Image-based automated defect detection for optical masks of sun sensors

The sun sensor is critical to aerospace attitude control systems, with its optical mask as a core element. Micron-scale mask defects disrupt imaging, reduce accuracy, or even disable the sensor. Traditional manual inspection, inefficient, inaccurate, and inconsistent, fails aerospace’s high-precision need. Thus, this paper proposes a novel automated sun sensor mask defect recognition method based on an improved MobileNetV3. The original MobileNetV3 is systematically optimized via activation function replacement, bottleneck layer reconfiguration, network depth expansion, and training strategy improvement. The experimental results demonstrate that the optimized model achieves significant performance enhancement on a self-built dataset, with an accuracy of 99.54%, precision of 99.47%, recall of 99.23%, and F1 score of 99.01%, representing an improvement over the baseline model and maintaining the lightweight characteristic of the model. A Gradio-based automated optical mask detection application supports batch input, real-time defect classification, and outputs abnormal information, offering a reliable solution for intelligent non-destructive sun sensor mask inspection.