Object Detection Improvement Through ResNet-U-Net Fusion and GIS Integration for Satellite Imagery

Convolution Neural Networks (CNNs) have shown great results in analyzing satellite images with high accuracy and performance. However, it struggles while detecting smaller objects, performing under obstruction and when handling complex backgrounds. To address these issues, we propose a deep learning based fine-tuned U-Net model that uses ResNet50 as its backbone for feature extraction which means automatically learning high-level and low-level patterns. The U-Net helps image segmentation which means classifying each pixel in an image to accurately detect different objects or regions within the image. The Proposed ResNet50 with U-Net collectively improves performance in challenging conditions where traditional CNNs may fail. The designed model typically improves the detection accuracy and reduces the false alarm and supports large-scale processing of high-resolution satellite images with the help of Geographical Information System (GIS) and data sets from EUROSAT.