Deep Convolutional Neural Networks For Classification of Satellite Images

Abstract

Deep learning algorithms that can learn from image, video, audio, and text data are becoming more successful as hardware power increases. Given the effectiveness and benefits of deep learning in many domains with more data, architecture should see similar implications. This study examined textures using particular rather than overall images. The deep convolutional neural network model classified 4500 satellite photos of clouds, deserts, greenery, and water. The constructed model classified previously unused test data (675 images) with 0.97 accuracies for cloud images, 0.98 for desert images, 0.96 for green areas, and 0.98 for water bodies. Although cloud and desert photos and green and water body images are comparable, this textural success shows that it can detect, analyze, and classify architectural elements. Deep convolutional neural networks can recognize, analyze, and classify architectural materials and elements, enabling shape recognition among many data to help architects collect helpful information. Thus, it will provide more extensive data than manual data analysis, enabling more accurate decisions. Understanding deep convolutional neural network data categorization characteristics explains architectural design differences and similarities. This condition reveals the hidden relationship in designs, allowing architects to create unique designs..