SARRSD: large-scale microwave radar remote sensing mapping-type and recognition-type object detection dataset

Object detection is a crucial task in the field of computer vision. Due to advancements in deep learning technologies over the past decade, significant progress has been made in object detection. However, these developments have lagged in their application to microwave radar remote sensing imagery. This gap arises not only from the unique characteristics and differences between microwave radar imaging and optical imaging but also because of the substantial disparities in shape and texture between targets in remote sensing images and ground objects in general scenarios. Additionally, there is a scarcity of high-quality annotated datasets for remote sensing scenarios, and microwave radar remote sensing target datasets remain limited in scale. To advance research on mapping-type and recognition-type target detection in microwave radar remote sensing, we introduce the Large-Scale Microwave Radar Remote Sensing Mapping-Type and Recognition-Type Object Detection Dataset (SARRSD). This dataset comprises 1,876 microwave radar remote sensing images, each with a resolution of approximately 1024×1024 pixels. It includes large-scale mapping-type targets and small-scale recognition-type targets, featuring extensive variations in target scales, diverse orientation distributions, and rich morphological characteristics. The dataset, annotated by aerial image interpretation experts, covers 18 common target categories (4 mapping-type and 14 recognition-type), containing a total of 10,923 instances. Each instance is annotated with rotated bounding boxes. To establish a benchmark for microwave radar remote sensing mapping and recognition datasets, we evaluated state-of-the-art object detection algorithms on SARRSD. Experimental results demonstrate that SARRSD effectively reflects real-world microwave radar visual application scenarios and presents significant challenges for object detection methods.