Scene Image Classification Based on Optimized Transfer Learning Using ResNet

Scene image classification is a process that involves automatically identifying the type of environment depicted in an image or video, such as a street, office, mountain, etc. This process is accomplished by analyzing various visual features of the image, including shapes, colors and textures. Machine learning (ML) algorithms are then utilized to classify the image into different categories. However, deep CNNs can be challenging due to the requirement of large-scale datasets, which may not always be available for small-scale datasets. To address this problem, Transfer Learning can be applied to fine-tune pre-trained models for scene image classification. Conventional methods that use handcrafted aspects have limited representation ability, whereas deep CNN model features contain additional semantic and structural details. Therefore, the proposed approach employs an integrated with five pre-trained convolutional neural networks (CNNs) and is fine-tuned by a ResNet50-based transfer learning model that utilizes interlayer discriminating features for scene image classification. The suggested model's effectiveness, accuracy and efficiency are evaluated on a small-scale scene dataset. Comparing with the various current approaches, our findings perform with high accuracy. Therefore, the proposed approach can be considered as an effective and efficient model for scene image classification, especially for small-scale datasets. The carried-out experimentations with complex context scene datasets, including the UC Merced Land Dataset (UCM), the UIUC sports event dataset, and the MIT scene dataset. The experimental results show that the suggested approach is superior to current research on scene image classification approaches.