Glass Object Detection Based on AI Training Platform and Visionmaster Software

As a typical inorganic non-metallic material, glass exhibits high light transmittance, minimal surface texture characteristics, and complex reflective and refractive behaviors. These inherent physical properties pose significant challenges for accurately detecting glass objects in smart grid inspection tasks. To improve detection accuracy and adaptability in realworld complex environments, this study adopts a platformlevel approach within the context of smart grid inspections. By utilizing an AI training platform in combination with the VisionMaster software, we design and implement a glass object detection pipeline. A mainstream open-source dataset is used to train and evaluate multiple detection models systematically. Through comparative analysis, the model with the best trade-off between accuracy and generalization is selected and successfully deployed to the VisionMaster 3D vision algorithm platform. The proposed method achieves highprecision recognition and stable detection of transparent glass objects. Experimental results demonstrate that the method maintains strong robustness across various scenarios, multiangle perspectives, and complex backgrounds.