Transfer learning for hostel image classification

Purpose Because of the fast-growing digital image collections on online platforms and transfer learning ability deep technology, classification could be improved implemented for hostel domain, which has complex clusters contents. This paper aims to test potential 11 pretrained convolutional neural network (CNN) with first database advance knowledge fill gap academically, as well suggest an alternative solution in optimal less labour cost human errors those who manage collections. Design/methodology/approach The is created data pre-processing steps, selection augmentation. Then, systematic comprehensive investigation divided into seven experiments CNNs was applied parameters were fine-tuned match this newly dataset. All conducted Google Colaboratory environment using PyTorch. Findings 7,350 labelled classes. Furthermore, its experiment results highlight that DenseNet 121 201 have greatest they outperform other terms accuracy training time. Originality/value fact there no existing academic work dedicating image-only made a novel contribution.