<p>The escalating demand for electricity is prompting substantial advancements in the power grid. Insulators, utilized as protective devices for transmission lines in elevated outdoor power systems, are extensively employed. The identification of defects in insulators under adverse conditions, including rain, snow, fog, sunlight, and the presence of rapidly moving drones during extensive photography, remains a significant challenge. To resolve this issue and enhance the precision of defect identification, this paper introduces a new system based on AI to detect insulator defects. The proposed system, called InsuDetNet that is based on deep learning. The proposed system passes through two stages: Image Preprocessing (IP) and (ii) Insulator Defect Detection (ID²). In the IP, the used images are preprocessed to be more suitable to feed to the trained YOLOv12 model, that used to detect and classify defects into three categories: insulator, pollution-flashover, and broken. Experimental results demonstrate the superiority of InsuDetNet over existing methods, achieving 97.44% F-measure, 95.0% Recall, 97.0% Precision, and 91.3% mAP@0.5, thus fulfilling the stringent accuracy and real-time requirements for outdoor insulator defect detection. This study highlights the novelty of integrating robust preprocessing with an advanced detection model for high-precision insulator monitoring.</p>