supervised learning mode, by DnCNN, sending each noise-added image and the corresponding cleaning image into a network together, wherein the network tries to learn a process of recovering the cleaning image from the noise-added image, namely, learning inverse mapping of noise, in the training process, the network adjusts weights of the predicted denoising image and the real cleaning image by taking a mean square error MSE as a loss function to improve the denoising effect, and step 3, namely, residual learning and network transmission, namely DnCNN, adopt residual learning, the network outputs an estimated value of noise instead of directly outputting the cleaning image, and then subtracts the estimated value from the noise-added image to obtain a denoised image, and finally, adopts a global residual block to subtract the input image and the denoising estimated value to obtain a final denoising result. And 4, after the network training is finished, applying DnCNN to the noisy image to denoise, inputting a noisy image into the network, and outputting the denoised image by the network.
4. The system of claim 1, wherein the platform system uses LSTM and GRU models in combination with holiday, weather and time information for deep vehicle behavior analysis.
5. The system of claim 1 wherein the PHTunnel component utilizes dynamic public network IP and port assignment techniques to quickly establish internal and external network connections while securing data transfer via SSL/TLS protocols.
6. The system of claim 1, wherein the handheld terminal is configured with a high-precision location service module to obtain law enforcement personnel location in real-time to optimize on-site evidence collection.
7. The system of claim 1, wherein the APP integrates natural language processing NLP and computer vision CV technology, automatically collates site evidence, forms an electronic evidence chain, and simplifies law enforcement procedures.
8. A system as claimed in claim 7, wherein the APP has electronic signature and blockchain certification functions, guaranteeing non-tamper-evident properties of legal documents and penalty vouchers.
9. A method of identifying an illegitimate operation vehicle using the system of any of claims 1 to 8, comprising the steps of: The method comprises the steps of denoising vehicle related information acquired by a front end in real time by using an AI-ISP denoising convolutional neural network; Analyzing the processed vehicle space-time track data through LSTM and GRU cyclic neural network models, and identifying an abnormal behavior mode; The handheld terminal and the custom APP acquire evidence information on site and encrypt and upload the evidence information to the platform system; the system generates an electronic evidence chain and legal documents according to the integrated vehicle image, the space-time track data and the evidence information; the handheld terminal prints the case processing receipt immediately.

Description

Illegal operation vehicle rapid identification system and method based on denoising convolutional neural network Technical Field The invention relates to the technical field of traffic comprehensive administration law enforcement, in particular to a system for rapidly checking and processing black vehicles, illegal operation, overrun vehicles, dangerous chemical transport vehicles and the like. Background At present, in traffic law enforcement, the problems of difficult verification, difficult case processing and low manual processing efficiency exist in the verification processing of vehicles such as black vehicles, illegal operations and the like. The traditional mode often needs to consume a great deal of time and manpower, one case can take two hours to finish processing, and the accuracy of the evidence chain is still to be improved. For example, it is difficult to accurately and quickly judge a suspicious vehicle that frequently occurs, and there is a hysteresis in information acquisition and processing. The present invention aims to solve these problems and improve law enforcement efficiency. Disclosure of Invention The present invention is directed to solving the above problems of the prior art. A system and a method for quickly identifying illegal operation vehicles based on a denoising convolutional neural network are provided. The technical scheme of the invention is as follows: an illegal operating vehicle rapid identification system based on a denoising convolutional neural network, comprising: The multi-module AI-ISP denoising convolutional neural network module is used for processing and denoising the vehicle related information acquired by the front end in real time; The LSTM and GRU circulating neural network model is used for analyzing the space-time track data of the vehicle processed by the multi-module AI-ISP denoising convolutional neural network module so as to identify a high-frequency abnormal behavior mode of the illegally operated vehicle; the real-time internal and external network penetration mapping PHTunnel component ensures that data is transmitted between the internal and external networks in real time and safely, and connects the LSTM and GRU circulating neural network model with the handheld terminal; the handheld terminal is integrated with a second-generation identity card identification module and an RFID module and is used for acquiring key evidence information on site; and customizing the APP, and deeply integrating with the handheld terminal to realize on-site data acquisition, encryption uploading and instant case processing. Further, the AI-ISP module integrates DnCNN algorithm, dnCNN is a deep learning framework that employs a supervised learning strategy using convolutional neural network structures with residual learning for removing Gaussian noise from vehicle images. Further, the DnCNN algorithm specifically includes the following steps: Step 1, firstly, preparing a vehicle image data set, wherein the vehicle images comprise corresponding cleaning images and noise-added versions, step 2, network training, namely, training in a supervised learning mode, by DnCNN, sending each noise-added image and the corresponding cleaning image into a network together, wherein the network tries to learn a process of recovering the cleaning image from the noise-added image, namely, learning inverse mapping of noise, in the training process, the network adjusts weights of the predicted denoising image and the real cleaning image by taking a mean square error MSE as a loss function to improve the denoising effect, and step 3, namely, residual learning and network transmission, namely DnCNN, adopt residual learning, the network outputs an estimated value of noise instead of directly outputting the cleaning image, and then subtracts the estimated value from the noise-added image to obtain a denoised image, and finally, adopts a global residual block to subtract the input image and the denoising estimated value to obtain a final denoising result. And 4, after the network training is finished, applying DnCNN to the noisy image to perform denoising, inputting a noisy image into the network, and outputting the denoised image by the network. Furthermore, the platform system adopts LSTM and GRU models, and combines holiday, weather and time information to perform deep vehicle behavior analysis. Furthermore, the PHTunnel component uses dynamic public network IP and port allocation technology to quickly establish the connection between the internal and external networks, and simultaneously ensures the data transmission safety through SSL/TLS protocol. Furthermore, the handheld terminal is provided with a high-precision position service module, so that the position of law enforcement personnel is obtained in real time, and the on-site evidence collection is optimized. Furthermore, the APP integrates NLP and CV technologies, automatically sorts on-site evidence, forms an electronic evidence