Accessing topological feature of polycrystalline microstructure using object detection technique

Faces-classes of grains, often referred to as topological features, largely dictate the evolution of polycrystalline microstructures during grain growth. Realising these topological features is generally an arduous task, often demanding sophisticated techniques. In the present work, a distinct machine-learning algorithm is extended for the first time to comprehend the topological distribution of the grains constituting a polycrystalline continuum. This regression-based object-detection approach, besides significantly reducing human-efforts and ensuring computational efficiency, predicts the face-class of the grains by introducing appropriate 'bounding boxes'. After sufficient training and validation, over 500 epochs, the current model exhibits a remarkable overlap with the ground truth that encompasses manually realised topological features of the polycrystalline microstructures. Accuracy of this treatment is further substantiated by relevant statistical studies including precision-recall analysis. The model is exposed to unknown test dataset and its performance is assessed by comparing its predictions with the labelled microstructures. Reflecting the statistical accuracy, a strong agreement between the algorithm-predictions and the ground truth is noticeable in these comparative studies involving polycrystalline systems with varying number of grains.