Classification of Environmental Images Using VGGNet

Abstract

Advances in digital image processing technology have encouraged the use of deep learning methods to perform automatic and accurate image classification. This study aims to build an environmental image classification model using a VGG16 architecture based on transfer learning and fine-tuning using the Intel Image Classification dataset, which consists of six categories, namely buildings, forest, glacier, mountain, sea, and street. The methods used include downloading the dataset via KaggleHub, pre-processing images by resizing, normalizing, and augmenting data, building a VGG16 model with a modified top layer, and training the model using the Adam optimizer and categorical cross- entropy loss function. Model performance was evaluated using a confusion matrix and classification report. The results of the study show that the model was able to achieve an overall accuracy of 90.16% on the test data, with the best performance obtained in the forest class, while relatively lower performance occurred in the glacier and mountain classes due to the similarity of visual characteristics between classes. Visualization of the prediction results shows that the model has fairly good generalization capabilities for images that were not used during training. These findings confirm that the application of transfer learning combined with fine-tuning on the VGG16 architecture is effective for environmental image classification and is suitable as a baseline in the development of image-based monitoring systems.