Optimization of Sample Plate Detection for Robotic Arm in Mass Spectrometry Experiment Instrument Based on Improved YOLOv5s

To address the challenges of improving recognition accuracy and speed while reducing system costs in the recognition technology for robotic arms grasping samples to be tested in mass spectrometer experimental instruments, this paper proposes an automatic recognition and detection method based on improved YOLOv5s. Firstly, a Coordinate Attention (CA) mechanism is introduced to enhance feature learning and extraction from sample plate images, while reducing the impact of the background of the sample plate surface on detection results, especially when it is similar to the sample plate. Secondly, a path that directly connects shallow feature maps to deep feature maps is added to the original backbone network structure to enhance the model’s capability of detecting small target objects. Finally, the original C3 module is replaced with the DSConv module to reduce model complexity and maintain its accuracy. Experimental results show that in the custom dataset, compared to YOLOv5s, the improved model exhibits slightly increased parameter count and computational load but achieves significantly higher detection accuracy. In the COCO public dataset, the model outperforms other models with a high mAP of 0.723, while maintaining a compact model size of 20.2 MB and a real-time processing speed of 35.81 FPS.