This paper provides a comprehensive comparative analysis of the performance of modern deep learning architectures for the object detection task. The research focuses on two main families of models: transformer architectures, including DETR (DEtection TRansformer) and its advanced variants such as RT-DETR (Real-Time DETR), D-FINE (DETR with Fine-grained Distribution Refinement) and DEIM (DETR with Improved Matching), as well as popular single-stage detectors of the YOLO (You Only Look Once) family, in particular, on the YOLOv11 and YOLOv12 versions. The models are evaluated on a specialized set of image data, which contains various defects of solar panels and consists of five classes, which makes it possible to identify the strengths and weaknesses of each architecture in the context of specific application tasks.

Keywords: solar panels, neural networks, detection, transformers, single-stage detectors, pattern recognition