Research on temperature detection method of liquor distilling pot feeding operation based on a compressed algorithm

Area of Science:

• Computer Vision
• Machine Learning
• Chemical Engineering

Background:

• Existing methods for detecting liquor vapor during distillation feeding are often delayed and lack precise location identification.
• Current computational models can be resource-intensive, hindering real-time deployment and performance.

Purpose of the Study:

• To develop a lightweight and efficient model for accurate, real-time detection of temperature changes on fermented grain surfaces.
• To improve the precision and speed of liquor vapor detection during the distillation feeding process.

Main Methods:

• Established an infrared image dataset of fermented grain surfaces.
• Fused YOLOv5n with knowledge distillation and model pruning algorithms to create the YOLOv5ns-DP model.
• Evaluated the model's performance in terms of parameters, computational load (GLOPs), model size, and mean Average Precision (mAP).

Main Results:

• The YOLOv5ns-DP model demonstrated significant reductions: 28.6% in parameters, 16.5% in GLOPs, and 26.4% in model size.
• Achieved a 0.6 improvement in mAP, indicating enhanced detection accuracy.
• The model effectively predicts and accurately locates liquor vapor, improving detection precision and speed.

Conclusions:

• The proposed YOLOv5ns-DP model offers a substantial improvement over existing methods for real-time temperature monitoring in fermented grains.
• This lightweight approach enhances computational efficiency and deployment performance for distillation processes.
• Accurate and early detection of liquor vapor is crucial for optimizing the distillation feeding process.