Applications of deep learning techniques in nanotechnology are particularly important since they allow for the optimal design of multi-characteristic nanomaterials. To differentiate between films composed of PS latex particles and MWCNTs in terms of their light-transmission characteristics connected DNN classifiers was fully trained. Intensity of transmitted light, particle size of PS latexes, percentage of MWCNT nanofillers by mass, and annealing temperature all played roles in classifying the spectroscopic data. Experiments were run to find the optimum values for the DNN classifier’s hyperparameters using a custom Bayesian optimizer. Proposed DNN classifiers may be evaluated using a variety of different measures, such as accuracy, the confusion matrix, cross-entropy loss, F1-score, recall, precision, and area under the curve calculated from receiver operating properties curvatures. When using sigmoid functions in the hidden layer units and setting the layer sizes to 30 and 20, respectively, the best results are shown on both the training and testing data sets for accuracy, precision, recall, and F1-score. Experimental data is few, but computational studies show that DNNs may be used to categorize the optical transparency of film samples.

PDF (Download Full text)

Download Citation (XML)