Satellite Image Classification Using Convolutional Neural Network

Summary

Image classification refers to segregating pictures based on their visual characteristics. This is a crucial area of study in computer vision and is widely used in various applications such as facial recognition, medical imaging, and object identification. With satellite image classification, this chapter aims to intelligently categorize satellite images based on their characteristics, which can benefit fields like urban planning, agribusiness, and environmental control. Several layers are used to classify satellite images, which begin with preparing the raw images to eliminate noise and enhance features. The next stage involves feature extraction, which can be done using various methods like custom feature descriptors and Convolutional Neural Networks (CNNs). These extracted features are then used to develop a classifier ranging from a simple Support Vector Machine (SVM) to a complex neural network. Classification accuracy can be impacted by choice of features and classifiers. Deep learning techniques, particularly CNNs, have performed exceptionally in tasks such as satellite image classification in recent years. These techniques can automatically learn features from data and identify complex relationships between input and output. Satellite image categorization is an essential issue with numerous applications, and recent advancements in deep learning have resulted in high levels of accuracy. However, it remains a subject of ongoing discussion, with ongoing efforts to improve the effectiveness and efficiency of classification algorithms. This chapter discusses creating a simple satellite image classification model with TensorFlow and ImageDataGenerator. The chapter describes the necessary steps for preparing data for the ImageDataGenerator, loading data using the ‘flow_from_directory’ function, visualizing the classification model's function, and improving its accuracy. The objective of the chapter is to aid readers in comprehending the fundamental concepts involved in building a satellite image classification model using deep learning methods. Adhering to the outlined steps allows readers to develop precise and effective satellite image classification models suitable for different purposes. This work implements two machine learning models, namely MobileNetV3 and EfficientNetB0. The accuracies achieved from both models are 98.4% and 99.65%, respectively.