self-prompting analogical reasoning for uav object detection

Abstract

Unmanned Aerial Vehicle Object Detection (UAVOD) presents unique challenges due to varying altitudes, dynamic backgrounds, and the small size of objects. Traditional de- tection methods often struggle with these challenges, as they typically rely on visual features only and fail to ex- tract the semantic relations between the objects. To address these limitations,we propose a novel approach named Self- Prompting Analogical Reasoning (SPAR). Ourmethod uti- lizes the vision-languagemodel (CLIP) to generate context- aware prompts based on image features, providing rich se- mantic information that guides analogical reasoning. SPAR includes two main modules: self-prompting and analogi- cal reasoning. Self-prompting module based on learnable description and CLIP-text encoder generates context-aware prompt by combining specific image feature; then an object- ness prompt scoremap is produced by computing the simi- laritybetweenpixel-level features andcontext-awareprompt. Withthis scoremap,multi-scale image features are enhanced and pixel-level features are chosen for graph construction. While for analogical reasoningmodule, graph nodes consist of category-level prompt nodes and pixel-level image feature nodes.Analogical inference is based on graph convolution. Under the guidance of category-level nodes, different-scale object features have been enhanced, which helps achieve more accuratedetectionof challengingobjects.Extensive ex- periments illustrate that SPARoutperforms traditionalmeth- ods,offeringamorerobustandaccuratesolutionforUAVOD.