The automation of the disassembly process for end-of-life products in the electrical industry is essential for establishing a sustainable circular economy. A central challenge in this context is the initial identification of installed components, as their wide variety imposes significant demands on conventional object recognition systems. Therefore, this paper investigates the effectiveness of a deep learning-based computer vision system for automating the identification and localization of electrical components in control cabinets. For this purpose, the models YOLOv8, YOLO11 and Co-DETR are implemented and compared. The evaluation is based on a specifically developed dataset containing images of typical electrical components of industrial control cabinets. The performance metrics achieved, including mean average precision values exceeding 95% and inference times of a few milliseconds, demonstrate the effectiveness of the selected object recognition models. Additionally, optical character recognition models are integrated to enable granular identification of components, achieving accuracies of over 97%. The proposed detection system provides new insights into the application of deep learning models for automated disassembly processes and marks another step towards a sustainable circular economy in the electrical industry.