Identification and characterisation of Type Ⅱ MnS via YOLO deep learning method

Abstract

In order to solve the problem that the current inclusion characterisation methods can only identify single-point inclusions, the identification and characterisation model of type II MnS inclusions was established, and related software was developed. Scanning electron microscopy was used for establish a database of type II MnS. You Only Look Once deep learning model was used to realise the recognition of type II MnS with the learning rate 0.01. Image post-processing technologies such as edge detection and grey value extraction were used to characterise the identified type II MnS, and the accuracy of the characterisation was confirmed by visualising the characterisation information. This method achieved up to 91.485% mAP0.5 in identifying Type II MnS inclusions, and the accuracy and recall rate achieved up to 85.924% and 83.333%, respectively. The characterisation of Type II MnS inclusions could be accurately completed by adjusting the grey threshold. The current method to detect type II MnS inclusions greatly saved time and reduced the recognition error.