Image Recognition Method and Device for Trace Scenarios Based on LSTM Neural Network
Mengjia Liu, Zhenya Shao, Chong Ma · 2024
The technique and apparatus for image recognition for trace situations presented in this research are based on the LSTM neural network and fall within the category of artificial intelligence/big data (image recognition/intelligent work order). When the work order system handles website access fault work orders (unreachable, packet loss, large delay, etc.), in the network monitoring and operation maintenance of telecommunication operators, the screenshots of routing trace information provided by customers cannot be directly used, and manually writing trace information is time-consuming and laborious. Furthermore, the total recognition accuracy for IP is low when using the conventional computer vision method for OCR recognition. Thus, based on work order annotation data and adaptive feature extraction, this research suggests an OCR recognition technique and device for work order trace information. The recognition accuracy is increased by the addition of an adaptive classifier and an LSTM neural network training process. This satisfies the high accuracy requirements of trace image recognition in the operators' online production environment and makes it easier to analyze network topology and fault location in the future using the trace recognition results.