A Smart Alcoholmeter Sensor Based on Deep Learning Visual Perception

Area of Science:

• Engineering
• Computer Science
• Chemistry

Background:

• Process automation enhances productivity and quality in industrial settings, including liquor production.
• Small and medium-sized liquor producers face challenges in modernizing equipment due to high costs.
• Alcohol concentration sensors are crucial for distillery automation, fraction separation, and process supervision.

Purpose of the Study:

• To propose a novel, low-cost method for sensing alcohol concentration in liquor production.
• To leverage deep learning for interpreting visual data from traditional alcoholmeters.
• To enable cost-effective automation and equipment modernization for small and medium producers.

Main Methods:

• Developed a dataset acquisition apparatus to capture labeled images of alcoholmeter readings.
• Treated alcohol concentration reading as both a regression and classification problem.
• Utilized the Resnet18 architecture for deep learning model development and performance evaluation.

Main Results:

• Achieved satisfying performance metrics for both regression and classification models.
• Demonstrated the feasibility of using deep learning on visual alcoholmeter data.
• The proposed system showed high accuracy in determining alcohol concentration from images.

Conclusions:

• The developed low-cost system, deployable on Raspberry Pi with a camera, is feasible for new and existing distillery equipment.
• Its non-invasive nature allows for retrofitting existing systems, promoting wider adoption.
• The approach is adaptable for reading other analog instruments by retraining the deep learning model.