Use of deep learning-based object detection technique in quality control of silicon particle detectors

We present the application of deep learning-based computer vision techniques for quality control (QC) of silicon-based particle detectors, focusing on wire bonds and sensor surfaces, characterizing the features into classes such as 'no wires,' 'glue,' 'broken wires,' and more. Manual QC of printed circuit boards, electronics components, and silicon sensor surfaces are generally costly, labor-intensive, and error-prone. The use of You Only See Once (YOLOV5) object detection for QC purposes is not only an effective tool for the current Compact Muon Solenoid Endcap Calorimeter Upgrade project at the Advanced Particle Detector Laboratory, but also for the future applications to other detector construction projects, promising increased efficiency and precision.