Changes in the Performance for Predicting Inappropriate Thermal Images according to the Composition of Datasets

초록

In this study, a deep neural network was used to classify normal and blurred thermal images. We conducted experiments to determine the types of performance changes caused by learning through datasets with different structures. In a data environment that uses images extracted from video, two methods were proposed: classifying training datasets and test datasets from all datasets and grouping and sorting them for each facility. Various deep neural network models were used to compare performances according to differences in dataset composition. The most suitable model among the trained models was applied to the new fuse dataset, which contained real data. Through the experiment using this dataset, the predictable performance in actual use was analyzed.

키워드