Prediction of Penetration Based on Feature Fusion of IR and CCD Images During Pulsed GTAW Process

Aiming at improving the recognition accuracy and robustness of the penetration state recognition model, a Coordination-Corrected-Mask-LCC-based (CCML-based) model with the input of original infrared thermal (IR) and visual (CCD) image was established. Since the arc flicker and irrelevant background thermal radiation during GTAW process are not propitious for predicting, while the industrial personal computer running model and the actual working scenario requires the model has light calculation burden and high operation efficiency, the adaptive target selection, synchronous feature extraction and feature fusion were designed in model. Adaptive target selection filtered interference information with the accurate frame selection of main targets in the input image. Synchronous feature extraction and feature fusion effectively and efficiently extracted the features of IR and CCD image. The recognition accuracy of CCML-based prediction model reached more than 99%, while the recognition time for each IR&CCD-image data pair was less than 212 ms.