FDAS alarm image text verification method based on deep learning

Abstract

The invention discloses an FDAS alarm picture text verification method based on deep learning. The method comprises the following steps: preprocessing collected FDAS alarm images containing characters to form an FDAS alarm image data set; performing feature labeling on characters in the FDAS alarm images in the FDAS alarm image data set, and dividing the characters into a test set and a training set; inputting the FDAS alarm images in the training set into an A-LSTM model to carry out context-based feature extraction; inputting the extracted features into a classifier for classification, and obtaining an accuracy rate according to a test set, thereby continuously adjusting and optimizing the A-LSTM model; and inputting the verified FDAS alarm image into the A-LSTM model, and comparing the identified alarm text with the target alarm information. The method has good universality and transportability, and the model can verify and confirm the alarm information only by reading the shot FDAS alarm image.

Description

technical field [0001] The invention relates to the verification field of avionics systems and display systems, in particular to a deep learning-based FDAS (Flight Deck Alert System, aircraft cockpit warning system) warning image text verification method. Background technique [0002] With the increasing scale of avionics display systems and the increasing number of connection information between systems, the verification task of each configuration item is becoming increasingly arduous. FDAS (Flight Deck Alert System, aircraft cockpit warning system) is a configuration item that is difficult to verify in the display system. It involves many associated configuration items and the internal logic is relatively complicated. Existing automated testing methods can automatically trigger alarms without manual parameter setting, but the inspection of verification results requires a lot of manual participation in the inspection results. This large amount of repetitive work has caused...

AI Technical Summary

Problems solved by technology

[0003] In view of the problem that the test results cannot be efficiently verified during the FDAS verification process in the current display system, the purpose of the invention is to provide a method for verifying the text of the FDAS warning image based on deep learning, and obtain enough information for the pictures on the FDAS warning page. The FDAS warning image data set is marked with image features, and the verification data set and test data set are obtained. After obtaining the primary features of the FDAS warning image through the self-encoder, the context information features of the FDAS warning image are obtained through the LSTM model, so as to be recognized by the classifier. After the letters in the FDAS image constitute the warning text, they are compared with the triggered warning text to complete the verification of the test results

Images