A quantitative study of aggregation behaviour and integrity of spray-dried microcapsules using three deep convolutional neural networks with transfer learning

Abstract Deep convolutional neural networks (DCNN) with transfer learning were employed to automatically identify, classify, and quantify two morphological characteristics (aggregated or non-aggregated, and intact or broken) of spray-dried microcapsules in scanning electron microscope (SEM) images. Three DCNN-based models with different network depths were compared in terms of classification accuracy, training and testing times, feature visualization, and strongest activations. The novelties of this study are 1) the introduction of DCNN to analyse SEM images of microcapsules and the classification accuracy of all models is above 91%, 2) the application of transfer learning not only reduces the dependence of DCNN on high-performance computers and large-scale datasets but also reduces training time, 3) feature visualization and strongest activations demonstrate the understanding of morphological characteristics of microcapsules from the perspective of DCNN, and 4) the proposed method can be run on a personal computer and is suitable for widespread use in routine laboratories.